On Fri, Nov 6, 2009 at 9:45 AM, Paul Burba <ptburba_at_gmail.com> wrote:
> On Fri, Jun 19, 2009 at 3:08 PM, Bert Huijben <rhuijben_at_sharpsvn.net> wrote:
>> Author: rhuijben
>> Date: Fri Jun 19 13:08:39 2009
>> New Revision: 38105
>> Make all response handlers in svn_ra_serf return svn_error_t* for
>> a much cleaner error handling. This fixes issue #3375, but some further
>> refactoring would be welcome.
>> * subversion/libsvn_ra_serf/commit.c
>> (handle_checkout): Updated for new handler prototype and replace abort()
>> with normal SVN_ERR_MALFUNCTION()
>> (post_response_handler): Updated for changed prototype.
>> * subversion/libsvn_ra_serf/locks.c
>> (handle_lock): Updated for new prototype. Return errors where possible.
>> * subversion/libsvn_ra_serf/options.c
>> (options_response_handler): Updated for new prototype.
>> * subversion/libsvn_ra_serf/ra_serf.h
>> (svn_ra_serf__response_handler_t): Updated to return svn_error_t * and
>> remove the temporary session argument.
>> (svn_ra_serf__handle_status_only, svn_ra_serf__handle_discard_body,
>> svn_ra_serf__handle_server_error, svn_ra_serf__handle_multistatus_only,
>> svn_ra_serf__handle_xml_parser): Updated to follow new prototype.
>> (SVN_SESSION_ERR): Removed macro.
>> * subversion/libsvn_ra_serf/update.c
>> (error_fetch): Return svn_error_t *
>> (handle_fetch, handle_stream): Update error handling for new prototype.
>> * subversion/libsvn_ra_serf/util.c
>> Implement new prototype.
>> Implement new prototype. Return errors instead of adding them to
>> the session. And remove an obsolete test.
>> Updated for new prototype. Clear ignored errors.
>> Use apr_status_t handler for discarding data, clear ignored
>> authentication errors. Compose session errors instead of overwriting
>> existing values. Add specific handling for serfs error handling.
> Hi Bert,
> Just an fyi:
> In addition to the error handling problem I already fixed in r40377,
> is seems this change is also responsible for triggering an assert in
> update_tests.py 57 'verify update of deleted locked files'. This test
> was not added until after your change, but on a hunch I checked if the
> test would pass with ra_serf back to r38104. It does, and then fails
> with the introduction of r38105.
> The test is marked as XFail so assert has gone unnoticed. I've
> attached the test run output for the test using ra_neon, where it
> fails "as expected" and from ra_serf where the assert occurs.
> This assert can quickly be replicated from the command line by locking
> a file, deleting the files containing directory and the updating.
> I haven't yet been able to figure out what abour r38105 is causing
> this. I'm out of the office today and will look at it more on Monday,
> but in the meantime if you have anything to add...
I couldn't get anywhere with this. Added issue #3532, which has a
reproduction recipe and a few more details
Received on 2009-11-17 20:14:55 CET