[PATCH] D38444: [compiler-rt] [cmake] Create dummy gtest target for stand-alone builds

[George Karpenkov via cfe-commits]

> On Oct 2, 2017, at 12:57 PM, Michał Górny via Phabricator <reviews at reviews.llvm.org> wrote:
> 
> mgorny added a comment.
> 
> In https://reviews.llvm.org/D38444#886138, @george.karpenkov wrote:
> 
>>> breaking stand-alone builds as a result
>> 
>> That's a strong statement. Could you clarify? We have a lot of buildbots performing standalone builds, and they are still green.
> 
> 
> I didn't know anyone actually added bots doing that. Are you sure we're talking about the same meaning of 'stand-alone'? Stand-alone == out of LLVM, against installed copy of LLVM.

Yes.
You are right though that bots I was referring to do not run unit tests.

> 
>  ninja -v -j16 -l0 check-all
>  ninja: error: '/var/tmp/portage/sys-libs/compiler-rt-sanitizers-9999/work/compiler-rt-sanitizers-9999/lib/asan/tests/gtest', needed by 'lib/asan/tests/dynamic/Asan-i386-calls-Dynamic-Test', missing and no known rule to make it
> 
> It's as broken as it could be since it depends on target that does not exist.

Right, by “works” I’ve meant that “it compiles”, not that “unit tests pass”.
My understanding is that running unit tests never meant to work,
as a freshly built clang is usually needed, and in standalone mode it is not available.
I could be wrong though, in that case I do not know.

Does everything magically work with this dummy target?

> 
>>> and the likeliness of people mistakenly adding more unconditional dependencies
>> 
>> That's a good point, however I'm not sure how your change would fix the problem.
>> As far as I remember targets in compiler-rt had quite a few dependencies which required checking for whether it is a standalone build.
>> 
>> In general, I'm not sure what this change would achieve, and the added complexity can always cause more bugs in the future.
> 
> My goal is to make things work again.
> 
> 
> Repository:
>  rL LLVM
> 
> https://reviews.llvm.org/D38444
> 
> 
>