Don't use clang by default when cross building for ARM/linux.

Don't use clang by default when cross building for ARM/linux.

Currently our cross-builders for ARM use the ubuntu gcc
cross compiler (on Precise this is gcc 4.6).

This is a temporary fix since 4.6 support is on the way out.
The longer term fix will be to switch to clang for cross
builds.

BUG=395832
TEST=arm_linux_cross_compile bot

[Sam Clegg, 2014-07-22 22:18:43]

The current bot failure is due to a different issue:
https://codereview.chromium.org/401243002/

[Mark Seaborn, 2014-07-22 22:23:31]

[+bradnelson and nodir]

https://codereview.chromium.org/411633003/diff/20001/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/411633003/diff/20001/build/common.gypi#newcod...
build/common.gypi:1783: ['OS=="linux" and target_arch!="mipsel" and
target_arch!="arm"', {
Does this affect Chrome OS, or do they set the compiler in some way that
bypasses this (e.g. by setting clang=0)?

https://codereview.chromium.org/411633003/diff/20001/build/common.gypi#newcod...
build/common.gypi:4271: ['gcc_version>=46 and gcc_version<48 and clang==0', {
Can you add some comments here?  I can't really tell what this is for.

[Nico, 2014-07-22 22:24:29]

This won't help. Trunk doesn't compile with 4.6.

[Nico, 2014-07-22 22:25:01]

https://codereview.chromium.org/411633003/diff/20001/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/411633003/diff/20001/build/common.gypi#newcod...
build/common.gypi:1783: ['OS=="linux" and target_arch!="mipsel" and
target_arch!="arm"', {
On 2014/07/22 22:23:31, Mark Seaborn wrote:
> Does this affect Chrome OS, or do they set the compiler in some way that
> bypasses this (e.g. by setting clang=0)?

I made the chromeos chrome.ebuild explicitly set clang=0

[Nico, 2014-07-22 22:26:26]

https://codereview.chromium.org/411633003/diff/40001/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/411633003/diff/40001/build/common.gypi#newcod...
build/common.gypi:4277: '-std=gnu++0x',
Oh, I missed this. I really don't want to have a bot that uses 4.6 and
std=gnu++0x. There are many known compiler bugs with c++11 and 4.6, which is why
we're not supporting this configuration in the first place.

The way forward for this bot is either clang or a qemu with gcc 4.8 and whatnot
as suggested by jam.

[Sam Clegg, 2014-07-23 00:35:50]

https://codereview.chromium.org/411633003/diff/40001/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/411633003/diff/40001/build/common.gypi#newcod...
build/common.gypi:4277: '-std=gnu++0x',
On 2014/07/22 22:26:26, Nico (away) wrote:
> Oh, I missed this. I really don't want to have a bot that uses 4.6 and
> std=gnu++0x. There are many known compiler bugs with c++11 and 4.6, which is
why
> we're not supporting this configuration in the first place.
> 
> The way forward for this bot is either clang or a qemu with gcc 4.8 and
whatnot
> as suggested by jam.


OK.  I'll keep pushing forward with clang.

Should we error out if gccversion>0 && gccversion<48 than?

[Nico, 2014-07-23 00:40:10]

https://codereview.chromium.org/411633003/diff/40001/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/411633003/diff/40001/build/common.gypi#newcod...
build/common.gypi:4277: '-std=gnu++0x',
On 2014/07/23 00:35:50, Sam Clegg wrote:
> On 2014/07/22 22:26:26, Nico (away) wrote:
> > Oh, I missed this. I really don't want to have a bot that uses 4.6 and
> > std=gnu++0x. There are many known compiler bugs with c++11 and 4.6, which is
> why
> > we're not supporting this configuration in the first place.
> > 
> > The way forward for this bot is either clang or a qemu with gcc 4.8 and
> whatnot
> > as suggested by jam.
> 
> 
> OK.  I'll keep pushing forward with clang.
> 
> Should we error out if gccversion>0 && gccversion<48 than?

Yes. That's blocked on someone (likely me :-) ) to figure out the "a better fix"
thing described at https://codereview.chromium.org/379693003/ , as under certain
conditions gcc_version is not 0 even with clang at the moment.