roll clang 155835:157931

roll clang 155835:157931

I manually merged in r158466 on OS X too, so that compiler-rt
could be built. As a consequence, the included compiler-rt no
longer supports iOS. We don't use that anywhere, so that shouldn't
be an issue.

Abandoned because of http://crbug.com/133507

BUG=none
TEST=none

[Nico, 2012-06-16 22:13:36]

Please review. Sending tryjobs from my box seems to not work atm, so I'll use
the cq box to send tryjobs. I'l uncheck it before they are done and won't keep
it checked until you reviewed.

[commit-bot: I haz the power, 2012-06-16 22:13:45]

No LGTM from a valid reviewer yet. Only full committers are accepted.
Even if an LGTM may have been provided, it was from a non-committer or
a lowly provisional committer, _not_ a full super star committer.
See http://www.chromium.org/getting-involved/become-a-committer
Note that this has nothing to do with OWNERS files.

[Nico, 2012-06-16 22:20:43]

Looks like goma can't handle the new binary yet, so this can't land for now.

[hans, 2012-06-18 10:07:42]

> I manually merged in r158466 on OS X

How did that work? You mean for the pre-built goma binaries? Won't that clash
when it compares the checksum with the locally build clang? Ah, or the checksum
wouldn't be affected by compiler-rt so it should work out anyway?

[glotov, 2012-06-18 10:24:11]

Niko, I noticed that gentoo repo began to have stable archives of clang, llvm,
compiler-rt for 3.1. 
Are we interested to use them (instead of building our own clang from source by
you)?

On 2012/06/18 10:07:42, hans wrote:
> > I manually merged in r158466 on OS X
> 
> How did that work? You mean for the pre-built goma binaries? Won't that clash
> when it compares the checksum with the locally build clang? Ah, or the
checksum
> wouldn't be affected by compiler-rt so it should work out anyway?

[glotov, 2012-06-18 10:25:36]

the link:
http://sources.gentoo.org/cgi-bin/viewvc.cgi/gentoo-x86/sys-devel/clang/clang...,
see SRC_URI

[hans, 2012-06-18 10:40:04]

On 2012/06/18 10:24:11, glotov wrote:
> Niko, I noticed that gentoo repo began to have stable archives of clang, llvm,
> compiler-rt for 3.1. 
> Are we interested to use them (instead of building our own clang from source
by
> you)?

Staying close to tip-of-tree allows us to get all the latest and greatest stuff
though, including the ASan stuff that goes into compiler-rt. It also means we're
using roughly the same version as Google uses internally, and we get a chance to
catch issues early.

[Nico, 2012-06-18 15:25:18]

On 2012/06/18 10:07:42, hans wrote:
> > I manually merged in r158466 on OS X
> 
> How did that work? You mean for the pre-built goma binaries? Won't that clash
> when it compares the checksum with the locally build clang? Ah, or the
checksum
> wouldn't be affected by compiler-rt so it should work out anyway?

Right, I just patched this in locally -- see the diff output in the build log
that's in the mac zip file (is.gd/chromeclang).

Locally built binaries have a hash that doesn't match goma's anyway due to the
build system embedding the build timestamp into the binary. Almost nobody builds
clang locally though (just me, as far as I know).

glotov: I don't think we're interested in this. Trunk clang already has several
features that are useful which aren't in 3.1, and this allows us to give
feedback to clang devs much earlier.

[hans, 2012-06-19 16:16:33]

lgtm

[glotov, 2012-06-19 16:23:41]

I tried to do the same roll for ChromeOS
(https://gerrit.chromium.org/gerrit/#change,25519) and started trybots with the
result. amd64-generic-asan works fine, but x86-generic-asan fails to start
autotests, I suspect it just cannot start the image
(http://chromegw.corp.google.com/i/chromiumos.tryserver/builders/x86-generic-a...).
I have not dig deeper yet. Did you face problems with 32-bit image built with
new Clang?

[Nico, 2012-06-19 16:33:35]

On 2012/06/19 16:23:41, glotov wrote:
> I tried to do the same roll for ChromeOS
> (https://gerrit.chromium.org/gerrit/#change%2C25519) and started trybots with
the
> result. amd64-generic-asan works fine, but x86-generic-asan fails to start
> autotests, I suspect it just cannot start the image
>
(http://chromegw.corp.google.com/i/chromiumos.tryserver/builders/x86-generic-a...).
> I have not dig deeper yet. Did you face problems with 32-bit image built with
> new Clang?

I don't think we have a 32-bit CrOs bot for clang (i.e. I haven't tried).

[Nico, 2012-06-19 17:07:26]

Abandoning this roll due to http://crbug.com/133507 . Will try with a new
revision later this week.