[PATCH] Generate a clang CompilationDatabase when running CMake.

[Justin Lebar via llvm-commits]

Thanks a lot, Justin.  I don't have commit access; would you mind
pushing this patch for me?

-Justin

On Mon, Dec 14, 2015 at 11:42 AM, Justin Bogner <mail at justinbogner.com> wrote:
> Justin Lebar <jlebar at google.com> writes:
>>> We already document to do this on the command line in all of the tooling
>>> docs
>>
>> FWIW that wasn't where I looked as someone new to the project; it
>> doesn't show up in http://llvm.org/docs/GettingStarted.html.  Of
>> course that page can't be a guide to all conceivable options.
>>
>>> does it actually make sense to default this to on?
>>
>> My thought was, one less thing you have to remember to do.  It's
>> annoying enough that you have to add a .ycm_extra_config.
>>
>> In fact, while I have your attention...do you have any thoughts on how
>> we might make adding a .ycm_extra_config file easier?  We could check
>> one in to llvm/utils with instructions to copy it down to the root
>> directory.  The problem with just sticking it in the root dir is that
>> this file needs to specify where your object files are located.
>
> Sorry, I don't use YCM so I'm not really familiar with the problem.
>
>> That seems better than asking people to cargo-cult from the canonical
>> .ycm_extra_config, which requires some minor but nontrivial
>> modifications to work with llvm.
>>
>>> I suppose as long as this is fast to do (I haven't checked, is it?)
>>> there's no harm in it.
>>
>> Doesn't appear to be significant.  On my macbook pro, after warming up
>> the disk cache, I ran
>>
>> $ time cmake -G Ninja -DCMAKE_BUILD_TYPE=RelWithDebInfo ..
>>
>> Without the patch (ran 3 times):
>>
>> real 0m17.875s
>> user 0m10.563s
>> sys 0m5.935s
>>
>> real 0m18.335s
>> user 0m10.948s
>> sys 0m5.981s
>>
>> real 0m17.757s
>> user 0m10.562s
>> sys 0m5.836s
>>
>> With the patch:
>>
>> real 0m18.461s
>> user 0m10.737s
>> sys 0m6.215s
>>
>> real 0m18.243s
>> user 0m10.805s
>> sys 0m6.022s
>>
>> real 0m17.724s
>> user 0m10.715s
>> sys 0m6.009s
>
> Fair enough. This LGTM.
>
>> On Mon, Dec 14, 2015 at 10:53 AM, Justin Bogner <mail at justinbogner.com> wrote:
>>> Justin Lebar via llvm-commits <llvm-commits at lists.llvm.org> writes:
>>>> This generates a compile_commands.json file, which tells tools like
>>>> YouCompleteMe and clang_complete exactly how to build each source file.
>>>> ---
>>>>  CMakeLists.txt | 4 ++++
>>>>  1 file changed, 4 insertions(+)
>>>> diff --git a/CMakeLists.txt b/CMakeLists.txt
>>>> index fc46413..208c0a8 100644
>>>> --- a/CMakeLists.txt
>>>> +++ b/CMakeLists.txt
>>>> @@ -86,6 +86,10 @@ set(CMAKE_MODULE_PATH
>>>>    "${CMAKE_CURRENT_SOURCE_DIR}/cmake/modules"
>>>>    )
>>>>
>>>> +# Generate a CompilationDatabase (compile_commands.json file) for our build,
>>>> +# for use by clang_complete, YouCompleteMe, etc.
>>>> +set(CMAKE_EXPORT_COMPILE_COMMANDS 1)
>>>
>>> We already document to do this on the command line in all of the tooling
>>> docs, does it actually make sense to default this to on? I suppose as
>>> long as this is fast to do (I haven't checked, is it?) there's no harm
>>> in it.
>>>
>>>> +
>>>>  option(LLVM_INSTALL_UTILS "Include utility binaries in the
>>>> 'install' target." OFF)
>>>>
>>>>  option(LLVM_INSTALL_TOOLCHAIN_ONLY "Only include toolchain files
>>>> in the 'install' target." OFF)
>>>> _______________________________________________
>>>> llvm-commits mailing list
>>>> llvm-commits at lists.llvm.org
>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits