| 35346dfd26cdceb35a56deb961e0db015ab43485 |
|
04-Dec-2015 |
Dan Albert <danalbert@google.com> |
Gingerbread support for the test runner.
Wow. Gingerbread didn't support `mkdir -p` or `rm -f`. Avoid using
either of those.
Change-Id: I9f79613465e8608f89bad9a77c232620a3ca8b4e
|
| 877b0ba5f2e0f6c6c729bf388910691264406bff |
|
04-Dec-2015 |
Dan Albert <danalbert@google.com> |
Fix doc, missed a --amend earlier.
Change-Id: I63cb52c17bef1ae83e75df896e1743239758622d
|
| 6abf81a055fa510794430457c77432ccf12475a4 |
|
03-Dec-2015 |
Dan Albert <danalbert@google.com> |
Pass the device's API level to device test_config.
We need to mark some tests as broken or unsupported based on what
device they are running on, as opposed to just what they were built
for.
Change-Id: I4818aa23d5af6f47f02d3bee5a22d299088e5562
|
| 4a1931270e6496f4b37551adaedc23c26f787bf5 |
|
24-Nov-2015 |
Dan Albert <danalbert@google.com> |
Replace cmp and diff with python equivalents.
Windows doesn't have these, and it's much easier to just use the
Python flavors rather than adding compatibility to shell out to the
right tools.
Change-Id: Icea144a61f0329314a6d5f7e2a92c9337bf2dba7
|
| 522c00403c6859aaf881b079456533e5b9bcce0a |
|
25-Nov-2015 |
Dan Albert <danalbert@google.com> |
Give test binaries execute permission.
Change-Id: I8874ebb201445a0d9df7ec2a3e813ba452370446
|
| ba50f01138ea9690a98cec4bb4abe6d88fdc7e85 |
|
25-Nov-2015 |
Dan Albert <danalbert@google.com> |
Use posixpath to build device path.
Change-Id: Iea106cff2a62bed35fb22a7d39eca1d64c2a8892
|
| b2c41e443cd1868c65a411dd1158accac63825c9 |
|
24-Nov-2015 |
Dan Albert <danalbert@google.com> |
Actually use the prebuilt awk...
I tried to do this before, but apparently failed miserably.
Change-Id: I4132e39ed7d1a3fa58a478651046b678bb4859d3
|
| ebc5fb43dae99b990eb96639fb1184b1deec3537 |
|
24-Nov-2015 |
Dan Albert <danalbert@google.com> |
Skip build.sh tests on Windows.
Bug: http://b/25293074
Change-Id: I3619462dc7bef5b315006cae84ae136be74631d9
|
| 9c7238cef1898ef61be6b40b9ca6654f104a082f |
|
27-Oct-2015 |
Dan Albert <danalbert@google.com> |
Add a new build test type: test.py.
We want to be able to run the build tests on Windows without requiring
cygwin (or similar). Since we already have Python available, we can
just rewrite the build.sh tests in that. This new test type makes that
possible.
Bug: http://b/25293074
Change-Id: I4649a79c51a44708dfefaeb2aa05e303031e80ca
|
| 1e53031916ac893fcf953321cdaa32fb91d84012 |
|
07-Oct-2015 |
Dan Albert <danalbert@google.com> |
Always capture test output for the report.
Before, the build process was shown on the terminal and the final
report only said "build failed". This patch makes the execution more
succinct and displays all the messages at the end.
Change-Id: I57228fe2006652870b25a32789e11e337b046732
|
| c8aa9d02ad6551b2fb9ae27314c54872da623914 |
|
05-Oct-2015 |
Dan Albert <danalbert@google.com> |
Don't count filtered tests as "skipped".
These are just clutter in the output. Skipped should be reserved for
things that were marked unsupported.
Change-Id: I1f533b351fc5c8d12cd28e0e5da00c3df822559d
|
| b8f95f59a07eb2d0f65ac5840a559087d835c305 |
|
05-Oct-2015 |
Dan Albert <danalbert@google.com> |
Don't fixup UnexpectedSuccess or ExpectedFailure.
These have already been fixed up on a per-subtest basis in DeviceTest
(we don't have any knowledge of what subtests exist without building
them) so doing it again here will actually swap the two of them
(ExpectedFailure becomes an UnexpectedSuccess and vice versa).
Change-Id: I75e0684e3b8ffe0323aa2c6d3ca1fd225f08cdf8
|
| 4b485bf4fbd62ce2be8f45958cd223230a5be538 |
|
24-Sep-2015 |
Dan Albert <danalbert@google.com> |
Add test_config.py for better test disabling.
Each test directory can contain a "test_config.py" file that describes
the configurations a test is not expected to pass for. Previously this
information could be captured in one of two places: the Application.mk
file, or a BROKEN_BUILD/BROKEN_RUN file.
Application.mk was used to state that a test was only to be run for a
specific platform version, specific toolchain, or a set of ABIs.
Unfortunately Application.mk could only specify a single toolchain or
platform, not a set.
BROKEN_BUILD/BROKEN_RUN files were too general. An empty file meant
the test should always be skipped regardless of configuration. Any
change that would put a test in that situation should be reverted
immediately. These also didn't make it clear if the test was actually
broken (and thus should be fixed) or just not applicable.
A test_config.py file is more flexible. It is a Python module that
defines the functions `match_broken` and `match_unsupported`. These
functions respectively identify if a test is broken or unsupported for
the given configuration.
Change-Id: Ibffcef610a32c49efef17eba1154f6a9f4dcf565
|
| a13bb8ae071e59e7881b6bf6f8386bfb0595d1b0 |
|
25-Sep-2015 |
Dan Albert <danalbert@google.com> |
Increase logging for device tests.
Many copies or many tests can make it look like the test is locked up.
Log each copy and execution so we can see what's going on.
Change-Id: I50e1f0d8f51dccd8fef8254a9ef09e3f8506e1de
|
| eb959f69781dffd602cfc551e612a3a8cf7f55d5 |
|
25-Sep-2015 |
Dan Albert <danalbert@google.com> |
Filter tests copied to the device.
Some tests like stlport's contain a lot of executables. If we're only
running a single test, there's no sense spending 30 seconds copying
all of them.
Change-Id: Ie78e9103934cd62a8f48f1ac144d16ec9d892b05
|
| 14395e23186db4a65b4943a6307d223bc3ac4232 |
|
24-Sep-2015 |
Dan Albert <danalbert@google.com> |
Make toolchain a command line test argument.
This lets us avoid calling in to ndk-build to determine which
toolchain we're using (which we need to know to figure out if we
should skip the test or not).
Change-Id: I8da595863e8a760f50367091d70e1cd4f5254c5b
|
| 5399d01ab35b4b1192891723c9e9a86f5783d13e |
|
27-Aug-2015 |
Dan Albert <danalbert@google.com> |
Use the prebuilt awk for the awk tests.
The old test runner didn't do this, but this was we should (in theory)
be able to run the tests on Windows too.
Change-Id: I6c11439e123af615ee158972bd98725ce682bf04
|
| 44873b38075d524c8f32754369085167852af711 |
|
06-Aug-2015 |
Dan Albert <danalbert@google.com> |
Move test details out of the driver script.
Only actual code change here is moving DEV_NULL from the global scope
down to the only function it's used in. Other than that it's just a
move.
Change-Id: I73d07aab4333bb36aa9a4f1029f36fa74029e9f7
|