Add 'profiling_full_stack_frames' build option not to optimize out stack frames when profiling.

Add 'profiling_full_stack_frames' build option not to optimize out stack frames when profiling.

It enables the heap-profiler in Release builds.

BUG=252560
R=phajdan.jr@chromium.org

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=208127

[Dai Mikurube (NOT FULLTIME), 2013-06-21 04:34:59]

Hi Paweł,

Could you take a look at this build option?  I thought you're a good reviewer
for this since you were working around profilers and stack traces.  Thanks.

[Dai Mikurube (NOT FULLTIME), 2013-06-21 06:49:52]

Ahh, sorry, it's not working.

[Dai Mikurube (NOT FULLTIME), 2013-06-21 07:30:38]

Sorry again.  I had a wrong command line locally. :(  I think it's working.

[Paweł Hajdan Jr., 2013-06-21 16:18:18]

LGTM

[Dai Mikurube (NOT FULLTIME), 2013-06-22 10:03:47]

Thanks!

[commit-bot: I haz the power, 2013-06-22 10:04:05]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/dmikurube@chromium.org/17115015/1

[commit-bot: I haz the power, 2013-06-22 11:41:06]

Step "update" is always a major failure.
Look at the try server FAQ for more details.
http://build.chromium.org/p/tryserver.chromium/buildstatus?builder=linux_aura...

[commit-bot: I haz the power, 2013-06-23 07:09:11]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/dmikurube@chromium.org/17115015/18001

[commit-bot: I haz the power, 2013-06-23 07:50:49]

Retried try job too often on linux_aura for step(s) content_unittests
http://build.chromium.org/p/tryserver.chromium/buildstatus?builder=linux_aura...

[Dai Mikurube (NOT FULLTIME), 2013-06-23 14:18:06]

Committed patchset #2 manually as r208127 (presubmit successful).

[eseidel, 2013-07-30 19:32:52]

I'm unable to get anything other than adresses when profiling content_shell with
"perf" on linux.

With either Release or Debug.

This seems to be a recent regression and I wonder if it could be related to this
change?

GYP_DEFINES="profiling=1" gclient runhooks
ninja -C ./out/Release content_shell
perf record -g ./out/Release/content_shell -- "javascript:while(true) {}"
--no-sandbox --disable-setuid-sandbox --dump-render-tree --single-process

perf report shows me all the time spent under a non-symboled address.

Similarly with Debug (which I would expect to work regardless of profiling=1).

Maybe you have theories?

[jochen (gone - plz use gerrit), 2013-07-30 19:48:22]

On 2013/07/30 19:32:52, eseidel wrote:
> I'm unable to get anything other than adresses when profiling content_shell
with
> "perf" on linux.
> 
> With either Release or Debug.
> 
> This seems to be a recent regression and I wonder if it could be related to
this
> change?
> 
> GYP_DEFINES="profiling=1" gclient runhooks
> ninja -C ./out/Release content_shell
> perf record -g ./out/Release/content_shell -- "javascript:while(true) {}"
> --no-sandbox --disable-setuid-sandbox --dump-render-tree --single-process
> 
> perf report shows me all the time spent under a non-symboled address.
> 
> Similarly with Debug (which I would expect to work regardless of profiling=1).
> 
> Maybe you have theories?

Are these the precise repro steps, or do you happen by chance to build component
mode? Maybe perf cannot symbolize in component mode with -rdynamic?

[eseidel, 2013-07-30 19:57:07]

Those are precise repro steps.  Or rather, I've repro'd using those steps.

I normally call gyp_chromium directly, and I normally export the variables, but
I have done exactly those steps and repro'd.

ojan is also unable to get stacks on his machine using perf.

Last time I used perf (a few weeks ago) I was able to get stacks just fine for
profiling=1 builds.

It's possible this change had nothing to do with this regression.  Just figured
commenting here might reach the right people.

I would encourage anyone to try:
perf record -g ./out/Debug/content_shell -- "javascript:while(true) {}"
--no-sandbox --disable-setuid-sandbox --dump-render-tree --single-process
^C
perf report

Note: That's on a Debug build, and see if they can get content_shell symbols. 
If they can, then something else may be wrong with their setup.  But so far at
least ojan can confirm my "perf doesn't work with Release or even Debug builds"
findings. :)

Happy to file a bug if that would be helpful.

[eseidel, 2013-07-30 19:58:29]

> Note: That's on a Debug build, and see if they can get content_shell symbols. 
> If they can, then something else may be wrong with their setup.
Sorry I meant *my* setup in that sentence. :(  If perf works for others, then
it's just my problem obviously. :)

[eseidel, 2013-07-30 20:01:22]

BTW, I've repro'd with both clang and gcc.

I've also tried getting stacks with base::debug::StackTrace and failed the other
day, but I may have only tried that with Release.  Now that I think about it,
that's probably related.

Likely something changed with the CC_FLAGS on linux recently?

[eseidel, 2013-07-30 20:05:59]

I've filed https://code.google.com/p/chromium/issues/detail?id=265996 and we can
move the discussion there.

Sorry for hijacking your review.

[Dai Mikurube (NOT FULLTIME), 2013-07-31 03:16:26]

I hope it's not related to the issue. ;)

[eseidel, 2013-07-31 03:55:28]

In the end I found out this was a multi-year old bug.  Although I think Julien
and I may be somewhat close to at least understanding what's going wrong.