On Jan 30, 2008 12:57 PM, C. Michael Pilato <cmpilato_at_collab.net> wrote:
> Augie Fackler wrote:
> >
> > On Jan 30, 2008, at 2:41 PM, C. Michael Pilato wrote:
> >
> >> Of interest (to me, at least) is how we managed to get ahold of the
> >> non-canonicalized path that we currently write to the entries file. I
> >> mean, the first thing our command-line client does with the paths
> >> provided to subcommands is canonicalize them, I thought. What gives?
> >
> > The path is the computed repository root - I haven't figured out in the
> > code where that happens yet, but it looks to my eye like the repository
> > root is being computed with the hostname and the path separate, so that
> > when the path is just / it isn't stored in a canonical form, but if it
> > was at /foo/ it would get stored as /foo
> >
> > I was seeing the bug when I was checking out
> > http://svn.perian.org/trunk, which isn't the repo root, so svn must be
> > computing the repo root of http://svn.perian.org/ on its own somewhere...
>
> Oh, yes, right -- forgot that you were probably checking out something
> deeper than the root. Yes, each RA implementation does this sort of root
> calculation. Your solution will work, but still leaves us with
> non-canonicalized paths being passed around the libraries in memory during
> the checkout. I would suggest making the four RA layers do this
> canonicalization at the moment they discern the repository root.
+1
Augie, do you think you'll get a chance to do this?
It's a shame that our test infrastructure doesn't provide an easy way
to write a test for this condition.
--dave
--
David Glasser | glasser@davidglasser.net | http://www.davidglasser.net/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe_at_subversion.tigris.org
For additional commands, e-mail: dev-help_at_subversion.tigris.org
Received on 2008-01-30 22:01:23 CET