[RFH] - git-svn auth bug (possibly SVN 1.5.0-related)

[Eric Wong <normalperson@yhbt.net>]

Brad King <brad.king@kitware.com> wrote:
> Eric Wong wrote:
> > Brad King <brad.king@kitware.com> wrote:
> >> Eric Wong wrote:
> >>> I haven't had the chance to look at this.   Can anybody else shed more
> >>> light on that bug?  It's really strange that the tests won't run because
> >>> of it.  Are you unable to run some git-svn tests or all of them?
> >> Just that one fails.  All others (including the one in the patch below) pass.
> > 
> > Exactly which test fails for you?  Perhaps it's some setting in your
> > ~/.subversion/config that's causing it to fail.  Maybe we should set
> > $HOME and use a clean ~/.subversion/config for git-svn tests regardless
> > if that turns out to be the case...
> 
> $ cd $gitsrc/t
> $ export SVNSERVE_PORT=5432
> $ ./t9113-git-svn-dcommit-new-file.sh
> *   ok 1: start tracking an empty repo
> * FAIL 2: create files in new directory with dcommit
> 
>                 mkdir git-new-dir &&
>                 echo hello > git-new-dir/world &&
>                 git update-index --add git-new-dir/world &&
>                 git commit -m hello &&
>                 start_svnserve &&
>                 git svn dcommit
> 
> * failed 1 among 2 test(s)
> 
> I hacked the test script to log the dcommit output to a file, and I see
> this:
> 
> Committing to svn://127.0.0.1:5432 ...
> Use of uninitialized value in concatenation (.) or string at
> /usr/lib/perl5/SVN/Core.pm line 584.
> Authorization failed:  at $gitsrc/t/../git-svn line 3329
> 
> (I replaced my git source dir full path with $gitsrc).
> 
> The version of libsvn-perl is:  1.5.0dfsg1-4

It could be another incompatibility introduced in SVN 1.5.0.

I'll try to dist-upgrade a machine to Lenny sometime in the next two
weeks so I can test; I'm pretty busy these days but if anybody else
wants to figure this out in the meantime, please do :)

> Please let me know if you need more info.
> 
> I tried moving my ~/.subversion/config out of the way but it makes no
> difference.  However, I agree we should block the user's home svn config
> when running other dcommit tests now that we have auto-props.  Perhaps
> just using the --config-dir option with an empty directory would be enough.

Thanks for the additional info!

-- 
Eric Wong