The toolchain_build git cache is now referenced as a git-alternate.

The toolchain_build git cache is now referenced as a git-alternate.

Having the origin fetch/push URL modified to point directly to the git
cache had some minor repercussions which forced us to implement various
work-arounds. By simply keeping the fetch/push URL to be the same and
using git-alternates to point to the git_cache, allows us to utilize
and turn off git_cache usage at will, and also makes our work-arounds
redundant.

R=dschuff@chromium.org
BUG= https://code.google.com/p/nativeclient/issues/detail?id=3852
TEST= toolchain_build trybots

Committed: https://src.chromium.org/viewvc/native_client?view=rev&revision=13411

[Derek Schuff, 2014-06-26 00:00:47]

https://codereview.chromium.org/355883002/diff/60001/toolchain_build/substitu...
File toolchain_build/substituter.py (right):

https://codereview.chromium.org/355883002/diff/60001/toolchain_build/substitu...
toolchain_build/substituter.py:76: if value:
how are empty paths getting in here? seems like we should avoid that?

[David Yen, 2014-06-26 00:12:53]

On 2014/06/26 00:00:47, Derek Schuff wrote:
>
https://codereview.chromium.org/355883002/diff/60001/toolchain_build/substitu...
> File toolchain_build/substituter.py (right):
> 
>
https://codereview.chromium.org/355883002/diff/60001/toolchain_build/substitu...
> toolchain_build/substituter.py:76: if value:
> how are empty paths getting in here? seems like we should avoid that?

Empty paths is how we are disabling the git cache. %(git_cache)s resolves to an
empty string and all the git_cache gets get turned off.

Seems like it is the cleanest way to handle it given the substitution pattern,
but I would be open to any ideas.

[Derek Schuff, 2014-06-26 18:17:56]

On 2014/06/26 at 00:12:53, dyen wrote:
> Empty paths is how we are disabling the git cache. %(git_cache)s resolves to
an empty string and all the git_cache gets get turned off.
> 
> Seems like it is the cleanest way to handle it given the substitution pattern,
but I would be open to any ideas.
Maybe in substituter.py it would be simpler just to make FixPath and AbsPath
handle empty input instead of having extra conditions in their callers.

[David Yen, 2014-06-26 19:03:20]

On 2014/06/26 18:17:56, Derek Schuff wrote:
> On 2014/06/26 at 00:12:53, dyen wrote:
> > Empty paths is how we are disabling the git cache. %(git_cache)s resolves to
> an empty string and all the git_cache gets get turned off.
> > 
> > Seems like it is the cleanest way to handle it given the substitution
pattern,
> but I would be open to any ideas.
> Maybe in substituter.py it would be simpler just to make FixPath and AbsPath
> handle empty input instead of having extra conditions in their callers.

The problem isn't FixPath, the problem is os.path.abspath and os.path.relpath.
Those functions raise an exception on empty values.

[Derek Schuff, 2014-06-26 20:31:31]

On 2014/06/26 at 19:03:20, dyen wrote:
> On 2014/06/26 18:17:56, Derek Schuff wrote:
> > On 2014/06/26 at 00:12:53, dyen wrote:
> > > Empty paths is how we are disabling the git cache. %(git_cache)s resolves
to
> > an empty string and all the git_cache gets get turned off.
> > > 
> > > Seems like it is the cleanest way to handle it given the substitution
pattern,
> > but I would be open to any ideas.
> > Maybe in substituter.py it would be simpler just to make FixPath and AbsPath
> > handle empty input instead of having extra conditions in their callers.
> 
> The problem isn't FixPath, the problem is os.path.abspath and os.path.relpath.
Those functions raise an exception on empty values.

After looking a little more, it does seem a little weird that the
toolchain_build scripts are passing the git cache path to SyncGitRepoCmds since
they otherwise know or care nothing about the caching. What about using
CommandOptions, which is already the thing we use to pass options controlled by
toolchain_main down to commands?

[Derek Schuff, 2014-06-26 21:24:43]

lgtm

[David Yen, 2014-06-26 21:52:57]

Committed patchset #6 manually as r13411 (presubmit successful).