Re: CGI for repository administration

Karl Fogel <kfogel@newton.ch.collab.net> writes:

> 3831 was committed for a reason. An actual bug had been encountered
> -- I don't remember what, but cmpilato may remember. We shouldn't
> just back it out.

I think the problem was that 'svn add foo' would schedule foo for
addition even if foo was a symbolic link. Then 'svn commit' would do
something nasty, I can't remember if it committed a file or produced
some nasty error message, but neither would be "correct". The current
behaviour is for 'svn add foo' to complain if foo is a symbolic link.

On reflection, svn_wc_add is not the only place that needs to reject
symbolic links. Consider

$ touch foo bar
$ svn add bar
$ rm bar
$ ln -s foo bar
$ svn commit

So commit still has to handle symbolic links, even if it just aborts.
[It does abort at present, but with a rather cryptic error
svn_io_copy_file: error copying /home/pm/sw/subversion/obj/wc/bar to /home/pm/sw/subversion/obj/wc/.svn/tmp/text-base/bar.svn-base.63132.00001.tmp
We really should do better than rely on a failed apr_file_copy.]

Here's another related bug

$ touch foo bar
$ svn add bar
$ rm bar
$ ln -s foo bar
$ svn st
? foo

The status command doesn't show bar at all.

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