Change ninja rpath generation to be more like make.

Change ninja rpath generation to be more like make.

Currently, the ninja generator adds -Wl,rpath in every executable for
linux targets, while the make generator includes it only for those
with .so dependencies. This patch changes the ninja generator to
behave like make.

Added a linux-specific unit test to verify the value and absence of
the rpath for executables with and without solib dependencies.

BUG=gyp:315
TEST=On linux: unit tests pass (45 no result), full
     Chrome build no longer has rpath on chrome_sandbox.

[cwolfe, 2013-01-10 20:44:05]

thakis: please review
davidjames: fyi

Not wedded to this approach, but more consistency seemed like the easy fix. Only
other difference I've hit so far is using 'lib' vs 'lib.target'.

[cwolfe, 2013-01-10 21:16:38]

Added unit test, and fixed my environment so that the unit tests actually pass.

Changed the rpath added by ninja to include a trailing slash. Don't think it has
any semantic meaning, but consistent with make.

[Nico, 2013-01-10 21:34:57]

lgtm!

(with comments addressed)

https://codereview.chromium.org/11821060/diff/3001/test/linux/gyptest-implici...
File test/linux/gyptest-implicit-rpath.py (right):

https://codereview.chromium.org/11821060/diff/3001/test/linux/gyptest-implici...
test/linux/gyptest-implicit-rpath.py:39: if (GetRpaths('shared_executable') !=
[expect]):
ifs usually have no parens in python

https://codereview.chromium.org/11821060/diff/3001/test/linux/gyptest-implici...
test/linux/gyptest-implicit-rpath.py:42: if (GetRpaths('static_executable') !=
[]):
likewise, and since lists are false exactly if they're not empty, this is
usually just written:

  if GetRpaths(...):
    ...

https://codereview.chromium.org/11821060/diff/3001/test/linux/implicit-rpath/...
File test/linux/implicit-rpath/test.gyp (right):

https://codereview.chromium.org/11821060/diff/3001/test/linux/implicit-rpath/...
test/linux/implicit-rpath/test.gyp:1: # Copyright (c) 2012 Google Inc. All
rights reserved.
nit: 2013

https://codereview.chromium.org/11821060/diff/3001/test/linux/implicit-rpath/...
test/linux/implicit-rpath/test.gyp:21: 'shared'
nit: i'd put a trailing , here

https://codereview.chromium.org/11821060/diff/3001/test/linux/implicit-rpath/...
test/linux/implicit-rpath/test.gyp:29: 'static'
here too

[Nico, 2013-01-10 21:35:26]

And try running `gcl try youchange` to send tryjobs. (I don't remember if that
works for non-committers – I think it does?)

[cwolfe, 2013-01-10 21:49:47]

Thanks!

'gcl try' does work, will wait for those to pass. Initially looked at the "Try
more bots" button in Reitveld, which wasn't promising.

https://codereview.chromium.org/11821060/diff/3001/test/linux/gyptest-implici...
File test/linux/gyptest-implicit-rpath.py (right):

https://codereview.chromium.org/11821060/diff/3001/test/linux/gyptest-implici...
test/linux/gyptest-implicit-rpath.py:39: if (GetRpaths('shared_executable') !=
[expect]):
On 2013/01/10 21:34:57, Nico wrote:
> ifs usually have no parens in python

Done. I blame test/mac/gyptest-rpath.py :)

https://codereview.chromium.org/11821060/diff/3001/test/linux/gyptest-implici...
test/linux/gyptest-implicit-rpath.py:42: if (GetRpaths('static_executable') !=
[]):
On 2013/01/10 21:34:57, Nico wrote:
> likewise, and since lists are false exactly if they're not empty, this is
> usually just written:
> 
>   if GetRpaths(...):
>     ...

Done.

https://codereview.chromium.org/11821060/diff/3001/test/linux/implicit-rpath/...
File test/linux/implicit-rpath/test.gyp (right):

https://codereview.chromium.org/11821060/diff/3001/test/linux/implicit-rpath/...
test/linux/implicit-rpath/test.gyp:1: # Copyright (c) 2012 Google Inc. All
rights reserved.
On 2013/01/10 21:34:57, Nico wrote:
> nit: 2013

Done.

https://codereview.chromium.org/11821060/diff/3001/test/linux/implicit-rpath/...
test/linux/implicit-rpath/test.gyp:21: 'shared'
On 2013/01/10 21:34:57, Nico wrote:
> nit: i'd put a trailing , here

Done.

https://codereview.chromium.org/11821060/diff/3001/test/linux/implicit-rpath/...
test/linux/implicit-rpath/test.gyp:29: 'static'
On 2013/01/10 21:34:57, Nico wrote:
> here too

Done.

[cwolfe, 2013-01-10 22:36:23]

gyp-mac failed once (825). Passed on the second try (826), so assuming it's
flake. Error was:

/usr/bin/python test/product/gyptest-product.py
None returned -11

Will commit tomorrow morning when I'm around to answer for breakage.

[Nico, 2013-01-10 23:03:05]

On 2013/01/10 22:36:23, cwolfe wrote:
> gyp-mac failed once (825). Passed on the second try (826), so assuming it's
> flake. Error was:
> 
> /usr/bin/python test/product/gyptest-product.py
> None returned -11

Can you check if go/gypbot is green (I'm not on vpn)? Your patch doesn't touch
any mac bits, so I agree that it's not your patch.

> 
> Will commit tomorrow morning when I'm around to answer for breakage.

Sounds good!

[cwolfe, 2013-01-11 20:24:25]

chrome-troopers got gyp-linux fixed. May be something else broken, as the Commit
checkbox doesn't seem to be triggering anything.

[M-A Ruel, 2013-01-15 16:26:25]

On 2013/01/11 20:24:25, cwolfe wrote:
> chrome-troopers got gyp-linux fixed. May be something else broken, as the
Commit
> checkbox doesn't seem to be triggering anything.

You have to commit manually. There's no CQ for gyp.

[cwolfe, 2013-01-15 16:41:34]

On 2013/01/15 16:26:25, Marc-Antoine Ruel wrote:
> On 2013/01/11 20:24:25, cwolfe wrote:
> > chrome-troopers got gyp-linux fixed. May be something else broken, as the
> Commit
> > checkbox doesn't seem to be triggering anything.
> 
> You have to commit manually. There's no CQ for gyp.

Thanks -- Nico already committed it for me (e-mails on the list).