Build on Linux with shared libraries (significant chunks courtesy craigsch):...

Build on Linux with shared libraries (significant chunks courtesy craigsch):
* Set $RPATH  to $LIB_DIR in the SCons configuration.
* Add missing dependencies:
  * net/net.gyp:net
    => testing/gtest.gyp:gtest
  * third_party/libxml/libxml.gyp:xmlcatalog
    => third_party/icu38/icu38.gyp:icuuc
  * chrome/chrome.gyp:perf_tests
    => renderer
    => views
    => webkit/webkit.gyp:glue
* Add files:
  * third_party/WebKit/WebCore/loader/icon/IconRecord.cpp
  * third_party/WebKit/WebCore/page/Coordinates.cpp
  * skia/sgl/SkUnPreMultiply.cpp
* Exclude on Linux:
  * chrome/views/controls/scroll_view.cc
  * chrome/views/focus/external_focus_tracker.cc
  * media/filter/ffmpeg_demuxer.{cc,h}
* Remove files:
  * third_party/WebKit/WebCore/Configurations/Version.xcconfig
* Sort the chrome.gyp:views linux exclusion list.
* DEPS roll for $SHLINKFLAGS settings in gyp.

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=14166

[Mark Mentovai, 2009-04-21 21:59:08]

LG otherwise

http://codereview.chromium.org/88058/diff/1008/11
File net/net.gyp (right):

http://codereview.chromium.org/88058/diff/1008/11#newcode21
Line 21: '../testing/gtest.gyp:gtest',
This one seems weird.  Care to explain?

[Evan Martin, 2009-04-21 22:00:05]

LG!

[Evan Martin, 2009-04-21 22:13:37]

http://codereview.chromium.org/88058/diff/1008/11
File net/net.gyp (right):

http://codereview.chromium.org/88058/diff/1008/11#newcode21
Line 21: '../testing/gtest.gyp:gtest',
On 2009/04/21 21:59:08, Mark Mentovai wrote:
> This one seems weird.  Care to explain?

I ran into this one before.  The net library has a "test_utils" module of some
sort, which needs gtest.  But for some reason you also can't split out that
single module into net_unittests... I believe because maybe users of net also
use that test helper module?

[Mark Mentovai, 2009-04-21 22:18:18]

Sounds like a good candidate for a comment.  "We know this is weird, here's why
we're doing it, and it's a good candidate for refactoring."

[sgk, 2009-04-21 22:48:19]

http://codereview.chromium.org/88058/diff/1008/11
File net/net.gyp (right):

http://codereview.chromium.org/88058/diff/1008/11#newcode21
Line 21: '../testing/gtest.gyp:gtest',
On 2009/04/21 22:13:37, Evan Martin wrote:
> On 2009/04/21 21:59:08, Mark Mentovai wrote:
> > This one seems weird.  Care to explain?
> 
> I ran into this one before.  The net library has a "test_utils" module of some
> sort, which needs gtest.  But for some reason you also can't split out that
> single module into net_unittests... I believe because maybe users of net also
> use that test helper module?

Went back and investigated.  Without this, disk_cache and stress_cache died with
(e.g.):

Linking
/usr/local/google/home/sgk/chrome/trunk1/src/sconsbuild/Debug/crash_cache
/usr/local/google/home/sgk/chrome/trunk1/src/sconsbuild/Debug/lib/libnet.so:
undefined reference to
`testing::internal::AssertHelper::operator=(testing::Message const&) const'
/usr/local/google/home/sgk/chrome/trunk1/src/sconsbuild/Debug/lib/libnet.so:
undefined reference to
`testing::internal::AssertHelper::AssertHelper(testing::TestPartResultType, char
const*, int, char const*)'
collect2: ld returned 1 exit status

But it makes more sense to add the gtest dependency to those executables, not to
libnet.so itself.  Uploaded new CL.

[Evan Martin, 2009-04-21 22:55:54]

On 2009/04/21 22:48:19, sgk wrote:
> But it makes more sense to add the gtest dependency to those executables, not
to
> libnet.so itself.  Uploaded new CL.

But those executables are not test programs!  The dependency lies in libnet,
which makes use of gtest directly.  Any program that uses libnet will need gtest
linked in, so I think the old way was correct.

[sgk, 2009-04-21 23:17:39]

http://codereview.chromium.org/88058/diff/1008/11
File net/net.gyp (right):

http://codereview.chromium.org/88058/diff/1008/11#newcode21
Line 21: '../testing/gtest.gyp:gtest',
On 2009/04/21 22:48:20, sgk wrote:
> On 2009/04/21 22:13:37, Evan Martin wrote:
> > On 2009/04/21 21:59:08, Mark Mentovai wrote:
> > > This one seems weird.  Care to explain?
> > 
> > I ran into this one before.  The net library has a "test_utils" module of
some
> > sort, which needs gtest.  But for some reason you also can't split out that
> > single module into net_unittests... I believe because maybe users of net
also
> > use that test helper module?
> 
> Went back and investigated.  Without this, disk_cache and stress_cache died
with
> (e.g.):
> 
> Linking
> /usr/local/google/home/sgk/chrome/trunk1/src/sconsbuild/Debug/crash_cache
> /usr/local/google/home/sgk/chrome/trunk1/src/sconsbuild/Debug/lib/libnet.so:
> undefined reference to
> `testing::internal::AssertHelper::operator=(testing::Message const&) const'
> /usr/local/google/home/sgk/chrome/trunk1/src/sconsbuild/Debug/lib/libnet.so:
> undefined reference to
> `testing::internal::AssertHelper::AssertHelper(testing::TestPartResultType,
char
> const*, int, char const*)'
> collect2: ld returned 1 exit status
> 
> But it makes more sense to add the gtest dependency to those executables, not
to
> libnet.so itself.  Uploaded new CL.

A full, clean-tree rebuild with shared libs shows that I'm incorrect above. 
Even linking 'chrome' itself with shared libs fails with the above undefined
reference(s).  Adding the gtest.gyp:gtest dependency back to libnet.so itself,
with a comment per mmentovai's suggestion.

        --SK