This improves response from kio_ftp, because the SIZE request will no longer
report 0 length. However, LIST etc may still report 0, which will confuse
some clients. Solution is non-trivial, because the length can only be
computed after the content is generated.
bzr revid: p_christ@hol.gr-20100713103102-v5uou8uuqvqxm6yc
At the nlst command, the paths from the API could have been unicode, in
which case the '\r\n' separator would be sent as '\00\r\00\n' over the
wire (that's utf-16 encoding).
For that, bring the utf-8 conversion functions inside ftpserver.
bzr revid: p_christ@hol.gr-20100708225433-fqpl911kqfsrm1ff
the FTP server has been limited to only serve databases that have the
document_ftp module "installed". However, this would break running the
tests, because the module would be marked as "to upgrade".
bzr revid: p_christ@hol.gr-20100706232928-5ckj97p61nfrq3rn
A big patch: nodes should not be calculated multiple times from the path
string, within an operation. Instead, they are resolved only once in a
node object (may reuse the cwd_node, if possible) and then op relative
to that node.
bzr revid: p_christ@hol.gr-20100702215249-kzucr3i1p5f6vvnt
The document_ftp is not really a database-specific server, but a
server-wide implementation. So, it could normally serve all the databases
of the server, once one of them has it installed. Limit that and instead
only serve the dbs that have the module marked as installed.
bzr revid: p_christ@hol.gr-20100701175132-dydehiw5m6nnvcwy
Trying a login on each db listed would result in the database loaded by
the pooler. This behaviour is not really welcome.
However, some clients (like kio_ftp) may still want to descend into the
databases listed, in which case we cannot avoid to load the db(s).
bzr revid: p_christ@hol.gr-20100629140328-2kqlfpo3oyhu0ufi