Re: svn commit: r18204 - trunk/contrib/hook-scripts

Martin Tomes <lists@tomes.org> writes:

> Philip Martin wrote:
>> martinto@tigris.org writes:
>>> Author: martinto
>>> Date: Tue Jan 24 10:16:30 2006
>>> New Revision: 18204
>>> Log:
>>> * contrib/hook-scripts/case-insensitive.py:
>>> Output the 'clash' text as utf-8, previously ascii was used and
>>> non-ascii characters in files names caused the error message at the
>>> subversion client to read 'svn: General svn error from server'.
>>> Non ascii names such as Àbc.txt and àbc.txt are now displayed at
>>> the client when they clash.
> >
>> Forcing the hook script to output UTF-8 is not correct, hook scripts
>> are supposed to output in the native encoding[1]. The first thing that
>> libsvn_repos does with the hook output is convert from the native
>> encoding to UTF-8, it makes no sense for the hook to output UTF-8
>> unless the native encoding is also UTF-8. I believe the original code
>> was correct.
>> [1] httpd might be an exception since it doesn't set up the locales,
>> in that case hook scripts probably need to restrict their output
>> to plain ascii.
>
> Would it be possible to detect whether the hook script is being fired
> off from apache or svnserve?

Not easily, I suppose you could do something platform specfic and look
at parent process IDs.

> We have companies in France and Italy
> using Subversion so this hook really should detect non-ascii case
> conflicts.

It does detect such conflicts. Using the r18203 version of the
script, i.e. without your change:

$ LANG=en_GB svnserve -dr.

$ LANG=en_GB svn mkdir svn://localhost/repo/xxx`printf "\xe3"`
Committed revision 1.

$ LANG=en_GB svn mkdir svn://localhost/repo/xxx`printf "\xc3"`
../svn/subversion/libsvn_repos/hooks.c:131: (apr_err=165001)
svn: 'pre-commit' hook failed with error output:
Clash: '/xxxÃ' '/xxxã'

../svn/subversion/svn/util.c:417: (apr_err=165001)
svn: Your commit message was left in a temporary file:
../svn/subversion/svn/util.c:417: (apr_err=165001)
svn: 'svn-commit.tmp'

I've shown the client and server using the same locale, but that's not
necessary. What is necessary is that the server and the hook must use
the same locale. [Note: I'm only dealing with message encoding here,
I'm ignoring the whole question of whether using a locale's
'convert-to-lowercase' is sufficient to detect name clashes and
whether client and server need to use the same locale for it to work.]

What I've shown above won't work with ra_dav because the httpd process
doesn't make a setlocale call and so httpd itself cannot do the
native/UTF-8 conversions for the non-ascii characters in the hook
output. To work with httpd the script needs to force the output to
ascii using something like (again using the r18203 version):

Index: contrib/hook-scripts/case-insensitive.py
===================================================================
--- contrib/hook-scripts/case-insensitive.py (revision 18203)
+++ contrib/hook-scripts/case-insensitive.py (working copy)
@@ -85,6 +85,7 @@
       clashes[canonical_path][join_path(dir, name_pair[1])] = True
 
 if (clashes):
+ native = 'ascii' # for httpd
   for canonical_path in clashes.iterkeys():
     sys.stderr.write(u'Clash:'.encode(native))
     for path in clashes[canonical_path].iterkeys():

$ LANG=en_GB svn mkdir http://localhost:8888/obj/repo/XXX`printf "\xe3"`
../svn/subversion/libsvn_ra_dav/util.c:827: (apr_err=165001)
svn: MERGE request failed on '/obj/repo'
../svn/subversion/libsvn_ra_dav/util.c:389: (apr_err=165001)
svn: 'pre-commit' hook failed with error output:
Clash: '/xxx?' '/xxx?'

../svn/subversion/svn/util.c:417: (apr_err=165001)
svn: Your commit message was left in a temporary file:
../svn/subversion/svn/util.c:417: (apr_err=165001)
svn: 'svn-commit.2.tmp'

Forcing ascii output has the disadvantage that any non-ascii
characters show up as question marks, but there is no alternative if
using httpd.

-- 
Philip Martin
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org