Re: [PATCH 0/2] git-p4: Improve client path detection

From: Vitor Antunes <vitor.hda@gmail.com>
Date: Apr 5, 2015, 22:57 UTC
Message-ID: <20150405235759.392c0f2b@pt-vhugo>
In-Reply-To: <55218C8F.209@diamand.org>

Luke Diamand <luke@diamand.org> wrote on Sun, 05 Apr 2015 20:27:11 +0100

Show 14 quoted lines

> On 28/03/15 12:28, Vitor Antunes wrote:
> > I'm adding a test case for a scenario I was confronted with when using branch
> > detection and a client view specification. It is possible that the implemented
> > fix may not cover all possible scenarios, but there is no regression in the
> > available tests.
>
> Vitor, one thing I wondered about with this part of the change:
>
> -            if entry["depotFile"] == depotPath:
> +            if entry["depotFile"].find(depotPath) >= 0:
>
> Does this mean that if 'p4 where' produces multiple lines of output that
> this will get confused, as it's just going to search for an instance of
> depotPath.

The reason why I introduced that was because in the test case I implemented (and which reflects a scenario I am confronted with in my workplace) the branches have a base directory that is removed in the client view mapping. As such, we will have a situation where depotPath is //depot/branch1/ while runninng "p4 where" will result in //depot/branch1/base/. To overcome this I used find() instead of a direct comparison. Now that I think about that, I could probably have used the simpler `if depotPath in entry["depotFile"]`...

Show 7 quoted lines

> The example in the Perforce man page for 'p4 where' would trigger this
> for example:
>
> http://www.perforce.com/perforce/r14.2/manuals/cmdref/p4_where.html
>
> -//a/b/file.txt //client/a/b/file.txt //home/user/root/a/b/file.txt
> //a/b/file.txt //client/b/file.txt /home/user/root/b/file.txt

These are examples where a simple comparison as was implemented would work.

> As an experiment, I hacked git-p4 to always use p4Where rather than
> getClientRoot(), which I would have thought ought to work, but while
> most of the tests passed, Pete's client-spec torture tests failed.

That was exactly my first approach and got to the same conclusion. I would have investigated it further but since I haven't had much free time to invest in solving this problem I decided to implement an intermediary solution that would not introduce any regressions.

Vitor