[Lldb-commits] [lldb] r210035 - Fix most of the remaining Windows build warnings.

[Todd Fiala]

Thanks, Zachary.

Ed - I'm testing the patch on my FreeBSD VM.  Finally firing it up again...


On Tue, Jun 17, 2014 at 1:13 PM, Zachary Turner <zturner at google.com> wrote:

> Yes, this was for building with MSVC, which does not support those linker
> options.  I guess either the patch you sent out earlier, or explicitly
> excluding those compilers which don't support this flag works.
>
>
> On Tue, Jun 17, 2014 at 1:11 PM, Todd Fiala <tfiala at google.com> wrote:
>
>> Zachary - were you trying to build on Windows without a gcc-compatible
>> compiler, and getting linker errors/warnings in that case?
>>
>> If so, a correct check might really be the final patch I sent out earlier:
>>
>> diff --git a/CMakeLists.txt b/CMakeLists.txt
>> index bad83e8..4cfa1d9 100644
>> --- a/CMakeLists.txt
>> +++ b/CMakeLists.txt
>> @@ -218,7 +218,8 @@ macro(add_lldb_library name)
>>     endif ()
>>
>>    if(LLDB_USED_LIBS)
>> -    if (LLVM_COMPILER_IS_GCC_COMPATIBLE)
>> +    # The Darwin linker doesn't understand --start-group/--end-group.
>> +    if (LLVM_COMPILER_IS_GCC_COMPATIBLE AND NOT "${CMAKE_SYSTEM_NAME}"
>> MATCHES "Darwin")
>>
>>
>>
>> On Tue, Jun 17, 2014 at 1:02 PM, Todd Fiala <tfiala at google.com> wrote:
>>
>>> This latest patch works on Ubuntu12.04/gcc4.8.2 and MacOSX10.9.3/clang.
>>>  But thinking more about it, the predicate as a whole (changing based on
>>> compiler rather than Linker) is just incorrect.  So the platform checks
>>> there before, which isn't a bad proxy for the system linker requirements,
>>> is a better check than the compiler test.
>>>
>>> Zachary - what were you originally intending to address with the change?
>>>  Which platform(s) were affected by the patch?  We can possibly handle that
>>> via another mechanism.  In hindsight I'm not a big fan of testing the
>>> compiler since it isn't enough to get this right on all platofmrs
>>>
>>> -Todd
>>>
>>>
>>> On Tue, Jun 17, 2014 at 12:43 PM, Todd Fiala <tfiala at google.com> wrote:
>>>
>>>> Keying off that link executable variable won't work.  Here's what it's
>>>> set to on MacOSX:
>>>>
>>>> CMAKE_CXX_LINK_EXECUTABLE=<CMAKE_CXX_COMPILER>  <FLAGS>
>>>> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS>  -o <TARGET> <LINK_LIBRARIES>
>>>>
>>>> Ok - so maybe what we need is this patch (which I'm testing over here
>>>> now):
>>>> diff --git a/CMakeLists.txt b/CMakeLists.txt
>>>> index bad83e8..4cfa1d9 100644
>>>> --- a/CMakeLists.txt
>>>> +++ b/CMakeLists.txt
>>>> @@ -218,7 +218,8 @@ macro(add_lldb_library name)
>>>>    endif ()
>>>>
>>>>    if(LLDB_USED_LIBS)
>>>> -    if (LLVM_COMPILER_IS_GCC_COMPATIBLE)
>>>> +    # The Darwin linker doesn't understand --start-group/--end-group.
>>>> +    if (LLVM_COMPILER_IS_GCC_COMPATIBLE AND NOT "${CMAKE_SYSTEM_NAME}"
>>>> MATCHES "Darwin")
>>>>        target_link_libraries(${name} ${cmake_2_8_12_PUBLIC}
>>>>                              -Wl,--start-group ${LLDB_USED_LIBS}
>>>> -Wl,--end-group)
>>>>      else()
>>>>
>>>>
>>>>
>>>> On Tue, Jun 17, 2014 at 12:31 PM, Todd Fiala <tfiala at google.com> wrote:
>>>>
>>>>> It looks like we might be able to test ${CMAKE_CXX_LINK_EXECUTABLE}
>>>>> to exclude those that don't need --start-group/--end-group linker flags.
>>>>>
>>>>> However, this is getting pretty far past the original contributor
>>>>> patch.  I'm more inclined to roll it back at this point until the whole
>>>>> thing gets more testing.
>>>>>
>>>>>
>>>>> On Tue, Jun 17, 2014 at 12:25 PM, Todd Fiala <tfiala at google.com>
>>>>> wrote:
>>>>>
>>>>>> Ok - so given it's a linker related flags issue, what is the correct
>>>>>> cmake way to test out the linker being used?
>>>>>>
>>>>>>
>>>>>> On Tue, Jun 17, 2014 at 11:49 AM, Ed Maste <emaste at freebsd.org>
>>>>>> wrote:
>>>>>>
>>>>>>> On 17 June 2014 14:33, Todd Fiala <tfiala at google.com> wrote:
>>>>>>> >
>>>>>>> > Hey Ed - can you test this out on FreeBSD?  This is a patch that
>>>>>>> fixes cmake on MacOSX.  I verified it works on MacOSX and Linux with
>>>>>>> cmake/ninja.
>>>>>>> >
>>>>>>> > The original code looks like it was opting in for the
>>>>>>> --start-group/--end-group linker options on FreeBSD, which this patch would
>>>>>>> change but might be a no-op on your end.
>>>>>>>
>>>>>>> It's odd that this is a compiler-dependent test, I'd expect it to be
>>>>>>> a
>>>>>>> linker issue.  In any case, we definitely needed --start-group and
>>>>>>> --end-group in the past.  If this test works with Linux though it
>>>>>>> will
>>>>>>> probably work with FreeBSD too.  I'll test it.
>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Todd Fiala | Software Engineer |  tfiala at google.com |  650-943-3180
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Todd Fiala | Software Engineer |  tfiala at google.com |  650-943-3180
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Todd Fiala | Software Engineer |  tfiala at google.com |  650-943-3180
>>>>
>>>
>>>
>>>
>>> --
>>> Todd Fiala | Software Engineer |  tfiala at google.com |  650-943-3180
>>>
>>
>>
>>
>> --
>> Todd Fiala | Software Engineer |  tfiala at google.com |  650-943-3180
>>
>
>


-- 
Todd Fiala | Software Engineer | tfiala at google.com | 650-943-3180
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-commits/attachments/20140617/a51d3eec/attachment.html>