Make minidump_stackwalk compile with target_arch=ia32 on a linux x64 host.

Make minidump_stackwalk compile with target_arch=ia32 on a linux x64 host.

This is necessary for the perf bots to build minidump_stackwalk which will
allow us to have stack traces when perf tests crash.

BUG=223572
TEST=GYP_DEFINES="branding=Chrome buildtype=Official target_arch=ia32 component=static_library" gclient runhooks && ninja -C out/Release minidump_stackwalk

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=218622

[tonyg, 2013-08-17 00:50:19]

This fixes the error:

/usr/local/google/code/chromium/src/chrome/installer/linux/debian_wheezy_i386-sysroot/usr/include/features.h:323:26:
fatal error: bits/predefs.h: No such file or directory

[Lei Zhang, 2013-08-18 09:41:03]

Are you doing an official build because you need an official build, or just to
get '-g' into cflags? If the latter, just  use linux_dump_symbols=1 instead.

[tonyg, 2013-08-18 17:24:34]

On 2013/08/18 09:41:03, Lei Zhang wrote:
> Are you doing an official build because you need an official build, or just to
> get '-g' into cflags? If the latter, just  use linux_dump_symbols=1 instead.

I want the perf bots to do official builds so that we are testing performance
against the bits we are actually shipping. The perf bots include
minidump_stackwalk in the target they build so that we can display a crash dump
when a bot crashes on a perf test.  Make sense?

[Mark Mentovai, 2013-08-19 14:35:52]

LGTM, I guess, but I’m trying to understand why you need this as much as I guess
Lei is.

Are the perf bots that you want to have do official builds somehow different
from the non-perf bots that are doing official builds?

[tonyg, 2013-08-19 15:48:45]

On 2013/08/19 14:35:52, Mark Mentovai wrote:
> LGTM, I guess, but I’m trying to understand why you need this as much as I
guess
> Lei is.
> 
> Are the perf bots that you want to have do official builds somehow different
> from the non-perf bots that are doing official builds?

The major differences are that they do not clobber every build and they build
the chromium_builder_perf target instead of the all target. These make them
cycle much faster. If we relied on the builders on the chromium.chrome waterfall
to produce the builds for the chromium.perf waterfall, there would be too many
CLs in each build.

Please confirm you are okay with me landing this.

[Mark Mentovai, 2013-08-19 15:49:27]

That doesn’t explain why you need this change. That’s what I’m trying to get at.

[tonyg, 2013-08-19 16:00:30]

On 2013/08/19 15:49:27, Mark Mentovai wrote:
> That doesn’t explain why you need this change. That’s what I’m trying to get
at.

Oh, master.chromium.chrome has two linux official builders: a 32 bit host that
builds a 32 bit target and a 64 bit host that builds a 64 bit target.
master.chromium.perf has one linux builder in which I'm trying to make a 64 bit
host build a 32 bit target.

[Mark Mentovai, 2013-08-19 16:05:13]

OK, LGTM

[commit-bot: I haz the power, 2013-08-19 16:09:23]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/tonyg@chromium.org/23064014/1

[Lei Zhang, 2013-08-19 20:24:23]

not lgtm

This only works because you just happen to have a machine with both the 32-bit
and 64-bit chroots. On a machine that only does a 32-bit official build, this
won't work.

Also, you shouldn't need the chroot unless you are doing an official+chrome
build. "Normal" builds shouldn't need the chroot.

[tonyg, 2013-08-19 21:14:09]

> On a machine that only does a 32-bit official build, this won't work.

But this is guarded by:

'conditions': [
  ['host_arch=="x86_64"', {
    ...

> Also, you shouldn't need the chroot unless you are doing an official+chrome
> build. "Normal" builds shouldn't need the chroot.

My basic goal is just to get the build working on a 64bit host cross-compiling
to 32bit. I don't really even care whether the minidump_stackwalk binary ends up
respecting the 32bit flag or just builds in host mode.

Could you suggest to me the right way to go about accomplishing this?

[Lei Zhang, 2013-08-19 21:31:10]

On 2013/08/19 21:14:09, tonyg wrote:
> > On a machine that only does a 32-bit official build, this won't work.
> 
> But this is guarded by:
> 
> 'conditions': [
>   ['host_arch=="x86_64"', {
>     ...

To clarify, on a 64-bit host doing a 32-bit build, this won't work because you
don't have the 64-bit chroot.

> > Also, you shouldn't need the chroot unless you are doing an official+chrome
> > build. "Normal" builds shouldn't need the chroot.
> 
> My basic goal is just to get the build working on a 64bit host cross-compiling
> to 32bit. I don't really even care whether the minidump_stackwalk binary ends
up
> respecting the 32bit flag or just builds in host mode.
> 
> Could you suggest to me the right way to go about accomplishing this?

Sure, let me test out some ideas and get back to you later today.

[Lei Zhang, 2013-08-20 08:32:36]

So how about have a condition inside the host_arch=="x86_64" condition like
this?

['target_arch=="ia32" and sysroot!=""', {
  'cflags!': ['--sysroot=<(sysroot)'],
}],

[tonyg, 2013-08-20 16:44:43]

On 2013/08/20 08:32:36, Lei Zhang wrote:
> So how about have a condition inside the host_arch=="x86_64" condition like
> this?
> 
> ['target_arch=="ia32" and sysroot!=""', {
>   'cflags!': ['--sysroot=<(sysroot)'],
> }],

That worked like a charm. Thanks for setting me straight. PTAL

[Lei Zhang, 2013-08-20 18:52:08]

lgtm

[commit-bot: I haz the power, 2013-08-20 18:53:50]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/tonyg@chromium.org/23064014/45001

[commit-bot: I haz the power, 2013-08-21 03:36:16]

Change committed as 218622