Re: [PATCH v3 2/5] setup: allow for prefix to be passed to git commands

From: Brandon Williams <bmwill@google.com>
Date: Mar 17, 2017, 21:00 UTC
Message-ID: <20170317210031.GA63813@google.com>
In-Reply-To: <xmqqy3w363yx.fsf@gitster.mtv.corp.google.com>

On 03/17, Junio C Hamano wrote:

Show 10 quoted lines

> Brandon Williams <bmwill@google.com> writes:
> 
> > ...  I was being cautious with this patch since git didn't currently
> > read GIT_PREFIX.
> 
> Ahh, I forgot about that.  Processes we spawn do expect GIT_PREFIX
> to tell them where the original $cwd was and they also do expect
> that "git" invoked by them would not be affected by GIT_PREFIX
> environment variable.  So we cannot change that now.
>

Yep. I just ran the test suite locally doing this...it fails on a bunch of stuff :)

> If you recurse into sub-sub module, it is likely that you would want
> to update the TOPLEVEL_PREFIX relative to that sub-sub module you
> are descending into.

So the idea would be that 'super_prefix' would keep getting updated when recursing deeper, but the 'prefix' should stay the same (in what I'm proposing here that is).

Show 10 quoted lines

> 
> That probably also means that processes we spawn now need to also
> pay attention to TOPLEVEL_PREFIX in addition to GIT_PREFIX, and we
> should NOT re-export what we got from TOPLEVEL_PREFIX to GIT_PREFIX.
> I.e. if a "git" process started from src/ subdirectory of the
> superproject that goes into module/sub1/ submodule, top-level prefix
> may export ../src/ to point at the original location, but the
> process that is running in the submodule will be running at the root
> level of the submodule working tree, so its prefix should be NULL or
> "", no?

I don't think that prefix can ever have ".." in it. From what I understand it is always a path from the root of the repository to the cwd that the git command was invoked by. So in your example prefix would be "src/".

Show 8 quoted lines

> 
> Adjusting pathspec and other file references on the caller's side,
> instead of exporting toplevel-prefix to have them adjusted by the
> callee, started to smell more and more like an easier/more correct
> approach to me, but perhaps I haven't thought things deeply enough.
> 
> I dunno.
>

Yeah, doing all of the adjusting on the caller's side is quite difficult. Getting the pathspecs and file references right seems to be overly complicated if done by the caller. That why I opted to punt to the callee; given enough information (e.g. prefix and super_prefix) the callee should be able to handle both the pathspecs and file references properly.

-- 
Brandon Williams