https://build.chromium.org/p/chromium.fyi/builders/ClangToTAndroidASan/builds/3054/steps/generate_build_files/logs/stdio

use_goma = true

:-/

7+?!

Also the bot builds with proprietary codecs enabled now, while it didn't previously (and shouldn't).

https://codereview.chromium.org/191824300

https://codereview.chromium.org/1918243002

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/d8deaff6dd714d1446757e7b4038d544289f5230

commit d8deaff6dd714d1446757e7b4038d544289f5230
Author: Nico Weber <thakis@chromium.org>
Date: Tue Apr 26 15:46:15 2016

Try to fix damage from switching ClangToTAndroidASan to mb and gn.

debug_bot_minimal_symbols implicitly enables goma (which doesn't work
with trunk clang), so don't use that.

android implicitly enables proprietary codecs; explicitly opt out from
that.

BUG= 606780 
R=dpranke@chromium.org

Review URL: https://codereview.chromium.org/1918243002 .

Cr-Commit-Position: refs/heads/master@{#389793}

[modify] https://crrev.com/d8deaff6dd714d1446757e7b4038d544289f5230/tools/mb/mb_config.pyl

It's back to just 3h: https://build.chromium.org/p/chromium.fyi/stats/ClangToTAndroidASan

I'll file a separate bug for the 1h->3h regression. That one feels unrelated to gn and mb.

Glad to hear, and sorry about the initial mb misconfiguration.

Actually: I think the "faster as 3h" was also a bug from the mb/gn transition. jbudorick switched the bot before mb knew about the LLVM_FORCE_HEAD_REVISION env var.

...yes, it took 3h before the first mb switch: https://build.chromium.org/p/chromium.fyi/builders/ClangToTAndroidASan/builds/2890


I'm not sure why things were fast after that first mb switch though. It did enable goma, but it did also build a local clang instead of downloading one (the env var was set during runhooks, just not when the build files are generated, but the generated build files should still point at the locally-built clang).

Here's a fast build: https://build.chromium.org/p/chromium.fyi/builders/ClangToTAndroidASan/builds/2987

Ah, because of the missing LLVM_FORCE_HEAD_REVISION, the -DCR_CLANG_REVISION didn't change, and so the bot usually didn't do clobber builds the first time round.

So:
- originally, builds took about 3h, all good
- mb lands for the first time, witout LLVM_FORCE_HEAD_REVISION => bot doesn't do full builds anymore, becomes fast
- mb reverted, back to 3h (good)
- mb relands, this time it forces on goma, and building with -j200 makes things very slow if you build locally (due to mismatching compilers) (7h)
- disable goma => all good again