GitLab

ARM devices generally claim support for armeabi-v7a, and armeabi, but
Intel ones without Houdini (in particular, the emulators), are just x86.

Bug: https://github.com/android-ndk/ndk/issues/78
Change-Id: I86d7d1810092d846a247db1e34f8fc06bb489ed5

Change-Id: I60b324f032f748a8685429224b87f9591438264d

Also clarify the usage text.

Bug: https://github.com/android-ndk/ndk/issues/63
Change-Id: I05e1186549aa78a52a3a2cdb5bf2adffa81c213f

Chrome will download logs instead of previewing them if they have no
extension :(

Change-Id: I7fc13498cedf56fcf6604c105c959f5013a67d9f

Change-Id: I32e4ca93475abc708b88ed20a7540920a04f5ee7

Not really sure how this got removed. My builds were working locally
because python is an asshole and was picking up the .pyc files.

Change-Id: I520756ede3d50769c662fd77b3505b00024d046f

* changes:
  Print the number of CPUs available.
  Appease pep8.
  Use pbzip2 if available.
  Time the internal steps of packaging.
  Stop compressing our intermediate packages.
  Add timing details to the build.

I'm seeing some strange variations in build time on the build server
and I'm not sure why.

Change-Id: I81a761f635a0f4883f91fdb16850eb302add0592

Change-Id: Iced37e6e18c9eb20fb593e22980d02a1d2445d95

Brings the final package time from 4 minutes to under a minute.

Bug: http://b/28214235
Change-Id: Ia382d70ca37109c9fa36948f07d8d5ec7b6fab0b

Change-Id: I228d58bd6607ec6fe8d6e101b7e3ab7347c77e9f

Change-Id: Id1e3110bb362c786d47a17f0a396d3bd5fa98308

The zips that pass through these two functions are only a part of the
modular NDK vaporware. We're compressing the hell out of them even
though they are never used. Just store them uncompressed to save time
in the build.

Before this change:

    Build: 0:14:16
    Packaging: 0:04:41
    Testing: 0:15:44
    Total: 0:34:43

After this change:

    Build: 0:13:38
    Packaging: 0:04:04
    Testing: 0:15:47
    Total: 0:33:31

Saves about a minute (the build step is affected by this too since
that's the part that creates the zips).

Bug: http://b/28214235
Change-Id: I7027aaa38efc78a0bdc4d2418e0f55f03360010b

When the build is finished, print the build, package, test, and total
time.

Bug: http://b/28214235
Change-Id: Ia0ea0b661a2bd9c11640be1c42375532ceef9f1d

The bad rendering here is actually a gerrit bug, and this makes it
worse.

This reverts commit 92b8744d.

Change-Id: Ieb94242a3862523c5cb23a107be313afb75ffc87

Change-Id: Ia7283ab6dc056d2e3470f8feab37c38af503e851

It's been named this since the dawn of time. Might have made more
sense then, but it's not very helpful now.

Change-Id: I11f0adf131afbd4dc973efa831513bfc2a219059

We don't support any of the multilib configurations (x32, mips32r2,
mips32r6), so don't waste time building them.

    # Before this change:
    $ time ./checkbuild.py --module gnustl
    real    22m24.606s
    user    32m50.704s
    sys     14m5.013s
    $ du -sh ../out/dist/android-ndk-r12-beta1-linux-x86_64.tar.bz2
    660M    ../out/dist/android-ndk-r12-beta1-linux-x86_64.tar.bz2

    # After this change:
    $ time ./checkbuild.py --module gnustl
    real    13m23.073s
    user    19m34.126s
    sys     8m36.052s$ du -sh
    $ du -sh ../out/dist/android-ndk-r12-beta1-linux-x86_64.tar.bz2
    605M    ../out/dist/android-ndk-r12-beta1-linux-x86_64.tar.bz2

Note that some of the time and space we're getting back is because
we're now only building the things we do support once. Previously
building x86_64 also build x86, which had already been built
separately. The same was true for mips64 (though it was an duplicate 3
ABIs instead of one).

Bug: http://b/23437346
Bug: http://b/28214235
Change-Id: I2a552f3ceab63db89b25a6a13ac9caf7ffbf09fb

We run `launch_build` from multiple processes. There can be a race
between the check for the log directory and the creation of it.
`os.makedirs` fails if the directory already exists, so that causes
it to fail.

Move the creation up to the parent process to avoid the issue.

Change-Id: Iac5f120c300db66c440a34b403af458f60bddce4

It seems the build servers run us in our own session, in which case we
get EPERM from `setpgrp`. No need to call this in that case because we
will already be the process group leader.

Change-Id: Ie694a45e48935fb15cded96e44c19d26c46c731b

This shows a roughly 2x speed up of the build step (ignoring tests
and packaging).

Both times run with:

    $ time ./checkbuild.py --arch arm64 --no-test --no-package

The serial result:

    real    16m19.286s
    user    19m44.367s
    sys     6m2.884s

The parallel result:

    real    7m44.093s
    user    26m16.921s
    sys     8m1.321s

The only problem is that SIGINT will spew a lot of garbage to the
terminal. From what I could figure out it looks like this is actually
just something to do with the multiprocessing pools. Might dig more
later, but it seems worth getting this in even with that.

Bug: http://b/28214235
Change-Id: Iff48f31f49255f5482c36fd86b1fc1ac0faa4958

This fixes the issue with building shaderc using Android.mk for Windows.

Change-Id: I796e2971e452f811fd39c84fe990a956ff2a3d5d

Change-Id: Ia6d474aff037ec6fc09f180a3a0a7d6e96c4ac92

* changes:
  Build the device tests during checkbuild too.
  Correct a couple test configs.

Right now we have the build server running the build tests, but we
should check that the device tests are at least buildable even if we
don't have a device to run them on.

This adds support for skipping the execution phase of the tests to
`run_single_configuration`. This is exposed from `run-all.py` as
`--skip-tests`.

Change-Id: Ie129041bc3104ad50b3e96c23158d6918d36b25a