linux: set a compiler flag that drops the eh_frame section

linux: set a compiler flag that drops the eh_frame section

eh_frame is data used for exception frames.  We already disable exceptions,
but it turns out this obscure extra flag was responsible for the eh_frame
making it into our binaries.

This shaves off another few megabytes from our binaries.

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=32254

[agl, 2009-11-18 01:01:26]

LGTM

[Joel Stanley (old), 2009-11-18 01:35:08]

LGTM.

Tested with gcc-4.4 and gold. Saved us 6.5MB on my x64 Release build (47665692
-> 40953476).

[Hironori Bono, 2009-11-18 09:13:29]

Evan,

Sorry for my stupid comment in advance.
If I recall correctly, gcc (x64) sometimes uses an eh_frame to display a stack
trace. I suspect this might be a reason why StackTrace.OutputToStream fails on
Linux (x64) since Build 2638 (*1).

(*1)
<http://build.chromium.org/buildbot/waterfall/builders/Chromium%20Linux%20x64/...>

Regards,

Hironori Bono

[Craig, 2009-11-18 12:22:56]

On 2009/11/18 09:13:29, hbono wrote:
> Evan,
> 
> Sorry for my stupid comment in advance.
> If I recall correctly, gcc (x64) sometimes uses an eh_frame to display a stack
> trace. I suspect this might be a reason why StackTrace.OutputToStream fails on
> Linux (x64) since Build 2638 (*1).

No stacktrace is not good :(

There is a nice summary of stack unwinding here btw. that I found useful:

http://sources.redhat.com/ml/frysk/2007-q3/msg00314.html
(search for eh_frame and read from that point on .. not sure how much of that is
still up-to-date and I'm guessing you all know this stuff already).

Maybe we should just use shorter function names? :)

[Evan Martin, 2009-11-18 19:09:29]

On 2009/11/18 09:13:29, hbono wrote:
> Sorry for my stupid comment in advance.

I have never seen a stupid comment from you.  :)

> If I recall correctly, gcc (x64) sometimes uses an eh_frame to display a stack
> trace. I suspect this might be a reason why StackTrace.OutputToStream fails on
> Linux (x64) since Build 2638 (*1).

Yes, that's correct.

[Evan Martin, 2009-11-18 19:09:53]

On 2009/11/18 12:22:56, Craig wrote:
> No stacktrace is not good :(

We already don't have symbols in release, so I'm not too sad about breaking
this.