android: Support host multilib builds.

android: Support host multilib builds.

The Android build system now supports multilib 32/64-bit host builds as
well as target builds, and Chromium needs to make use of this to
properly support compiling V8 for host. Introduce a GYP_HOST_VAR_PREFIX
variable that works equivalently to GYP_VAR_PREFIX and use it in all the
places where the first/second host architecture must be selected, and
also introduce GYP_HOST_MULTILIB to enable the top level makefile to
specify whether it wants 32-bit or 64-bit host binaries.

BUG=chromium:358141
r1970

[Torne, 2014-06-02 12:51:37]

Anyone on gyp-developer able to review this? I need this change to unblock the
android build system maintainers enabling host multilib :/

I realise the android backend continues to be behind on testing; we have a
chrome test engineer currently working on updating the test infrastructure so we
can implement test coverage for the new features added to the backend like
multilib, as well as improve existing coverage.

[Torne, 2014-06-30 11:28:42]

Hi Nico,

I've added a (simple) test case for this functionality: it tests that it can
build a host executable which depends on a host static library, and does this
twice, forcing the variables that control 32 vs 64 bit host binaries both ways
and checking the result is as expected.

The separate library is necessary to test the GYP_HOST_VAR_PREFIX code as this
is only relevant when one target depends on another. Without the functional
changes included in this patch, the test fails to build (as expected) when run
against a current version of AOSP where host multilib is enabled by default, but
passes with the changes here.

However, the gyp buildbot currently uses an old version of AOSP which doesn't
support host multilib and so this test won't pass there. I'm sending it to the
trybot to establish exactly in what way it fails.

One of our SETs is working on updating the gyp buildbot script to be able to
check out a newer AOSP build. I'll check up with him on how long that's going to
take and decide what the next step should be here.

[Torne, 2014-06-30 11:46:03]

OK, so it fails "as expected", namely that both attempts build (the same) 32-bit
binary because the multilib variables don't do anything on older versions of the
android build system.

[Torne, 2014-08-08 17:08:29]

If we can land https://codereview.chromium.org/426783004/ then the test here
should work correctly, however that CL currently causes different random tests
to hang on each run and I'm not sure why :/

[Primiano Tucci (use gerrit), 2014-08-22 16:52:34]

FWIW, LGTM
Not a gyp owner nor really familiar with the android generator, but the changes
look consistent to me with the changes in the Android build system.

[Torne, 2014-08-22 16:58:05]

Hooray! The test now passes thanks to Primiano's test harness fixes in
https://codereview.chromium.org/493743002 so it's looking good.

Nico, are you okay with this CL now?

[Nico, 2014-08-25 01:18:05]

lgtm

Thanks!!

[Primiano Tucci (use gerrit), 2014-08-25 17:56:09]

Nico landed this manually in r1970 (Thanks)