Greg Hudson wrote:
> On Sun, 2004-07-04 at 11:21, Josh Pieper wrote:
> > Yes, I considered that case; this also stopped generating an error
> > when you try "invalid_path/../real_path." Apache does this when
> > resolving URLs to filesystem paths, so at least one other application
> > handles paths the same way. Of course that doesn't mean it is a valid
> > method, but collapsing the ".." makes the copy code more robust to
> > differing path specifications in the common cases.
>
> Apache can dictate that directory symlinks are disallowed in document
> roots. I don't think it's fair for us to dictate that about all paths
> on the command line, only about the area within working copies (and, of
> course, the virtual area within repositories).
>
> I followed the pre-commit discussion about this change, but the
> discussion of "differing path specifications" never seemed to become
> concrete. Where is it necessary to ensure that canon(p1) == canon(p2)
> if p1 and p2 are two different paths expressing the same object?
Ok, in IRC ghudson proposed:
<ghudson> Well, I'd agree with your initial judgement that we
shouldn't be collapsing .. segments, and then if we changed the
apr_filepath_merge() call in opt.c to an svn_path_canonicalize, we'd
fix the case I mentioned was broken.
I don't believe we considered the symlink case when breser and I
initially talked on IRC about supporting collapsing "/../" elements.
Does anyone have objections to just removing the ".." collapsing
support from svn_path_canonicalize? That would also resolve the issue
of implementing svn_path_canonicalize in terms of apr_filepath_merge.
-Josh
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sun Jul 4 17:59:22 2004