ARB: Run "gclient sync"

ARB: Run "gclient sync"

BUG=379065

[borenet, 2014-05-30 13:19:03]

Should choose one of this or https://codereview.chromium.org/307103002/

https://codereview.chromium.org/303123007/diff/1/tools/safely-roll-deps.py
File tools/safely-roll-deps.py (right):

https://codereview.chromium.org/303123007/diff/1/tools/safely-roll-deps.py#ne...
tools/safely-roll-deps.py:127: prnt_subprocess.check_call(['gclient', 'sync'])
Should I add --nohooks here?

[M-A Ruel, 2014-05-30 14:01:59]

https://codereview.chromium.org/303123007/diff/1/tools/safely-roll-deps.py
File tools/safely-roll-deps.py (right):

https://codereview.chromium.org/303123007/diff/1/tools/safely-roll-deps.py#ne...
tools/safely-roll-deps.py:127: prnt_subprocess.check_call(['gclient', 'sync'])
On 2014/05/30 13:19:04, borenet wrote:
> Should I add --nohooks here?

If you are going to sync the whole tree, add this "detail" in the docstring.

But is it necessary? For a bot, it simply increases the points of failures and
hooks run arbitrary code, which doesn't excite me at all. For a user, it's more
useful and hooks should be run.

[eseidel, 2014-05-30 17:32:30]

It also seems reasonable that we depend on the calling environment to have
sync'd the tree.  (ARB's wrapper or buildbot for Skia's RB).

However I tried gclient sync manually on the ARB bot and then running
tools/safely-roll-deps.py and it still failed.

This may not be the last time that other python depends on gclient having been
run.  Maybe SRD is simple enough it doesn't need gclient, but eventually it
probably will.

[jochen (gone - plz use gerrit), 2014-06-02 11:21:51]

On 2014/05/30 17:32:30, eseidel wrote:
> It also seems reasonable that we depend on the calling environment to have
> sync'd the tree.  (ARB's wrapper or buildbot for Skia's RB).
> 
> However I tried gclient sync manually on the ARB bot and then running
> tools/safely-roll-deps.py and it still failed.
> 
> This may not be the last time that other python depends on gclient having been
> run.  Maybe SRD is simple enough it doesn't need gclient, but eventually it
> probably will.

I think we need to check_call with shell=True, otherwise, the python script is
executed in the context of the current script which won't work

[borenet, 2014-06-02 14:26:44]

I'm going to close this in favor of https://codereview.chromium.org/307103002/. 
I don't feel safe calling "gclient sync" since it introduces lots of new failure
modes.

[eseidel, 2014-06-02 16:00:50]

It's also possible to pass env= to control the environment.  I don't think we
want use_shell=True, that is just security holes and strange x-platform bugs
waiting to happen.