[PATCH] clang-check should ignore -no-integrated-as because certain drivers can't handle it

[Artyom Skrobov]

Thank you Simon,

Do you agree that ignoring -no-integrated-as (and therefore, only creating the external assembler job if that's the default behaviour) is the right solution for clang-check?

It seems to me to be equally applicable to all three possible cases: external-by-default, internal-by-default, and internal-only.


-----Original Message-----
From: Simon Atanasyan [mailto:simon at atanasyan.com] 
Sent: 11 December 2013 13:44
To: Artyom Skrobov
Cc: cfe-commits at cs.uiuc.edu; Scott Douglass; revane at gmail.com; geek4civic at gmail.com
Subject: Re: [PATCH] clang-check should ignore -no-integrated-as because certain drivers can't handle it

Some targets (for example MIPS and maybe ARM) do dot use integrated
assembler by default. If we do not pass --integrated-as argument to
the driver it always invokes external assembler if the driver run on
MIPS. The purpose of the commit r156478 was to solve this problem.

It looks like your change solves an opposite problem - fix the test
for targets which always use integrated assembler.

On Tue, Dec 10, 2013 at 7:20 PM, Artyom Skrobov <Artyom.Skrobov at arm.com> wrote:
> Simon, thank you for this explanation.
>
> If I understand it correctly, our proposed change would prevent the second job from being created -- defeating the purpose of this test?
>
> OTOH, how would you expect this test to be handled by the drivers that are incapable of invoking external assembler?
> Becoming somewhat pointless with the drivers that cannot create the second job, does it need a permanent REQUIRES: statement, in place of the XFAIL?
>
>
> -----Original Message-----
> From: Simon Atanasyan [mailto:simon at atanasyan.com]
> Sent: 10 December 2013 11:28
> To: Artyom Skrobov
> Cc: cfe-commits at cs.uiuc.edu; Scott Douglass
> Subject: Re: [PATCH] clang-check should ignore -no-integrated-as because certain drivers can't handle it
>
> On Tue, Dec 10, 2013 at 2:59 PM, Artyom Skrobov <Artyom.Skrobov at arm.com> wrote:
>>> Do we know why Clang::Tooling/multi-jobs.cpp has '--no-integrated-as'
>>> in the first place?
>>
>> No, but it's been there since the beginning (r156478).
>> I'm copying Simon Atanasyan and Manuel Klimek who created/reviewed the
>> original test, and may be able to shed some light on the purpose of
>> --no-integrated-as there.
>
> This test checks ClangSyntaxOnlyAdjuster. If we pass -no-integrated-as
> option to the clang, it starts two jobs. The first one produces
> assembler file. The second one runs external assembler to generate an
> object file. The goal of the ClangSyntaxOnlyAdjuster is to add
> -fsyntax-only option to the command line. With that option clang
> changes behaviour and does not start external assembler.

-- 
Simon