Split test_support_common into test_support_common[_base]

Split test_support_common into test_support_common[_base]

test_support_common depends on liballocator.a, which contains tcmalloc. It
cannot be used in .so files. I split this library into two parts: one that
doesn't depend on tcmalloc and the original one (which does depend on tcmalloc).
_pyautolib.so links against test_support_common_base which doens't depend on
tcmalloc. This way it can be dlopen()'d without any problems.

BUG=none
TEST=Rebuilt chrome + _pyautolib.so and ran tests to make sure _pyautolib.so
works.
Also ran trybots.

[asharif1, 2012-04-20 20:03:37]

This is an alternative to http://codereview.chromium.org/10035012/.

I refactored out all source code from test_support_common and made another
library called test_support_common_base. I put in just enough dependencies to
run pyautolib but avoided linking in liballocator.a.

My technique here was just trial-and-error. If you know of more efficient ways
of doing this sort of dependency removal, let me know.

[Nirnimesh, 2012-04-20 20:43:13]

http://codereview.chromium.org/10159006/diff/17003/chrome/chrome_tests.gypi
File chrome/chrome_tests.gypi (right):

http://codereview.chromium.org/10159006/diff/17003/chrome/chrome_tests.gypi#n...
chrome/chrome_tests.gypi:40: 'chrome_resources.gyp:chrome_strings',
please keep this list sorted

http://codereview.chromium.org/10159006/diff/17003/chrome/chrome_tests.gypi#n...
chrome/chrome_tests.gypi:69: # copied from skia.
why is skia stuff necessary?

[asharif1, 2012-04-20 22:19:50]

On 2012/04/20 20:43:13, Nirnimesh wrote:
> http://codereview.chromium.org/10159006/diff/17003/chrome/chrome_tests.gypi
> File chrome/chrome_tests.gypi (right):
> 
>
http://codereview.chromium.org/10159006/diff/17003/chrome/chrome_tests.gypi#n...
> chrome/chrome_tests.gypi:40: 'chrome_resources.gyp:chrome_strings',
> please keep this list sorted
> 
>
http://codereview.chromium.org/10159006/diff/17003/chrome/chrome_tests.gypi#n...
> chrome/chrome_tests.gypi:69: # copied from skia.
> why is skia stuff necessary?

This approach is pretty tedious. I haven't been able to get it to work for all
platforms.

I first tried to simply add:

'dependencies!': [
  '../base/allocator/allocator.gyp:allocator'
]

to exclude tcmalloc from pyautolib, that seems like dependencies! can only work
on immediate dependencies.

What needs to happen is all targets need to be separated surgically into
tcmalloc dependent and independent ones (like skia, browser, etc.). pyautolib
needs all those targets and they automatically pull in tcmalloc.

I'm currently trying out a different approach. Adding this to pyautolib:

          'all_dependent_settings': {
            'linux_use_tcmalloc': '0',
          },

Not sure whether this will work or not. If this doesn't work I'll file a Chrome
bug and move on. Perhaps I'll write a different autotest in chromeos (non
pyautolib) that generates the right PGO data.

[asharif1, 2012-04-20 22:20:00]

http://codereview.chromium.org/10159006/diff/17003/chrome/chrome_tests.gypi
File chrome/chrome_tests.gypi (right):

http://codereview.chromium.org/10159006/diff/17003/chrome/chrome_tests.gypi#n...
chrome/chrome_tests.gypi:69: # copied from skia.
On 2012/04/20 20:43:14, Nirnimesh wrote:
> why is skia stuff necessary?

It doesn't build without it since I removed skia from the deps (skia pulls in
tcmalloc).

[asharif1, 2013-04-23 00:07:26]

On 2012/04/20 22:20:00, asharif1 wrote:
> http://codereview.chromium.org/10159006/diff/17003/chrome/chrome_tests.gypi
> File chrome/chrome_tests.gypi (right):
> 
>
http://codereview.chromium.org/10159006/diff/17003/chrome/chrome_tests.gypi#n...
> chrome/chrome_tests.gypi:69: # copied from skia.
> On 2012/04/20 20:43:14, Nirnimesh wrote:
> > why is skia stuff necessary?
> 
> It doesn't build without it since I removed skia from the deps (skia pulls in
> tcmalloc).

I worked around this, so closing this issue for now.