[PATCH] TargetInfo: Use DataLayout to compute pointer width in getPointerWidth()

[Alp Toker]

On 26/06/2014 00:22, Rafael Espíndola wrote:
> On 25 June 2014 17:19, Alp Toker <alp at nuanti.com> wrote:
>> On 26/06/2014 00:13, Rafael Espíndola wrote:
>>>> Hi Tom,
>>>>
>>>> TargetInfo is meant to be independent of DataLayout IIRC, otherwise
>>>> frontend
>>>> functionality like semantic analysis would become dependent on backend
>>>> targets.
>>> Well, on the target datalayout. That is currently constructed in
>>> clang. My only concern is if this would make it harder to change clang
>>> in the future to use llvm provided datalayouts.
>>
>> Oh, this is a clang DataLayout? If there's no backend dependency it's
>> probably OK.
>>
>> It's interesting no other target has done it this way before, but I guess if
>> it can be tested with a -fsyntax-only test that doesn't have REQUIRES:
>> r600-registered-target there's no fundamental problem
> I agree.

In fact we should probably disable REQUIRES support in Sema* tests. All 
but one of the REQUIRES lines throughout these tests seem to be flawed, 
and that one kinda belongs elsewhere:

$ git grep REQUIRES: -- test/Sema*
test/Sema/array-init.c:// REQUIRES: LP64
test/Sema/compound-literal.c:// REQUIRES: LP64
test/Sema/ms-inline-asm.c:// REQUIRES: x86-registered-target
test/SemaCXX/cstyle-cast.cpp:// REQUIRES: LP64
test/SemaCXX/dcl_ambig_res.cpp:// REQUIRES: LP64
test/SemaCXX/functional-cast.cpp:// REQUIRES: LP64
test/SemaObjC/nonnull.m:// REQUIRES: LP64
test/SemaObjCXX/pointer-to-objc-pointer-conv.mm:// REQUIRES: LP64

Alp.


>
> Cheers,
> Rafael

-- 
http://www.nuanti.com
the browser experts