The fetcher will try:
1) PREMIRROR
2) Upstream
3) MIRROR
If it fails to download from the Upstream, but succeeds from the MIRROR,
and ud.localpath != origud.localpath (for example, the git tarball),
then we will get the error (e.g.: xf86-video-omapfb):
ERROR: Function failed: Fetcher failure for URL: 'xxx'. Unable to fetch URL from any source.
ERROR: Logfile of failure stored in: /path/to/log.do_fetch.28024
It should not show the error and let the build go on since it succeeds.
(e.g.: xf86-video-omapfb)
[YOCTO #5686]
(Bitbake rev: 3bb3f1823bdd46ab34577d43f1e39046a32bca77)
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
A typo was meaning that the mirror creation method wasn't being called
when it should have been. Fix the type to fix mirror tarball creation.
[YOCTO #5284]
(Bitbake rev: 66cdc2e21660847c50317e8bfd28cf3595422e28)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
* currently decode_url regexp parses branch=@foo as username so it ends like this:
- ('git', '', 'foo', 'git.openembedded.org/bitbake;branch=', '', {})
+ ('git', 'git.openembedded.org', '/bitbake', '', '', {'branch': '@foo'})
* http://hg.python.org/cpython/file/2.7/Lib/urlparse.py also assumes
that there is at least one '/' as separator between netloc and path,
params, so it looks reasonable to prevent including '/' in username
(Bitbake rev: 3c694e20df3b1d442603300786580e4b2f4bf5f3)
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If an SCM mirror is in PREMIRRORS, the tarball is downloaded and then found
by the "upstream" check and handled correctly.
If an SCM mirror is in MIRRORS, the tarball is downloaded but not used
since there is no "upstream" run after MIRRORS completes. It therefore
sits there useless and unused. This code change forces the upstream to
run after a mirror tarball is found and fixes the usage of SCM mirrors
in MIRRORS.
(Bitbake master rev: a66ee0994645aa5658b2f5ea134ed17d89f8751a)
(Bitbake rev: 98d2cd8576a8d035e2b073cd54bb737a3c22bc4d)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If a file ends with .xz, it currently gets overwritten during unpack:
The decompress command for .xz files is:
'xz -dc %s > %s' % (file, efile)
and as efile == file, we end up overwriting file (the source).
Fix this by adding .xz to the list of suffixes that that need to
be removed from a file name for an extract command, leaving the
bare file name. Now, for a given file foo.xz,
file == foo.xz and efile == foo, similar to how .gz .bz2 and .Z
files are treated.
(Bitbake rev: 2cd2d0a48e12ab4358fb967eaf7a56c17993f48d)
Signed-off-by: André Draszik <andre.draszik@linaro.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
BB_NO_NETWORK can be set by bitbake internally by the use of
BB_FETCH_PREMIRRORONLY so update the error message to give users a
hint about this.
[YOCTO #3222]
(Bitbake rev: cac3060d0bf8c7deeacda18d06d92787911380d0)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
AUTOINC was meant to appear once at the start of the version string.
The list of names may not be sorted meaning it could get inserted in
the middle. This patch simplifies the code and ensures it appears at
the start.
Include cache version bump to ensure the cache picks up these changes.
(Bitbake rev: ad8bf10d873abb94d987860a3f6d06b134fb8a99)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The assignment to True was missing from the code, well spotted Saul!
(Bitbake rev: e493fe8cb4953935f01361ffc0240e5818ebb283)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The long srcrevs are mainly used or the workdir construction as well as
the package version. The long entries are hashes generated by the git fetcher
and other scms using a similar revision mechanism.
We need these to change when the package changes however collisions are
unlikely to happen within the domains we care about. The long revisions
have generated negative user feedback due to the use in path and file
names.
This patch therefore truncates the revisions to 10 characters maximum.
This should be safe in the contexts where these revisions are used as
the chances of spatially close collisions is very low (distant
collisions are not a major issue in the way we use these).
(Bitbake rev: 43a8319cda7fae37862dae323eeb24cb39ca21b7)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Now we no longer try and provide increasing values from the fetcher,
we can simplify the function structure for the sortable_revision
pieces and move the AUTOINC handling directly into the function
which needs it, simplifying the code.
(Bitbake rev: fb068bee47bb1a06f02447daf16c2b2a79c03288)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Move various random imports to the start of the modules as cleanup
and avoid an import issue with bb.process on python 2.6.
(Bitbake rev: aed4adfbe3a591ca4f8e41fb763c9f961bf2e6d5)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
* this can be useful when someone wan't to compare old file with
bad checksum and new one
(Bitbake rev: 33c6b93597dd43ab03ce7b62ba3eeb1893a68c38)
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This adds very basic git submodule support to the fetcher. It can be
used by replacing a git:// url prefix with a gitsm:// prefix, otherwise
behaviour is the same as the git fetcher. Whilst this code should be
functional, its not as efficient as the usual git fetcher due to the
need to checkout the tree to fetch/update the submodule information. git
doesn't support submodule operations on the bare clones the standard git
fetcher uses which is also problematic.
This code does however give a starting point to people wanting to use
submodules.
(Bitbake rev: 25e0b0bc50114f1fbf955de23cc0c96f5f7a41e3)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This reverts commit 21fe2683aefde10e847e66c11c26d4f4c1e07cfd
since bitbake-selftest doesn't pass when this is applied and
we're seeing multiple build failures from this change.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This fetcher differs from the SSH fetcher in that it adheres more
strictly to the SECSH URI internet draft --- it uses the sftp://
instead of the ssh:// scheme, and it uses sftp instead of scp.
(Bitbake rev: d240baeb7a4107d2eba3f08c411c0f086674d8e2)
Signed-off-by: Olof Johansson <olof.johansson@axis.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
A class representing a generic URI, with methods for accessing the URI
components, and stringifies to the URI. This class should be a bit more
flexible than the existing {encode,decode}_url functions in that it
supports more components (e.g. port) and that it does not rely on a
specific order on the return values. This makes it easy to add new
properties without affecting the API.
(Bitbake rev: bd824da8a7eafe27310e410807319628378caeca)
Signed-off-by: Olof Johansson <olof.johansson@axis.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Currently stamps end up at the top level of DL_DIR even if the files
themselves are in a directory structure. This patch preserves path
components allowing the top level directory to be less populated
which is an advantage for large sstate caches.
(Bitbake rev: 59921ce3ed7a4c0b7f8ef1a101ad9127469bf1fd)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
For example, this allows us to specify downloadfilename=PATH in sstate
mirror urls to improve directory structures in DL_DIR when using sstate.
(Bitbake rev: 7850a1364b6b37c58664d84f9c14806b4479b45c)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The following variables perform no function outside of bitbake:
GIT_CONFIG
GIT_PROXY_HOST
GIT_PROXY_PORT
GIT_PROXY_IGNORE
GIT_CONFIG only affects the git-config command which is not relevant to
the fetcher. This was previously used with the OE GIT_CORE_CONFIG
variable which would provide a basic git config to use instead of the
user's config. This usage was deprecated by git for over a year now:
http://git.661346.n2.nabble.com/Overriding-gitconfig-using-GIT-CONFIG-td6680977.html
GIT_PROXY_HOST and GIT_PROXY_PORT are not used by git.
GIT_PROXY_IGNORE was an OE construct used to create the custom git
config and had no meaning outside of the OE environment. It is not used
by git.
Remove these variables from the fetcher environment.
Users wishing to configure git to work with a proxy should define the
GIT_PROXY_COMMAND environment variable to use an external script.
NO_PROXY can be used within this script to skip the proxy for certain
hosts.
(Bitbake rev: e60270bdce6b8c2f8da1a4838aa374da9db3c86a)
Signed-off-by: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Applications are inconsistent in their use of upper and lower case proxy
variables. Curl, for example, specifies NO_PROXY (not no_proxy) in the
man page (changed in 2009 [1]). Avoid proxy issues by ensuring both the
upper and lower case versions of each proxy variable is available in the
environment for the fetcher commands.
Add FTPS_PROXY and ftps_proxy to the list as well.
1. http://curl.haxx.se/mail/tracker-2009-04/0012.html
(Bitbake rev: c3e6b2c5ec81d5ad7dcf606fff16fd5552bd267c)
Signed-off-by: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If a particular SRCREV (say for a particular branch) is missing, the
fetcher will currently just report an obtuse error about the "SRCREV"
being invalid. If there is more information is to be had (say from name,
i.e. branch, and pn) then display that as well.
The new error looks something like this:
ERROR: ExpansionError during parsing /home/dvhart/source/poky/meta/recipes-kernel/linux/linux-yocto_3.4.bb: Failure expanding variable do_patch: ExpansionError: Failure expanding variable SRCPV, expression was ${@bb.fetch2.get_srcrev(d)} which triggered exception FetchError: Fetcher failure for URL: 'git://otcgit.jf.intel.com/dvhart/linux-yocto-minnow-3.4.git;protocol=git;nocheckout=1;branch=standard/minnow,meta,emgd-1.14;name=machine,meta,emgd'. Please set SRCREV_emgd_pn-linux-yocto to a valid value
Note the variable listed as invalid is
"SRCREV_emgd_pn-linux-yocto", making it explicit what is wrong.
(Bitbake rev: 63774f5b4edb999300bddd891233f6050f4af877)
Signed-off-by: Darren Hart <dvhart@linux.intel.com>
Cc: bitbake-devel@lists.openembedded.org
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Changing the path to a file could change the task hash even if the file
still has the same checksum. This occurs when the task depends on
multiple files and the sort order of their paths changes. Usually the
sorting is consistent because layers tend to have the same relative
paths, but this should take care of other configuations.
The problem arose when using a .bbappend to add files to a recipe in
another layer. If the layer is located alongside the other layers and
their parent directory is moved, the hash does not change. However,
moving the .bbappend layer outside of the common directory can change
the path sort order and the task hash.
(Bitbake rev: 22bd19d208f0251f5a1f9b98f3cac66181f3fc07)
Signed-off-by: Tyler Hall <tylerwhall@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
- do not use the BB_URI_LOCALCOUNT database for computing revision
incremental numbers anymore
- sortable_revision now generates "AUTOINC+${latest_rev}"
- use one incrementing value rather than several
- PV becomes 0.1+gitAUTOINC+deadbeefdecafbad_decafbaddeadbeef
- remove all localcount code and simplify the fetcher
- this patch addresses the following proposal:
http://lists.linuxtogo.org/pipermail/bitbake-devel/2012-November/003878.html
(Bitbake rev: 61cf01c5c236b4218f40cfae7c059c2b86765dbd)
Signed-off-by: Constantin Musca <constantinx.musca@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Currently there is code which uses FILESDIR in unpack to ensure
parent directories are created, leading to differing behaviour depending on
which search path is used to locate the directory.
This change standardises the code and takes the data from the fetcher in
question meaning we can standardise the code and deprecate FILESDIR.
(Bitbake rev: 1cccb3bd01ed82e4978acfef0fda1bd797eef72a)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add "-d" option to cpio since it is useful:
-d
--make-directories
Create leading directories where needed.
[YOCTO #3137]
(Bitbake rev: a78f9ded7896432b107f34c0bb608b389fdb676a)
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
* Don't print the full exception in the initial warning - if we later
succeed in fetching the file from a mirror, we won't usually need the
details (which are in the fetch log if they are needed); otherwise the
full error will be printed when the fetch operation fails. Also adjust
the conditional block so that we don't print another warning just
mentioning we're going to try mirrors.
* Call logger.error() so that with knotty the full log is not printed
* Provide an explanation around the lines we print for easily updating
the checksums in the recipe. We don't want users to be just blindly
updating the recipe in case of a transient failure or deliberately
altered remote file.
(Bitbake rev: 2793413106c925b06783beb7413aa87cbcf246c3)
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Most of the time we don't need to see the fetch command; the fetch log
includes the command as a debug message in any case, so omit it. Also
adjust the way command output is printed (we don't need stderr/stdout
labelled, and print "no output" instead of "output:\nNone" when there is
no output.
(Bitbake rev: a75505a52e4da918222100221f79e8a658f90446)
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The implementation of NoMethodError and MalformedUrl was broken - if you
just set self.args in an exception class to a string it treats it as a
list and then fails later on with a TypeError due to the number of
arguments not matching up.
This nasty exception during exception handling was breaking the normal
exception flow (fixed separately), which meant that if you had a
malformed URL or invalid protocol in SRC_URI you would get the
following:
ERROR: Command execution failed: Traceback (most recent call last):
File "/home/user/poky/poky/bitbake/lib/bb/command.py", line 84, in runAsyncCommand
self.cooker.updateCache()
File "/home/user/poky/poky/bitbake/lib/bb/cooker.py", line 1207, in updateCache
if not self.parser.parse_next():
File "/home/user/poky/poky/bitbake/lib/bb/cooker.py", line 1694, in parse_next
logger.error('Unable to parse %s', value.recipe,
AttributeError: 'exceptions.TypeError' object has no attribute 'recipe'
A specific fix for [YOCTO #2977].
(Bitbake rev: 9d4150d99051d24ff218e8a43664ceaf524b19c7)
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This ensures that if all a MIRRORS entry does is add a slash, this does
not result in a circular loop.
Fixes [YOCTO #3073].
(Bitbake rev: 57055d337a2c9997a6e5d5bdabaec396e3e128e9)
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
* Unpack the ".rpm" binary package (only .src.rpm in the past)
* Unpack the .deb and .ipk binary package, their unpack commands are the same.
* This is useful for binary package recipe.
[YOCTO #1592]
(Bitbake rev: de7ceb9459574f33920ccc06255b533434f0ec25)
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
* in form which can be copied into a recipe without modification
* like oe-classic did since:
http://git.openembedded.org/openembedded/commit/?id=68abc465559a68e9201c9525be3c4acc6297eaed
* it shows them in right form when they are missing completely, but in
more verbose form when different
* it needs to print that only when checksums were requested, e.g.
fetching from sstate mirror sets both md5mismatch and sha256mismatch,
but your checksums shouldn't be shown
(Bitbake rev: 9eb34fc866775fd8310759a0111f232a9dc98981)
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If a file:// mirror is being used, the fetcher will create a symlink to the
local file. However, if the local file gets removed, that link will be dead,
and os.path.exists() returns False in that case, so it tries and fails to
recreate the link. Now we unlink such a dead link if it exists.
(Bitbake rev: 229ed3857e826e3e215e843cb51f729c1e13ed37)
Signed-off-by: Christopher Larson <chris_larson@mentor.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Without the new exception, when the system is configured to use premirrors,
but not allow network access (via BB_NO_NETWORK), when a recipe was lacking a
checksum the wrong error message(s) were being generated.
Instead of complaining about trying to perform network access, if the system
was able to find the item in the premirror, it should inform the user of the
lack of checksums, and the two SRC_URI fields they should use to update their
recipe.
(Bitbake rev: cb10e9c03a3f96d94e27e18330009616dde5e2b3)
Signed-off-by: Mark Hatle <mark.hatle@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
As well, if "proto" is used, get the associated value as "protocol"
(Bitbake rev: 53e6b630f0463d2d07cdaa9c9eb36794dc9b6b69)
Signed-off-by: Andrei Gherzan <andrei@gherzan.ro>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
People are using regexps in the url type field so we need to preserve
this bitbake behaviour. To address the issues with https:// urls mapping
badly to file:// urls we anchor the regexp if its not already anchored.
There should be no expressions in the wild which would break with this
change.
(Bitbake rev: ce0579dc256251e523c6330641f98b9f5a0e5761)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If '*' does end up in mirror urls accidently, some strange things
can break since supports_checksum() looks for this, ud.localpath can
then get ignored and this can lead to empty directories being downloaded
"successfully". '*' is a special case for file urls only at this point
so remove any entries that accidentlly make it in through url mapping.
(Bitbake rev: 1369bec2404d942acc3618a8d005ec6868dcfd41)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
When writing mirror specifications, the current regexp syntax can be awkward
and hard to get it to do what you want. For example, extracting the 'basename'
of a repository:
PREMIRRORS = "git://.*/([^/]+/)*([^/]*) git://somewhere.org/somedir/\\2;protocol=file"
can now become:
PREMIRRORS = "git://.*/.* git://somewhere.org/somedir/BASENAME;protocol=file"
which is much clearer. A MIRRORNAME substitution is also added which contains
an encoded form of both host and path. One of the problems with the existing
regexp syntax is you couldn't access HOST information from PATH and vice-versa
which is an issue this patch also addresses.
Tests for the new syntax are also added.
(Bitbake rev: c6b1acbad7b3d2698530eb8b5249adb4ab95da21)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
When we build the mirror urls, its possible an error will occur. If it
does, it should just mean we don't attempt this mirror url. The current
code actually aborts *all* the mirrors, not just the failed url.
This patch catches and logs the exception allowing things to continue.
(Bitbake rev: c35cbd1a1403865cf4f59ec88e1881669868103c)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
For mirror mapping, we never use regexps for the type component of
the url. Doing so causes various slightly bizarre behaviour such as
https:// urls being translated to files:// urls which we have no handler
for. This patch forces the type matches to be direct, not regexp based
and gives the expected bahvriour.
(From Poky rev: 695d8aca0999d2d61970b990e83c3132ba6f12cb)
(Bitbake rev: 604df1b25cf114e083f52917df2df64e01279c25)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Using ud.localpath as a basename when it points at a directory causes
problems. The supports_checksum() method gives a good indication of whether
ud.localpath can be used in the way we need.
(From Poky rev: 933ec8a44634e33f92f6f76de3a34094c3d63aa6)
(Bitbake rev: dcd79ae20ab2c72c3312b2251c2b6dc4cabe988e)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We only consider mirror tarballs when the source and target urls are of
differing types. We also should clear all url paramters when handling
mirror tarballs.
(From Poky rev: da140b8b0b3dda5429f9eee68829ef5247cdfe12)
(Bitbake rev: 7619dcad29a6c2405b15a8fbadfa11e81b399ae1)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This means that parameters in the source expression are used as part of the
match. Parameters in the destination are used explicitly in the final
url.
(From Poky rev: c465cb0c5c927dd41d96ad6d6fa1566349574bb7)
(Bitbake rev: 3c468ac3ed6d045561afce19b85ae9dd18d87cea)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Robert Yang
Richard Purdie
Martin Jansa
André Draszik
Peter Kjellerstedt
Matthew McClintock
Olof Johansson
Darren Hart
Tyler Hall
Constantin Musca
Paul Eggleton
Christopher Larson
Mark Hatle
Andrei Gherzan