Make the Clang and ASan GYP definitions available for iOS.

Make the Clang and ASan GYP definitions available for iOS.
The intention is to allow to build iOS Chrome/tests with asan=1, which is only possible with a custom Clang at the moment.
The resulting binaries can't be shipped to users, but can be used to debug memory corruptions.
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=164619

[Alexander Potapenko, 2012-10-24 16:31:08]

Mike, Stuart, WDYT?
I understand that we can't use the custom Clang in production, but allowing to
run it locally will (probably) improve the development experience: the devs will
get all the newest compile-time checks as well as the power of AddressSanitizer.

[Nico, 2012-10-24 18:12:10]

https://codereview.chromium.org/11273012/diff/1/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/11273012/diff/1/build/common.gypi#newcode3011
build/common.gypi:3011: ['clang==1', {
Now that I think about this, a problem with enabling this for iOS is that this
block often contains flags only understood by a very new clang, so building with
xcode's clang will likely fail if this is enabled and the ios bots build with
clang==1.

[Alexander Potapenko, 2012-10-24 20:23:37]

Do ios bots define clang=1 at all? I don't see any evidence of iOS-specific
clang flags in common.gypi now.

[Alexander Potapenko, 2012-10-25 08:55:10]

On 2012/10/24 20:23:37, Alexander Potapenko wrote:
> Do ios bots define clang=1 at all? I don't see any evidence of iOS-specific
> clang flags in common.gypi now.

If you believe we may occasionally have clang=1 on the iOS bots at some  point,
let's just introduce the ios_use_custom_clang GYP flag so that those who want
the very new Clang will state that explicitly.

[stuartmorgan, 2012-10-25 09:07:35]

On 2012/10/24 20:23:37, Alexander Potapenko wrote:
> Do ios bots define clang=1 at all? I don't see any evidence of iOS-specific
> clang flags in common.gypi now.

I'd be very surprised if this were true. IIRC we found that 'clang=1' really
means 'the_chromium_ToTish_clang=1', and thus breaks things if we set it in
general. So conditionalizing on 'clang' seems correct here.

[Nico, 2012-10-25 20:10:20]

On Thu, Oct 25, 2012 at 2:07 AM,  <stuartmorgan@chromium.org> wrote:
> On 2012/10/24 20:23:37, Alexander Potapenko wrote:
>>
>> Do ios bots define clang=1 at all? I don't see any evidence of
>> iOS-specific
>> clang flags in common.gypi now.
>
>
> I'd be very surprised if this were true. IIRC we found that 'clang=1' really
> means 'the_chromium_ToTish_clang=1', and thus breaks things if we set it in
> general. So conditionalizing on 'clang' seems correct here.

Ok. Another question for Stuart: It sounds like glider wants to use
asan on ios, which requires using chromium's clang to build chrome/ios
(only on the asan bot). Are you fine with having the ios code built
with two slightly different compilers?

If so: glider: Please add a comment saying that the clang block is
only used on ios for asan, not for regular builds.

[Alexander Potapenko, 2012-10-26 08:32:32]

I've proactively added the comment.
Stuart?

[stuartmorgan, 2012-10-26 09:08:28]

LGTM. We're fine with a nonstandard config that uses custom clang (but we may
accidentally break it at any time unless it ends up with a non-experimental bot
eventually).

My one concern is that we might accidentally pick up a clang=1 someday, and not
notice. I wonder if we should put something in the ios section that will explode
if clang is 1 and asan isn't?

But we can revisit that if/when it becomes an issue, I guess.

[stuartmorgan, 2012-10-26 09:11:23]

(I should also warn you that we've had a surprising number of problems with
broken code generation for arm with clang; you should expect that living on ToT
you are likely to hit that more often.)

[Alexander Potapenko, 2012-10-26 09:17:11]

As suggested above, I can add the ios_use_custom_clang flag to prevent that.
In this case the patch actually becomes a one-liner.
OTOH I guess you'll notice the accidental switch to new clang because of the
numerous problems.

[stuartmorgan, 2012-10-26 09:19:39]

On 2012/10/26 09:17:11, Alexander Potapenko wrote:
> OTOH I guess you'll notice the accidental switch to new clang because of the
> numerous problems.

I think that's probably true, which is why I think we should hold off on adding
yet another flag.

[Alexander Potapenko, 2012-10-26 09:25:07]

Ok, I'll wait for Nico's LGTM before submitting.

[Nico, 2012-10-26 22:05:10]

lgtm