Missing svn commit log message in ticket comment

[joop.van.bussel@…]

Ocassionally we are missing svn commit log messages referring a ticket as a ticket comment. There are no strange messages in the trac log. Is this a know bug? Are there conditions when the creation of the comment is left out?

Configuration of the commit updater:

commit_ticket_update_envelope =
commit_ticket_update_commands.close = #!NO#!
commit_ticket_update_commands.refs = <ALL>
commit_ticket_update_check_perms = false
commit_ticket_update_notify = true

[Ryan J Ollos]

We see this issue from time to time on trac-hacks.org. I have no clues as to what the cause might be.

[fschophuizen@…]

The issue is not listed as a known issue on the CommitTicketUpdater page. Where was this issue discussed before on track-hacks.org?

[Ryan J Ollos]

Replying to fschophuizen@…:

The issue is not listed as a known issue on the CommitTicketUpdater page. Where was this issue discussed before on track-hacks.org?

It was mentioned in several tickets on trac-hacks.org over the years, but there isn't any useful information in those tickets so I won't bother trying to search for them now.

I wonder if the issue could be similar to #11767, see comment:2:ticket:11767. Are you using explicit synchronization in the post-commit hook and have a non-empty value for [trac] repository_sync_per_request?

After reading comment:2:ticket:11767 I checked the configuration on trac-hacks.org and see that we do have [trac] repository_sync_per_request = (default) and are using explicit synchronization in SVN's post-commit hook (using tags/trac-1.0.1/contrib/trac-svn-hook​). I'll change to [trac] repository_sync_per_request = later this evening and do some testing.

[Jun Omae]

Replying to rjollos:

After reading comment:2:ticket:11767 I checked the configuration on trac-hacks.org and see that we do have [trac] repository_sync_per_request = (default) and are using explicit synchronization in SVN's post-commit hook (using tags/trac-1.0.1/contrib/trac-svn-hook​). I'll change to [trac] repository_sync_per_request = later this evening and do some testing.

Should we check whether repository is not listed in the option when changeset added and changeset modified are called, like this?

trac/versioncontrol/api.py

@@ -671 +671 @@
                           base or reponame)
             return
         for repos in sorted(repositories, key=lambda r: r.reponame):
+            reponame = repos.reponame or '(default)'
+            if reponame in self.repository_sync_per_request:
+                self.log.warn("Repository '%s' should be removed from [trac] "
+                              "repository_sync_per_request for explicit "
+                              "synchronization", reponame)
             repos.sync()
             for rev in revs:
                 args = []
@@ -685 +690 @@
                     except NoSuchChangeset:
                         continue
                 self.log.debug("Event %s on %s for revision %s",
-                               event, repos.reponame or '(default)', rev)
+                               event, reponame, rev)
                 for listener in self.change_listeners:
                     getattr(listener, event)(repos, changeset, *args)
 

[Ryan J Ollos]

Replying to jomae:

Should we check whether repository is not listed in the option when changeset added and changeset modified are called, like this?

Yeah, that looks like a good idea. Your changes test out well for me on the trunk.

[Ryan J Ollos]

Replying to rjollos:

I wonder if the issue could be similar to #11767, see comment:2:ticket:11767. Are you using explicit synchronization in the post-commit hook and have a non-empty value for [trac] repository_sync_per_request?

Using explicit synchronization on every request seems like it would reduce performance, but I haven't yet found a mechanism by which it might prevent the post-commit hook from inserting a changeset message. The mechanisms I had in mind don't seem probable based on further study of the code, but I'll do some more testing to be sure.

[Ryan J Ollos]

Replying to rjollos:

Replying to jomae:

Should we check whether repository is not listed in the option when changeset added and changeset modified are called, like this?

Yeah, that looks like a good idea. Your changes test out well for me on the trunk.

Change committed to 1.0-stable in [13337], merged in [13338].