Another attempt to fix debug info stripping for ASan on OSX

Another attempt to fix debug info stripping for ASan on OSX

Unce upon a time the ASan runtime library used to be statically linked into
the executables on OSX.
Because that library provided a number of API functions needed by the shared
libraries, we had custom ASan-specific .saves files that told the `strip`
utility to preserve those functions when stripping the executables.

Then ASan switched to using dynamic runtime library, which instantly broke the
stripping step, because the executables weren't necessarily referencing all the
ASan API functions (issue 242503).
As a result, stripping has been disabled, and we haven't had .dSYM debug info
for ASan builds for more than a year now (issue 148383).

Because the ASan API functions are now undefined in the executables, it's
actually senseless to use the .saves files to preserve those functions in each
executable. However __asan_default_options still must be preserved to be accessible
from the ASan runtime.

This CL removes all ASan symbols except for __asan_default_options from the .saves files
and makes Chromium.app and Chromium Helper.app correctly use app_asan.saves.
It also turns mac_strip_release on for ASan builds.

BUG=148383, 242503, 170739, 166857
R=mark@chromium.org
TBR=cpu@chromium.org

Committed: https://crrev.com/d051b21c75afc414190db331fd608cc61e3056dc
Cr-Commit-Position: refs/heads/master@{#296413}

[Alexander Potapenko, 2014-09-24 13:43:35]

Mark, may I ask you to take a look?

[Mark Mentovai, 2014-09-24 13:48:30]

LGTM

https://codereview.chromium.org/560323004/diff/80001/build/asan.saves
File build/asan.saves (right):

https://codereview.chromium.org/560323004/diff/80001/build/asan.saves#newcode5
build/asan.saves:5: # This file lists symbols that should not be stripped from
the Mac executables
Rietveld shows this file with a “D” for “delete” but I’m seeing a diff to it. I
think that the diff is right. Just watch out when you commit this that the right
thing happens.

[Alexander Potapenko, 2014-09-24 13:50:10]

On 2014/09/24 13:48:30, Mark Mentovai wrote:
> LGTM
> 
> https://codereview.chromium.org/560323004/diff/80001/build/asan.saves
> File build/asan.saves (right):
> 
> https://codereview.chromium.org/560323004/diff/80001/build/asan.saves#newcode5
> build/asan.saves:5: # This file lists symbols that should not be stripped from
> the Mac executables
> Rietveld shows this file with a “D” for “delete” but I’m seeing a diff to it.
I
> think that the diff is right. Just watch out when you commit this that the
right
> thing happens.

Yeah, saw this too. Don't know what's going on, but I've double-checked this
file is tracked by git.

[Alexander Potapenko, 2014-09-24 13:50:25]

The CQ bit was checked by glider@chromium.org

[commit-bot: I haz the power, 2014-09-24 13:51:23]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/560323004/80001

[commit-bot: I haz the power, 2014-09-24 14:00:06]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2014-09-24 14:00:08]

Try jobs failed on following builders:
  chromium_presubmit on tryserver.chromium.linux
(http://build.chromium.org/p/tryserver.chromium.linux/builders/chromium_presub...)

[Alexander Potapenko, 2014-09-24 14:01:39]

glider@chromium.org changed reviewers:
+ cpu@chromium.org

[Alexander Potapenko, 2014-09-24 14:02:00]

On 2014/09/24 14:00:08, I haz the power (commit-bot) wrote:
> Try jobs failed on following builders:
>   chromium_presubmit on tryserver.chromium.linux
>
(http://build.chromium.org/p/tryserver.chromium.linux/builders/chromium_presub...)

TBRing cpu@, the chrome/app/ owner.

[Alexander Potapenko, 2014-09-24 14:02:05]

The CQ bit was checked by glider@chromium.org

[commit-bot: I haz the power, 2014-09-24 14:02:30]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/560323004/80001

[commit-bot: I haz the power, 2014-09-24 14:49:21]

Committed patchset #5 (id:80001) as e8c3f8add29a74e6280ae0e33f10d4500c95c1b5

[commit-bot: I haz the power, 2014-09-24 14:49:50]

Patchset 5 (id:??) landed as
https://crrev.com/d051b21c75afc414190db331fd608cc61e3056dc
Cr-Commit-Position: refs/heads/master@{#296413}

[jiayl, 2014-09-24 16:24:12]

A revert of this CL (patchset #5 id:80001) has been created in
https://codereview.chromium.org/597873004/ by jiayl@chromium.org.

The reason for reverting is: Broke mac asan content_browsertests:
http://build.chromium.org/p/chromium.memory/builders/Mac%20ASan%20Tests%20%28....

[Alexander Potapenko, 2014-09-24 16:25:09]

Thanks to this CL the file:line info is now available in ASan reports on OSX.
However now the symbol names are gone because the symbol tables llvm-symbolizer
relies onto are stripped (we can't just take symbol info from the debug info
because we're building with -gline-tables-only, which produces cropped symbol
debug info, e.g. "Function()" instead of "namespace::Class::Function()")
I'm inclining towards not stripping the symbols from ASan builds when producing
.dSYM files. Mark, WDYT?

[Alexander Potapenko, 2014-09-24 16:28:25]

jiayl: there's clearly something wrong with my CL since there's no debug info in
the logs (I must've seen it somewhere else)
However I'm not quite sure this was the main reason for the tests to start
failing.