Fix the corner case issue with Android test sharder.

Fix the corner case issue with Android test sharder.

Previously all gtests would run if the shard list was empty.

BUG=165529
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=172630

[frankf, 2012-12-12 00:50:05]

+craigdh for him to get familiar with our scripts

[Yaron, 2012-12-12 01:30:10]

So this is just a left over vestige from before sharding? Now with sharding we
later grab the correct number of tests if _gtest_filter isn't set? From a brief
scan of the code it looks right but not fully aware of how the test runner works
anymore.

[frankf, 2012-12-12 01:51:37]

Yep, the code is convoluted, my goal is to simplify it. How it works now:

TestSharder (in run_tests.py) gets all the enabled tests and shards them across
SingleTestRunner objects by setting the gtest_filter param. If there is only 1
test case and 2 devices, then one of the SingleTestRunner objects will have an
empty gtest_filter. I'm not sure why we had that logic to run all the tests,
it's before we upstreamed.

On 2012/12/12 01:30:10, Yaron wrote:
> So this is just a left over vestige from before sharding? Now with sharding we
> later grab the correct number of tests if _gtest_filter isn't set? From a
brief
> scan of the code it looks right but not fully aware of how the test runner
works
> anymore.

[Yaron, 2012-12-12 02:18:33]

lgtm

[bulach, 2012-12-12 11:38:59]

lgtm, thanks frank!

details below:

https://codereview.chromium.org/11557004/diff/1/build/android/pylib/single_te...
File build/android/pylib/single_test_runner.py (left):

https://codereview.chromium.org/11557004/diff/1/build/android/pylib/single_te...
build/android/pylib/single_test_runner.py:332: if not self._gtest_filter:
yeah, I think this was done before we unified the sharder: it was possible to
hit here without a "-f" param, and so we'd run all the non-disabled tests..
now that we always go through sharding, even with one device, we always have the
filter set here.

btw, if you're willing to cleanup: the rebaseline option was done so when we
merged, we'd re-generate the foo_disabled_tests at once...

this is now truly obsolete, there's no scenario in which we'd regenerate that
file.. this "rebaseline" option pollutes and traverses a lot of layers, feel
free to clean up when you have a moment!

[craigdh, 2012-12-12 18:17:26]

lgtm.

[frankf, 2012-12-12 18:54:41]

On 2012/12/12 11:38:59, bulach wrote:
> lgtm, thanks frank!
> 
> details below:
> 
>
https://codereview.chromium.org/11557004/diff/1/build/android/pylib/single_te...
> File build/android/pylib/single_test_runner.py (left):
> 
>
https://codereview.chromium.org/11557004/diff/1/build/android/pylib/single_te...
> build/android/pylib/single_test_runner.py:332: if not self._gtest_filter:
> yeah, I think this was done before we unified the sharder: it was possible to
> hit here without a "-f" param, and so we'd run all the non-disabled tests..
> now that we always go through sharding, even with one device, we always have
the
> filter set here.
> 
> btw, if you're willing to cleanup: the rebaseline option was done so when we
> merged, we'd re-generate the foo_disabled_tests at once...
> 
> this is now truly obsolete, there's no scenario in which we'd regenerate that
> file.. this "rebaseline" option pollutes and traverses a lot of layers, feel
> free to clean up when you have a moment!

@bulach: You read my mind.