trac-post-commit-hook and codepages

[GregZ]

Sorry for bad English.

I'am writing comments for SVN revisions on Russian.
To see this comments correctly in timeline page, I'am set UNICODE UTF-8 encoding in my web browser (Opera).

And whan I'am using trac-post-commit-hook the same coomments that fixed and closed tickets displayed in Codepage 866.

Switching codepages very inconvenient to use.

Server and Workstation OS: Window XP SP2

Буду рад любой помощи.

[cmlenz]

Why did you close this, anonymous?

[GregZ]

To solve this problem, i'am added 2 lines to the trac-post-commit-hook:

    def __init__(self, project=options.project, author=options.user, rev=options.rev,msg=options.msg):
        self.author = author
        self.rev = rev

this>>        csmsg = "(In [%s]) %s" % (rev, msg)
this>>        self.msg = util.to_utf8(csmsg, 'windows-1251')

        self.now = int(time.time()) 
        self.con = sqlite.connect(os.path.join(project, 'db', 'trac.db'), autocommit=0) 

[cmlenz]

This is not high-priority, being just a contrib script.

[anonymous]

What svn client you use for writing comments ? It's strange since nornal subversion encoding is utf-8.

PS Можно пообщаться об этом напрямую - email/jabber:vyt@vzljot.ru

[vyt@…]

I think that problem is a way for getting log message.

svnlook does recoding to current locale charset and no way for disable this recoding. Charset detection in post-commit hook is overhead, so best way, IMHO - using subversion python API for getting log message.

[dans]

vyt >>

That sounds reasonable, but I'm feeling a bit lost here.

I'm on Windows XP, running the "post-commit.cmd" from #1602. I guess I should replace the line

FOR /F "usebackq delims==" %%i IN (`%%SVNLOOK%% log -r %TXN% %REPOS%`) DO SET LOG=%%i

with something calling the Python-SVN API, but I neither fully understand the current line nor have any clue about the API.

Can I just replace the part between the parentheses? And if so, what do I replace it with?

[vyt@…]

At least trac-post-commit-hook should contain note about invoking svnlook in non UTF-8 locales like LANG=ru_RU.UTF-8 svnlook...

[szaman]

i did:

LC_ALL="pl_PL.UTF-8"

in MYREPO/hooks/post-commit

Think, that should be pointed in hook's documentation.

(I'm working on trac-0.9.3 and trunk/contrib/trac-post-commit-hook)

[Markus Tacker <m@…>]

#1625 has been marked as duplicate of this bug.

[cboos]

#2845 has also been marked as duplicate of this bug.

[cboos]

#2352 is yet another duplicate.

[Markus Tacker <m@…>]

#3732 has been marked as duplicate of this bug.

[cboos]

He, not sure that #3732 is really a duplicate of this one, as I wrote a patch that would have fixed #3732, I believe. We'll see.

Can you try out the following patch?
(be careful, it's for 0.10, don't apply on 0.9)

Index: trac-post-commit-hook
===================================================================
--- trac-post-commit-hook	(revision 3720)
+++ trac-post-commit-hook	(working copy)
@@ -78,6 +78,7 @@
 from trac.ticket import Ticket
 from trac.ticket.web_ui import TicketModule
 # TODO: move grouped_changelog_entries to model.py
+from trac.util.text import to_unicode
 from trac.web.href import Href
 
 try:
@@ -101,6 +102,8 @@
                   help='The user who is responsible for this action')
 parser.add_option('-m', '--msg', dest='msg',
                   help='The log message to search.')
+parser.add_option('-c', '--encoding', dest='encoding',
+                  help='The encoding used by the log message.')
 parser.add_option('-s', '--siteurl', dest='url',
                   help='The base URL to the project\'s trac website (to which '
                        '/ticket/## is appended).  If this is not specified, '
@@ -132,7 +135,9 @@
                        'see':        '_cmdRefs'}
 
     def __init__(self, project=options.project, author=options.user,
-                 rev=options.rev, msg=options.msg, url=options.url):
+                 rev=options.rev, msg=options.msg, url=options.url,
+                 encoding=options.encoding):
+        msg = to_unicode(msg, encoding)
         self.author = author
         self.rev = rev
         self.msg = "(In [%s]) %s" % (rev, msg)

As normally I think you'd get UTF-8 strings when Subversion calls the post-commit hook, the -c/--encoding option shouldn't be needed.

However it can be useful when testing the script from the command line, or if for some reason the post-commit hook is actually given the message with a different encoding.

[sier@…]

Replying to cboos:

He, not sure that #3732 is really a duplicate of this one, as I wrote a patch that would have fixed #3732, I believe. We'll see.

The patch works. Thanks!!

[cboos]

Ok, patch applied in r3743, and I'm going to close this, so we're going to find out if #3732 was really a duplicate ;)

[Marksu Tacker <m@…>]

Confirmed on r3747

[cboos]

Confirmed what, the fix or the bug? I assume the former, otherwise I guess you'd have reopened ;)

[blaufalke@…]

doesn't work for me and my trac 0.10.
The changset itself displays the log message with the correct German umlauts, but the fixed / addressed tickets are generated faulty - even with trac-post-commit-hook from the current trunk.

Expected result

(In [353]) "Testcommit für den trac-post-commit-hook addresses #2 fixes #1"

Actual result:

(In [353]) "Testcommit f?\195?\188r den trac-post-commit-hook addresses #2 fixes #1"

[anonymous]

Replying to GregZ:

Sorry for bad English.

I'am writing comments for SVN revisions on Russian.
To see this comments correctly in timeline page, I'am set UNICODE UTF-8 encoding in my web browser (Opera).

And whan I'am using trac-post-commit-hook the same coomments that fixed and closed tickets displayed in Codepage 866.

Switching codepages very inconvenient to use.

Server and Workstation OS: Window XP SP2

Буду рад любой помощи.

[l2k]

I have the same problem as blaufalke

[cboos]

See TracDev/Proposals/Journaling. I think that we could eventually get rid of the trac-post-commit-hook in the future (and hence get rid of the encoding issues as well).

[stock@…]

I did the following changes to the script, which now uses the trac api to get the change message. So the conversion to locale is avoided, I think.
It works for me with german umlauts entered through TortoiseSVN.

trac-post-commit-hook

@@ -80 +80 @@
 # TODO: move grouped_changelog_entries to model.py
 from trac.util.text import to_unicode
 from trac.web.href import Href
+from trac.versioncontrol.svn_fs import SubversionRepository, SubversionChangeset
+from trac.log import logger_factory
 
 try:
     from optparse import OptionParser
@@ -137 +139 @@
     def __init__(self, project=options.project, author=options.user,
                  rev=options.rev, msg=options.msg, url=options.url,
                  encoding=options.encoding):
-        msg = to_unicode(msg, encoding)
+        self.env = open_environment(project)
+        repos_dir = self.env.config.get('trac', 'repository_dir')
+        repos = SubversionRepository(repos_dir,
+                             '', logger_factory('test'))
+        change = repos.get_changeset(rev)
+        msg = change.message
+        to_unicode(msg)
+        msg = msg.decode('utf-8')
         self.author = author
         self.rev = rev
         self.msg = "(In [%s]) %s" % (rev, msg)
         self.now = int(time.time()) 
-        self.env = open_environment(project)
         if url is None:
             url = self.env.config.get('project', 'url')
         self.env.href = Href(url)
@@ -194 +202 @@
 
 
 if __name__ == "__main__":
-    if len(sys.argv) < 5:
+    if len(sys.argv) < 4:
         print "For usage: %s --help" % (sys.argv[0])
     else:
         CommitHook()

[markus]

The above trac api solution works for me!

You could also get the author name that way, this would simplify the calling batch file:

self.author = author

becomes

self.author = change.author

as well as

if len(sys.argv) < 4:

becomes

if len(sys.argv) < 3:

The batch files in the hooks directories of all repositories simply call an other batch script in a directory outside the repository directory.
post-commit.cmd:

%~dp0\..\..\..\hooks\trac-post-commit-hook.cmd %1 %2

The trac-post-commit-hook.cmd (purified version from #1602) calls the python script. This script can be used for all the repositories as long as the trac environments have the same names as the subversion repositories:

@ECHO OFF

SET REV=%2
SET REPNAME=%~nx1

:: Modify paths and port number here
SET TRAC_ENV=D:\trac\trac_env\%REPNAME%
SET PYTHON_DIR=C:\Python24
SET TRAC_URL=http://%COMPUTERNAME%:8080/%REPNAME%
SET PYTHON="%PYTHON_DIR%\python.exe"

:: Do not execute hook if trac environment does not exist
IF NOT EXIST %TRAC_ENV% GOTO :EOF

%PYTHON% "%~dp0\trac-post-commit-hook.py" -p "%TRAC_ENV%" -r "%REV%" -s "%TRAC_URL%"

[markus]

Patch for international characters in changeset and tickets

[markus]

I attached a patch attachment:trac-post-commit-hook.0.10-stable.diff​ which applies all the changes from stock@… as well as the one I described above to the latest version from 0.10-stable. It works for me, cyrillic characters in changeset and the ticket comments are correctly displayed. The same changes could be applied to the trunk of 0.11.

[cboos]

A small fix to your patch:

repos_dir = self.env.config.get('trac', 'repository_dir') 
repos = self.env.get_repository() # will do a `sync` if needed 

(and then the log related import is not needed)

Besides the above, I think it's a step in the right direction.

[cboos]

woops, and I didn't see the msg = msg.decode('utf-8') line. This is unnecessary and potentially harmful.

[markus]

Patch for international characters in changeset and tickets v2

[markus]

I made a second patch attachment:trac-post-commit-hook.0.10-stable.2.diff​ containing your fixes. I hope this is better (it's my first day with python, I'm just trying to put together a patch from what works on my windows box…). I tested it with trac 0.10.3 and subversion 1.4.2, french accents and cyrillic characters work fine.

[cboos]

Thanks for the patch and the tests. I've committed a slightly different version, which won't break compatibility with the previous script.

Fixed in r4532, r4533 (trunk) and r4534 (0.10-stable).

Please try out once again to let me know if everything works for you.

[markus]

Works great for me, thanks!