[PATCH] clang-cl: Add support for -o

Alexey Samsonov samsonov at google.com
Mon May 19 15:06:24 PDT 2014


On Mon, May 19, 2014 at 12:03 PM, Reid Kleckner <rnk at google.com> wrote:

> On Mon, May 19, 2014 at 5:10 AM, Timur Iskhodzhanov <timurrrr at google.com>wrote:
>
>> 2014-05-15 11:00 GMT+04:00 Reid Kleckner <rnk at google.com>:
>> > Maybe we should just run these tests with the gcc driver frontend.
>>
>> So the problem with that is a combination of the following factors:
>> a) The ASan lit tests use RUN lines like "%clangxx_asan %s -o %t"
>>   [e.g. see test/asan/TestCases/stack-buffer-overflow.cc]
>>
>> b) The clang-cl driver doesn't support "-o"
>>
>> c) By default, %clangxx_asan is COMPILER_RT_TEST_COMPILER
>>
>> d) We have to use clang-cl to build ASan gtest-based tests as gtest
>> itself uses exceptions.
>>   Currently clang doesn't support exception, thus clang-cl falls back.
>>
>
> Try using -DGTEST_HAS_SEH=0, this should compile out any use of __try.  I
> use that to self-host.
>
> It's my fault for not documenting this more widely, but it's not really
> clean yet.
>
>
>> e) COMPILER_RT_TEST_COMPILER is apparently used to build gtest-based tests
>>   [see cmake/Modules/CompilerRTCompile.cmake]
>>
>>
>> Looks like making any of (a-c,e) false fixes the problems.
>>
>> Possible solutions are (at least):
>> A) Use a %exe_out macro in asan lit tests.
>>   This implies a big (yet trivial) diff and a small-but-annoying
>> maintenance burden for new tests.
>>
>> B) Add "-o" support to clang-cl, this is being discussed here.
>>
>> C) We can un-override %clangxx_asan to be clang rather than clang-cl on
>> Windows
>>
>
> That could work.
>
>
>>   Or we could use a wrapper for clang-cl which replaces "-o" with
>> "-Fo" when lit tests are run...
>>   (there's a "android_commands/android_compile.py" wrapper around anyways)
>>
>
> No wrapper please, a macro would be better.
>
>
>> E) We could set COMPILER_RT_TEST_COMPILER to clang (rather than
>> clang-cl) and override it for gtest-based tests.
>>
>> Opinions?
>>
>
I'm fine with using clang instead of clang-cl for lit tests. Option E
sounds slightly better.


-- 
Alexey Samsonov, Mountain View, CA
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140519/c645e2f8/attachment.html>


More information about the cfe-commits mailing list