clang/win bots: Pass fastbuild=1.

clang/win bots: Pass fastbuild=1.

fastbuild=1 means we don't pass a /Z flag to clang-cl, but we still pass
/DEBUG to link.exe (or lld-link.exe).  So stacks will still have function names,
but no line numbers any more.

All other Windows bots already get this through the "goma" config, but the
clang/win bots don't use goma yet and hence need an explicit flag for this.

BUG=455966
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=293998

[Nico, 2015-02-10 00:11:41]

thakis@chromium.org changed reviewers:
+ rnk@chromium.org

[Nico, 2015-02-10 00:11:41]

timurrrr, ruiu: fyi

[Reid Kleckner, 2015-02-10 00:17:41]

lgtm

Looks like it doesn't affect asan.

[Nico, 2015-02-10 00:21:57]

Thanks!

> Looks like it doesn't affect asan.

You mean it doesn't affect the asan test since function names are still
available in backtraces, right? (Since it does affect the asan builder, see
expectation updates in full_chromium_fyi_CrWinAsan.json and
full_chromium_fyi_CrWinAsan_dll_.json – chromium_win_clang_asan inherits from
the chromium_win_clang config)

[Nico, 2015-02-10 00:21:59]

The CQ bit was checked by thakis@chromium.org

[commit-bot: I haz the power, 2015-02-10 00:22:28]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/915453002/1

[commit-bot: I haz the power, 2015-02-10 00:23:44]

Committed patchset #1 (id:1) as
http://src.chromium.org/viewvc/chrome?view=rev&revision=293998

[Reid Kleckner, 2015-02-10 00:28:42]

On 2015/02/10 00:21:57, Nico wrote:
> Thanks!
> 
> > Looks like it doesn't affect asan.
> 
> You mean it doesn't affect the asan test since function names are still
> available in backtraces, right? (Since it does affect the asan builder, see
> expectation updates in full_chromium_fyi_CrWinAsan.json and
> full_chromium_fyi_CrWinAsan_dll_.json – chromium_win_clang_asan inherits
from
> the chromium_win_clang config)

Hm, I just failed to read this code:

@config_ctx(includes=['chromium_win_clang', 'asan', 'static_library'])
def chromium_win_clang_asan(c):
  # Clear lsan configuration for win.
  del c.gyp_env.GYP_DEFINES['lsan']	 491   del c.gyp_env.GYP_DEFINES['lsan']

It didn't occur to me that the string could reference that function.

I guess changing the *bot* config for asan is OK. It won't change what end users
get. They'll still get line numbers by default.

[Timur Iskhodzhanov, 2015-02-10 06:29:49]

Wait, why do we want to change the bot?

On Tue Feb 10 2015 at 3:28:43 AM <rnk@chromium.org> wrote:

> On 2015/02/10 00:21:57, Nico wrote:
> > Thanks!
>
> > > Looks like it doesn't affect asan.
>
> > You mean it doesn't affect the asan test since function names are still
> > available in backtraces, right? (Since it does affect the asan builder,
> > see
> > expectation updates in full_chromium_fyi_CrWinAsan.json and
> > full_chromium_fyi_CrWinAsan_dll_.json –&nbsp;chromium_win_clang_asan
> > inherits
> from
> > the chromium_win_clang config)
>
> Hm, I just failed to read this code:
>
> @config_ctx(includes=['chromium_win_clang', 'asan', 'static_library'])
> def chromium_win_clang_asan(c):
>    # Clear lsan configuration for win.
>    del c.gyp_env.GYP_DEFINES['lsan']     491   del
> c.gyp_env.GYP_DEFINES['lsan']
>
> It didn't occur to me that the string could reference that function.
>
> I guess changing the *bot* config for asan is OK. It won't change what end
> users
> get. They'll still get line numbers by default.
>
> https://codereview.chromium.org/915453002/
>

To unsubscribe from this group and stop receiving emails from it, send an email
to chromium-reviews+unsubscribe@chromium.org.

[Timur Iskhodzhanov, 2015-02-10 06:29:55]

*the ASan bot

On Tue Feb 10 2015 at 9:29:48 AM Timur Iskhodzhanov <timurrrr@chromium.org>
wrote:

> Wait, why do we want to change the bot?
>
> On Tue Feb 10 2015 at 3:28:43 AM <rnk@chromium.org> wrote:
>
>> On 2015/02/10 00:21:57, Nico wrote:
>> > Thanks!
>>
>> > > Looks like it doesn't affect asan.
>>
>> > You mean it doesn't affect the asan test since function names are still
>> > available in backtraces, right? (Since it does affect the asan builder,
>> > see
>> > expectation updates in full_chromium_fyi_CrWinAsan.json and
>> > full_chromium_fyi_CrWinAsan_dll_.json –&nbsp;chromium_win_clang_asan
>> > inherits
>> from
>> > the chromium_win_clang config)
>>
>> Hm, I just failed to read this code:
>>
>> @config_ctx(includes=['chromium_win_clang', 'asan', 'static_library'])
>> def chromium_win_clang_asan(c):
>>    # Clear lsan configuration for win.
>>    del c.gyp_env.GYP_DEFINES['lsan']     491   del
>> c.gyp_env.GYP_DEFINES['lsan']
>>
>> It didn't occur to me that the string could reference that function.
>>
>> I guess changing the *bot* config for asan is OK. It won't change what end
>> users
>> get. They'll still get line numbers by default.
>>
>> https://codereview.chromium.org/915453002/
>>
>

To unsubscribe from this group and stop receiving emails from it, send an email
to chromium-reviews+unsubscribe@chromium.org.

[Timur Iskhodzhanov, 2015-02-10 13:09:48]

timurrrr@chromium.org changed reviewers:
+ timurrrr@chromium.org

[Timur Iskhodzhanov, 2015-02-10 13:09:48]

I've just checked and the
fasbuild=1 .zip from
http://build.chromium.org/p/chromium.fyi/builders/CrWinAsan/builds/93
doesn't have filename/linenumber information,
whilst the older
http://build.chromium.org/p/chromium.fyi/builders/CrWinAsan/builds/90
has them.
Simple command-line to test:
$ browser_tests.exe --gtest_filter=*CrashRenderer* --no-sandbox

Any reason why we want to have GYP flags on the ASan FYI bot different from what
our end-users have?  Why don't we consider (MFYI) sheriffs / bot spectators as
end users as well?

[Nico, 2015-02-10 15:02:16]

See bug, it cycles faster (1 hour less time spent on "package builds"). The
smaller binaries also enable us to not run into crbug.com/419862 when using
swarming, which means we can now swarm browser tests, which cuts off hours
of test execution time.  All regular bots use fastbuild=1.

If you prefer line numbers on the asan bots, let me know and I can change
it back for them. (But from what I understand, pdb files aren't packaged
right for asan at the moment and the bot doesn't have and pdb info
currently anyways.)

On Tue, Feb 10, 2015 at 5:09 AM, <timurrrr@chromium.org> wrote:

> I've just checked and the
> fasbuild=1 .zip from
> http://build.chromium.org/p/chromium.fyi/builders/CrWinAsan/builds/93
> doesn't have filename/linenumber information,
> whilst the older
> http://build.chromium.org/p/chromium.fyi/builders/CrWinAsan/builds/90
> has them.
> Simple command-line to test:
> $ browser_tests.exe --gtest_filter=*CrashRenderer* --no-sandbox
>
> Any reason why we want to have GYP flags on the ASan FYI bot different
> from what
> our end-users have?  Why don't we consider (MFYI) sheriffs / bot
> spectators as
> end users as well?
>
> https://codereview.chromium.org/915453002/
>

To unsubscribe from this group and stop receiving emails from it, send an email
to chromium-reviews+unsubscribe@chromium.org.

[Timur Iskhodzhanov, 2015-02-10 15:15:32]

I think we should have linenumbers on the ASan bots.

The builder was processing PDBs fine, and I'm currently working on a patch
to ASan RTL that should fix the tester as well.

To unsubscribe from this group and stop receiving emails from it, send an email
to chromium-reviews+unsubscribe@chromium.org.

[Timur Iskhodzhanov, 2015-02-10 16:26:51]

Just FTR, "package build" is only 20 minutes faster on the ASan bots, not 1
hour.

To unsubscribe from this group and stop receiving emails from it, send an email
to chromium-reviews+unsubscribe@chromium.org.

[Nico, 2015-02-10 16:37:47]

The 1h win on CrWinClang looks like an anomaly too:
http://build.chromium.org/p/chromium.fyi/stats/CrWinClang The two runs
before the change were just very slow.

However, unpacking the build on CrWinClang tester got faster, and as I said
this enables of swarming for browser_tests, and that did drop the cycle
time of CrWinClang tester from 2 hours to 20 minutes:
http://build.chromium.org/p/chromium.fyi/stats/CrWinClang%20tester

On Tue, Feb 10, 2015 at 8:26 AM, Timur Iskhodzhanov <timurrrr@chromium.org>
wrote:

> Just FTR, "package build" is only 20 minutes faster on the ASan bots, not
> 1 hour.
>

To unsubscribe from this group and stop receiving emails from it, send an email
to chromium-reviews+unsubscribe@chromium.org.