asterisk/main/hashtab.c

905 lines
21 KiB
C
Raw Normal View History

This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
/*
* Asterisk -- An open source telephony toolkit.
*
* Copyright (C) 2007, Digium, Inc.
*
* Steve Murphy <murf@digium.com>
*
* See http://www.asterisk.org for more information about
* the Asterisk project. Please do not directly contact
* any of the maintainers of this project for assistance;
* the project provides a web site, mailing lists and IRC
* channels for your use.
*
* This program is free software, distributed under the terms of
* the GNU General Public License Version 2. See the LICENSE file
* at the top of the source tree.
*/
/*! \file
*
* \brief code to implement generic hash tables
*
* \author Steve Murphy <murf@digium.com>
*/
/*** MODULEINFO
<support_level>core</support_level>
***/
#define ASTMM_LIBC ASTMM_REDIRECT
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
#include "asterisk.h"
#include <ctype.h>
#include "asterisk/lock.h"
#include "asterisk/frame.h"
#include "asterisk/channel.h"
#include "asterisk/cli.h"
#include "asterisk/term.h"
#include "asterisk/utils.h"
#include "asterisk/threadstorage.h"
#include "asterisk/linkedlists.h"
#include "asterisk/hashtab.h"
#ifdef __AST_DEBUG_MALLOC
static void _ast_hashtab_resize(struct ast_hashtab *tab, const char *file, int lineno, const char *func);
#define ast_hashtab_resize(a) _ast_hashtab_resize(a,__FILE__, __LINE__, __PRETTY_FUNCTION__)
#else
static void ast_hashtab_resize(struct ast_hashtab *tab);
#endif
static void *ast_hashtab_lookup_internal(struct ast_hashtab *tab, const void *obj, unsigned int h);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
/* some standard, default routines for general use */
int ast_hashtab_compare_strings(const void *a, const void *b)
{
return strcmp(a, b);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
int ast_hashtab_compare_strings_nocase(const void *a, const void *b)
{
return strcasecmp(a, b);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
int ast_hashtab_compare_ints(const void *a, const void *b)
{
int ai = *((int *) a);
int bi = *((int *) b);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (ai < bi)
return -1;
return !(ai == bi);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
int ast_hashtab_compare_shorts(const void *a, const void *b)
{
short as = *((short *) a);
short bs = *((short *) b);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (as < bs)
return -1;
return !(as == bs);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
int ast_hashtab_resize_java(struct ast_hashtab *tab)
{
double loadfactor = (double) tab->hash_tab_elements / (double) tab->hash_tab_size;
return (loadfactor > 0.75);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
int ast_hashtab_resize_tight(struct ast_hashtab *tab)
{
return (tab->hash_tab_elements > tab->hash_tab_size); /* this is quicker than division */
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
int ast_hashtab_resize_none(struct ast_hashtab *tab) /* always return 0 -- no resizing */
{
return 0;
}
int ast_is_prime(int num)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
int tnum, limit;
if (!(num & 0x1)) /* even number -- not prime */
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return 0;
/* Loop through ODD numbers starting with 3 */
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
tnum = 3;
limit = num;
while (tnum < limit) {
if (!(num % tnum))
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
/* really, we only need to check sqrt(num) numbers */
limit = num / tnum;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
/* we only check odd numbers */
tnum = tnum + 2;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
/* if we made it through the loop, the number is a prime */
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return 1;
}
int ast_hashtab_newsize_java(struct ast_hashtab *tab)
{
int i = (tab->hash_tab_size << 1); /* multiply by two */
while (!ast_is_prime(i))
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
i++;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return i;
}
int ast_hashtab_newsize_tight(struct ast_hashtab *tab)
{
int x = (tab->hash_tab_size << 1);
int i = (tab->hash_tab_size + x);
while (!ast_is_prime(i))
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
i++;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return i;
}
int ast_hashtab_newsize_none(struct ast_hashtab *tab) /* always return current size -- no resizing */
{
return tab->hash_tab_size;
}
unsigned int ast_hashtab_hash_string(const void *obj)
{
unsigned char *str = (unsigned char *) obj;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
unsigned int total;
for (total = 0; *str; str++) {
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
unsigned int tmp = total;
total <<= 1; /* multiply by 2 */
total += tmp; /* multiply by 3 */
total <<= 2; /* multiply by 12 */
total += tmp; /* multiply by 13 */
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
total += ((unsigned int)(*str));
}
return total;
}
unsigned int ast_hashtab_hash_string_sax(const void *obj) /* from Josh */
{
const unsigned char *str = obj;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
unsigned int total = 0, c = 0;
while ((c = *str++))
total ^= (total << 5) + (total >> 2) + (total << 10) + c;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return total;
}
unsigned int ast_hashtab_hash_string_nocase(const void *obj)
{
const unsigned char *str = obj;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
unsigned int total;
for (total = 0; *str; str++) {
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
unsigned int tmp = total;
unsigned int charval = toupper(*str);
/* hopefully, the following is faster than multiplication by 7 */
/* why do I go to this bother? A good compiler will do this
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
anyway, if I say total *= 13 */
/* BTW, tried *= 7, and it doesn't do as well in spreading things around! */
total <<= 1; /* multiply by 2 */
total += tmp; /* multiply by 3 */
total <<= 2; /* multiply by 12 */
total += tmp; /* multiply by 13 */
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
total += (charval);
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return total;
}
unsigned int ast_hashtab_hash_int(const int x)
{
return x;
}
unsigned int ast_hashtab_hash_short(const short x)
{
/* hmmmm.... modulus is best < 65535 !! */
return x;
}
struct ast_hashtab *
#ifdef __AST_DEBUG_MALLOC
_ast_hashtab_create
#else
ast_hashtab_create
#endif
(int initial_buckets,
int (*compare)(const void *a, const void *b),
int (*resize)(struct ast_hashtab *),
int (*newsize)(struct ast_hashtab *tab),
unsigned int (*hash)(const void *obj),
int do_locking
#ifdef __AST_DEBUG_MALLOC
, const char *file, int lineno, const char *function
#endif
)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
struct ast_hashtab *ht;
#ifdef __AST_DEBUG_MALLOC
if (!(ht = __ast_calloc(1, sizeof(*ht), file, lineno, function)))
#else
if (!(ht = ast_calloc(1, sizeof(*ht))))
#endif
return NULL;
while (!ast_is_prime(initial_buckets)) /* make sure this is prime */
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
initial_buckets++;
#ifdef __AST_DEBUG_MALLOC
if (!(ht->array = __ast_calloc(initial_buckets, sizeof(*(ht->array)), file, lineno, function))) {
#else
if (!(ht->array = ast_calloc(initial_buckets, sizeof(*(ht->array))))) {
#endif
ast_free(ht);
return NULL;
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
ht->hash_tab_size = initial_buckets;
ht->compare = compare;
ht->resize = resize;
ht->newsize = newsize;
ht->hash = hash;
ht->do_locking = do_locking;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (do_locking)
ast_rwlock_init(&ht->lock);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (!ht->resize)
ht->resize = ast_hashtab_resize_java;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (!ht->newsize)
ht->newsize = ast_hashtab_newsize_java;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return ht;
}
#ifdef __AST_DEBUG_MALLOC
struct ast_hashtab *_ast_hashtab_dup(struct ast_hashtab *tab, void *(*obj_dup_func)(const void *obj), const char *file, int lineno, const char *func)
#else
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
struct ast_hashtab *ast_hashtab_dup(struct ast_hashtab *tab, void *(*obj_dup_func)(const void *obj))
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
struct ast_hashtab *ht;
unsigned int i;
if (!(ht = ast_calloc(1, sizeof(*ht))))
return NULL;
if (!(ht->array =
#ifdef __AST_DEBUG_MALLOC
__ast_calloc(tab->hash_tab_size, sizeof(*(ht->array)), file, lineno, func)
#else
ast_calloc(tab->hash_tab_size, sizeof(*(ht->array)))
#endif
)) {
free(ht);
return NULL;
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
ht->hash_tab_size = tab->hash_tab_size;
ht->compare = tab->compare;
ht->resize = tab->resize;
ht->newsize = tab->newsize;
ht->hash = tab->hash;
ht->do_locking = tab->do_locking;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (ht->do_locking)
ast_rwlock_init(&ht->lock);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
/* now, dup the objects in the buckets and get them into the table */
/* the fast way is to use the existing array index, and not have to hash
the objects again */
for (i = 0; i < ht->hash_tab_size; i++) {
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
struct ast_hashtab_bucket *b = tab->array[i];
while (b) {
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
void *newobj = (*obj_dup_func)(b->object);
if (newobj)
#ifdef __AST_DEBUG_MALLOC
_ast_hashtab_insert_immediate_bucket(ht, newobj, i, file, lineno, func);
#else
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
ast_hashtab_insert_immediate_bucket(ht, newobj, i);
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
b = b->next;
}
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return ht;
}
static void tlist_del_item(struct ast_hashtab_bucket **head, struct ast_hashtab_bucket *item)
{
/* item had better be in the list! or suffer the weirdness that occurs, later! */
if (*head == item) { /* first item in the list */
*head = item->tnext;
if (item->tnext)
item->tnext->tprev = NULL;
} else {
/* short circuit stuff */
item->tprev->tnext = item->tnext;
if (item->tnext)
item->tnext->tprev = item->tprev;
}
}
static void tlist_add_head(struct ast_hashtab_bucket **head, struct ast_hashtab_bucket *item)
{
if (*head) {
item->tnext = *head;
item->tprev = NULL;
(*head)->tprev = item;
*head = item;
} else {
/* the list is empty */
*head = item;
item->tprev = NULL;
item->tnext = NULL;
}
}
/* user-controlled hashtab locking. Create a hashtab without locking, then call the
following locking routines yourself to lock the table between threads. */
void ast_hashtab_wrlock(struct ast_hashtab *tab)
{
ast_rwlock_wrlock(&tab->lock);
}
void ast_hashtab_rdlock(struct ast_hashtab *tab)
{
ast_rwlock_rdlock(&tab->lock);
}
void ast_hashtab_initlock(struct ast_hashtab *tab)
{
ast_rwlock_init(&tab->lock);
}
void ast_hashtab_destroylock(struct ast_hashtab *tab)
{
ast_rwlock_destroy(&tab->lock);
}
void ast_hashtab_unlock(struct ast_hashtab *tab)
{
ast_rwlock_unlock(&tab->lock);
}
void ast_hashtab_destroy(struct ast_hashtab *tab, void (*objdestroyfunc)(void *obj))
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
/* this func will free the hash table and all its memory. It
doesn't touch the objects stored in it */
if (tab) {
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_wrlock(&tab->lock);
if (tab->array) {
/* go thru and destroy the buckets */
struct ast_hashtab_bucket *t;
int i;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
while (tab->tlist) {
t = tab->tlist;
if (t->object && objdestroyfunc) {
/* I cast this because I'm not going to MOD it, I'm going to DESTROY
* it.
*/
(*objdestroyfunc)((void *) t->object);
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
tlist_del_item(&(tab->tlist), tab->tlist);
free(t);
}
for (i = 0; i < tab->hash_tab_size; i++) {
/* Not totally necessary, but best to destroy old pointers */
tab->array[i] = NULL;
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
free(tab->array);
}
if (tab->do_locking) {
ast_rwlock_unlock(&tab->lock);
ast_rwlock_destroy(&tab->lock);
}
free(tab);
}
}
#ifdef __AST_DEBUG_MALLOC
int _ast_hashtab_insert_immediate(struct ast_hashtab *tab, const void *obj, const char *file, int lineno, const char *func)
#else
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
int ast_hashtab_insert_immediate(struct ast_hashtab *tab, const void *obj)
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
unsigned int h;
int res=0;
if (!tab || !obj)
return res;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_wrlock(&tab->lock);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
h = (*tab->hash)(obj) % tab->hash_tab_size;
#ifdef __AST_DEBUG_MALLOC
res = _ast_hashtab_insert_immediate_bucket(tab, obj, h, file, lineno, func);
#else
res = ast_hashtab_insert_immediate_bucket(tab, obj, h);
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_unlock(&tab->lock);
return res;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
#ifdef __AST_DEBUG_MALLOC
int _ast_hashtab_insert_immediate_bucket(struct ast_hashtab *tab, const void *obj, unsigned int h, const char *file, int lineno, const char *func)
#else
int ast_hashtab_insert_immediate_bucket(struct ast_hashtab *tab, const void *obj, unsigned int h)
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
int c;
struct ast_hashtab_bucket *b;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (!tab || !obj)
return 0;
for (c = 0, b = tab->array[h]; b; b= b->next)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
c++;
if (c + 1 > tab->largest_bucket_size)
tab->largest_bucket_size = c + 1;
if (!(b =
#ifdef __AST_DEBUG_MALLOC
__ast_calloc(1, sizeof(*b), file, lineno, func)
#else
ast_calloc(1, sizeof(*b))
#endif
)) return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
b->object = obj;
b->next = tab->array[h];
tab->array[h] = b;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (b->next)
b->next->prev = b;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
tlist_add_head(&(tab->tlist), b);
tab->hash_tab_elements++;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if ((*tab->resize)(tab))
ast_hashtab_resize(tab);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return 1;
}
#ifdef __AST_DEBUG_MALLOC
int _ast_hashtab_insert_safe(struct ast_hashtab *tab, const void *obj, const char *file, int lineno, const char *func)
#else
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
int ast_hashtab_insert_safe(struct ast_hashtab *tab, const void *obj)
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
/* check to see if the element is already there; insert only if
it is not there. */
/* will force a resize if the resize func returns 1 */
/* returns 1 on success, 0 if there's a problem, or it's already there. */
unsigned int bucket = 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_wrlock(&tab->lock);
if (!ast_hashtab_lookup_bucket(tab, obj, &bucket)) {
#ifdef __AST_DEBUG_MALLOC
int ret2 = _ast_hashtab_insert_immediate_bucket(tab, obj, bucket, file, lineno, func);
#else
int ret2 = ast_hashtab_insert_immediate_bucket(tab, obj, bucket);
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_unlock(&tab->lock);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return ret2;
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_unlock(&tab->lock);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return 0;
}
void *ast_hashtab_lookup(struct ast_hashtab *tab, const void *obj)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
/* lookup this object in the hash table. return a ptr if found, or NULL if not */
unsigned int h;
void *ret;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (!tab || !obj)
return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_rdlock(&tab->lock);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
h = (*tab->hash)(obj) % tab->hash_tab_size;
ret = ast_hashtab_lookup_internal(tab,obj,h);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_unlock(&tab->lock);
return ret;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
void *ast_hashtab_lookup_with_hash(struct ast_hashtab *tab, const void *obj, unsigned int hashval)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
/* lookup this object in the hash table. return a ptr if found, or NULL if not */
unsigned int h;
void *ret;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (!tab || !obj)
return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_rdlock(&tab->lock);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
h = hashval % tab->hash_tab_size;
ret = ast_hashtab_lookup_internal(tab,obj,h);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_unlock(&tab->lock);
return ret;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
void *ast_hashtab_lookup_bucket(struct ast_hashtab *tab, const void *obj, unsigned int *bucket)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
/* lookup this object in the hash table. return a ptr if found, or NULL if not */
unsigned int h;
void *ret;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (!tab || !obj)
return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
h = (*tab->hash)(obj) % tab->hash_tab_size;
ret = ast_hashtab_lookup_internal(tab,obj,h);
*bucket = h;
return ret;
}
static void *ast_hashtab_lookup_internal(struct ast_hashtab *tab, const void *obj, unsigned int h)
{
struct ast_hashtab_bucket *b;
for (b = tab->array[h]; b; b = b->next) {
if (!(*tab->compare)(obj,b->object)) {
/* I can't touch obj in this func, but the outside world is welcome to */
return (void*) b->object;
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
return NULL;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
void ast_hashtab_get_stats(struct ast_hashtab *tab, int *biggest_bucket_size, int *resize_count, int *num_objects, int *num_buckets)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
/* returns key stats for the table */
if (tab->do_locking)
ast_rwlock_rdlock(&tab->lock);
*biggest_bucket_size = tab->largest_bucket_size;
*resize_count = tab->resize_count;
*num_objects = tab->hash_tab_elements;
*num_buckets = tab->hash_tab_size;
if (tab->do_locking)
ast_rwlock_unlock(&tab->lock);
}
/* this function returns the number of elements stored in the hashtab */
int ast_hashtab_size(struct ast_hashtab *tab)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
return tab->hash_tab_elements;
}
/* this function returns the size of the bucket array in the hashtab */
int ast_hashtab_capacity( struct ast_hashtab *tab)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
return tab->hash_tab_size;
}
/* the insert operation calls this, and is wrlock'd when it does. */
/* if you want to call it, you should set the wrlock yourself */
#ifdef __AST_DEBUG_MALLOC
static void _ast_hashtab_resize(struct ast_hashtab *tab, const char *file, int lineno, const char *func)
#else
static void ast_hashtab_resize(struct ast_hashtab *tab)
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
/* this function is called either internally, when the resize func returns 1, or
externally by the user to force a resize of the hash table */
int newsize = (*tab->newsize)(tab), i, c;
unsigned int h;
struct ast_hashtab_bucket *b,*bn;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
/* Since we keep a DLL of all the buckets in tlist,
all we have to do is free the array, malloc a new one,
and then go thru the tlist array and reassign them into
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
the bucket arrayj.
*/
for (i = 0; i < tab->hash_tab_size; i++) { /* don't absolutely have to do this, but
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
why leave ptrs laying around */
tab->array[i] = 0; /* erase old ptrs */
}
free(tab->array);
if (!(tab->array =
#ifdef __AST_DEBUG_MALLOC
__ast_calloc(newsize, sizeof(*(tab->array)), file, lineno, func)
#else
ast_calloc(newsize, sizeof(*(tab->array)))
#endif
))
return;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
/* now sort the buckets into their rightful new slots */
tab->resize_count++;
tab->hash_tab_size = newsize;
tab->largest_bucket_size = 0;
for (b = tab->tlist; b; b = bn) {
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
b->prev = 0;
bn = b->tnext;
h = (*tab->hash)(b->object) % tab->hash_tab_size;
b->next = tab->array[h];
if (b->next)
b->next->prev = b;
tab->array[h] = b;
}
/* recalc the largest bucket size */
for (i = 0; i < tab->hash_tab_size; i++) {
for (c = 0, b = tab->array[i]; b; b = b->next)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
c++;
if (c > tab->largest_bucket_size)
tab->largest_bucket_size = c;
}
}
#ifdef __AST_DEBUG_MALLOC
struct ast_hashtab_iter *_ast_hashtab_start_traversal(struct ast_hashtab *tab, const char *file, int lineno, const char *func)
#else
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
struct ast_hashtab_iter *ast_hashtab_start_traversal(struct ast_hashtab *tab)
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
/* returns an iterator */
struct ast_hashtab_iter *it;
if (!(it =
#ifdef __AST_DEBUG_MALLOC
__ast_calloc(1, sizeof(*it), file, lineno, func)
#else
ast_calloc(1, sizeof(*it))
#endif
))
return NULL;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
it->next = tab->tlist;
it->tab = tab;
if (tab->do_locking)
ast_rwlock_rdlock(&tab->lock);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return it;
}
/* use this function to get a write lock */
#ifdef __AST_DEBUG_MALLOC
struct ast_hashtab_iter *_ast_hashtab_start_write_traversal(struct ast_hashtab *tab, const char *file, int lineno, const char *func)
#else
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
struct ast_hashtab_iter *ast_hashtab_start_write_traversal(struct ast_hashtab *tab)
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
{
/* returns an iterator */
struct ast_hashtab_iter *it;
if (!(it =
#ifdef __AST_DEBUG_MALLOC
__ast_calloc(1, sizeof(*it), file, lineno, func)
#else
ast_calloc(1, sizeof(*it))
#endif
))
return NULL;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
it->next = tab->tlist;
it->tab = tab;
if (tab->do_locking)
ast_rwlock_wrlock(&tab->lock);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return it;
}
void ast_hashtab_end_traversal(struct ast_hashtab_iter *it)
{
if (!it)
return;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (it->tab->do_locking)
ast_rwlock_unlock(&it->tab->lock);
free(it);
}
void *ast_hashtab_next(struct ast_hashtab_iter *it)
{
/* returns the next object in the list, advances iter one step */
struct ast_hashtab_bucket *retval;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (it && it->next) { /* there's a next in the bucket list */
retval = it->next;
it->next = retval->tnext;
return (void *) retval->object;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return NULL;
}
static void *ast_hashtab_remove_object_internal(struct ast_hashtab *tab, struct ast_hashtab_bucket *b, int h)
{
const void *obj2;
if (b->prev)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
b->prev->next = b->next;
else
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
tab->array[h] = b->next;
if (b->next)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
b->next->prev = b->prev;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
tlist_del_item(&(tab->tlist), b);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
obj2 = b->object;
b->object = b->next = (void*)2;
free(b); /* free up the hashbucket */
tab->hash_tab_elements--;
#ifdef DEBUG
{
int c2;
struct ast_hashtab_bucket *b2;
/* do a little checking */
for (c2 = 0, b2 = tab->tlist; b2; b2 = b2->tnext) {
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
c2++;
}
if (c2 != tab->hash_tab_elements) {
printf("Hey! we didn't delete right! there are %d elements in the list, and we expected %d\n",
c2, tab->hash_tab_elements);
}
for (c2 = 0, b2 = tab->tlist; b2; b2 = b2->tnext) {
unsigned int obj3 = (unsigned long) obj2;
unsigned int b3 = (unsigned long) b;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (b2->object == obj2)
printf("Hey-- you've still got a bucket pointing at ht_element %x\n", obj3);
if (b2->next == b)
printf("Hey-- you've still got a bucket with next ptr pointing to deleted bucket %x\n", b3);
if (b2->prev == b)
printf("Hey-- you've still got a bucket with prev ptr pointing to deleted bucket %x\n", b3);
if (b2->tprev == b)
printf("Hey-- you've still got a bucket with tprev ptr pointing to deleted bucket %x\n", b3);
if (b2->tnext == b)
printf("Hey-- you've still got a bucket with tnext ptr pointing to deleted bucket %x\n", b3);
}
}
#endif
return (void *) obj2; /* inside this code, the obj's are untouchable, but outside, they aren't */
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
void *ast_hashtab_remove_object_via_lookup(struct ast_hashtab *tab, void *obj)
{
/* looks up the object; removes the corresponding bucket */
const void *obj2;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (!tab || !obj)
return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_wrlock(&tab->lock);
obj2 = ast_hashtab_remove_object_via_lookup_nolock(tab,obj);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_unlock(&tab->lock);
return (void *)obj2;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
void *ast_hashtab_remove_object_via_lookup_nolock(struct ast_hashtab *tab, void *obj)
{
/* looks up the object; removes the corresponding bucket */
unsigned int h;
struct ast_hashtab_bucket *b;
if (!tab || !obj)
return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
h = (*tab->hash)(obj) % tab->hash_tab_size;
for (b = tab->array[h]; b; b = b->next) {
if (!(*tab->compare)(obj, b->object)) {
const void *obj2;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
obj2 = ast_hashtab_remove_object_internal(tab, b, h);
return (void *) obj2; /* inside this code, the obj's are untouchable, but outside, they aren't */
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
return 0;
}
void *ast_hashtab_remove_this_object(struct ast_hashtab *tab, void *obj)
{
/* looks up the object by hash and then comparing pts in bucket list instead of
calling the compare routine; removes the bucket -- a slightly cheaper operation */
/* looks up the object; removes the corresponding bucket */
const void *obj2;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (!tab || !obj)
return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_wrlock(&tab->lock);
obj2 = ast_hashtab_remove_this_object_nolock(tab,obj);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (tab->do_locking)
ast_rwlock_unlock(&tab->lock);
return (void *)obj2;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
void *ast_hashtab_remove_this_object_nolock(struct ast_hashtab *tab, void *obj)
{
/* looks up the object by hash and then comparing pts in bucket list instead of
calling the compare routine; removes the bucket -- a slightly cheaper operation */
/* looks up the object; removes the corresponding bucket */
unsigned int h;
struct ast_hashtab_bucket *b;
if (!tab || !obj)
return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
h = (*tab->hash)(obj) % tab->hash_tab_size;
for (b = tab->array[h]; b; b = b->next) {
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
if (obj == b->object) {
const void *obj2;
obj2 = ast_hashtab_remove_object_internal(tab, b, h);
return (void *) obj2; /* inside this code, the obj's are untouchable, but outside, they aren't */
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@89129 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-11-09 16:00:22 +00:00
}
}
return 0;
}