[PATCH] [CMake] Introduce libLLVM.so as LLVM_ENABLE_SHARED

[NAKAMURA Takumi]

2014-03-05 8:15 GMT+09:00 Tobias Grosser <tobias at grosser.es>:
> On 03/05/2014 12:12 AM, NAKAMURA Takumi wrote:
>>
>> 2014-03-05 2:42 GMT+09:00 Tobias Grosser <tobias at grosser.es>:
>>>
>>> On 03/04/2014 06:06 PM, NAKAMURA Takumi wrote:
>>>>
>>>>
>>>> Hi chandlerc,
>>>>
>>>> Note, this includes; http://llvm-reviews.chandlerc.com/D2941
>>>
>>>
>>>
>>> Why do you include this? The patch in general looks unnecessary big. Is
>>> there a need to keep all this together? Some changes like the one above)
>>> are
>>> really cleanups, that could be committed ahead of time. This would make
>>> the
>>> actual patch review easier.
>>
>>
>> Sorry for my laziness since I didn't eliminate D2941 stuff. I will
>> work again later.
>>
>>>> **Why using OBJLIB?**
>>>> - --whole-archive is available only for GNU ld. Rather, I chose linking
>>>> whole OBJLIBs.
>>>> - No need to create *.a(s). Archive files may be created in parallel.
>>>>
>>>> **Why introducing DEPENDS instead of add_dependencies?**
>>>> - add_dependencies(target) works only for the target. Implicitly
>>>> generated
>>>> obj.LLVM*** are required to add deps to generated files.
>>>>
>>>> **Why llvm-tblgen and clang-tblgen are tweaked?**
>>>> - They should be built out of LINK_COMPONENTS with -static.
>>>>
>>>> **Why LLVM_ENABLE_SHARED turns off LLVM_DEAD_STRIP?**
>>>> - I think -ffunction-sections would not make sense for the big
>>>> libLLVM.so.
>>>
>>>
>>>
>>> Out of interest, what is the motivation for introducing a libLLVM.so? I
>>> currently use BUILD_SHARED_LIBS=ON to avoid long link times in my
>>> incremental development builds. I assume this is not the motivation for
>>> libLLVM.so? Would BUILD_SHARED_LIBS still work after this patch?
>>
>>
>> Autoconf's --enable-shared provides the functionality and some people
>> have wanted in cmake.
>> http://llvm.org/bugs/show_bug.cgi?id=15493
>
>
> That is motivation enough for me. Would be good to put this in the commit
> message.

Yes, sure. Since it would resolve the PR.

>> I am a fan of such "the big DLL".
>
>
> You did not answer the last question, will BUILD_SHARED_LIBS still work as
> today? This functionality is something I would like to maintain, as it can
> reduce linking time a lot.

I have not tried with both LLVM_ENABLE_SHARED and BUILD_SHARED_LIBS.
Although it would work, it might be redundant.
I'll rework for it.

IMO, BUILD_SHARED_LIBS might be hidden as ADVANCED.
It confuses novices.