[cfe-commits] r164992 - /cfe/trunk/Makefile

[David Blaikie]

On Tue, Oct 2, 2012 at 1:34 PM, Bill Wendling <isanbard at gmail.com> wrote:
> On Oct 2, 2012, at 11:49 AM, David Blaikie <dblaikie at gmail.com> wrote:
>
>> On Tue, Oct 2, 2012 at 11:39 AM, Bill Wendling <isanbard at gmail.com> wrote:
>>> On Oct 2, 2012, at 11:35 AM, David Blaikie <dblaikie at gmail.com> wrote:
>>>
>>>> On Mon, Oct 1, 2012 at 11:19 PM, Bill Wendling <isanbard at gmail.com> wrote:
>>>>> Author: void
>>>>> Date: Tue Oct  2 01:19:15 2012
>>>>> New Revision: 164992
>>>>>
>>>>> URL: http://llvm.org/viewvc/llvm-project?rev=164992&view=rev
>>>>> Log:
>>>>> Use an environment variable instead of what's in the make cmd goals.
>>>>
>>>> Still makes me sad that this is an environment variable (admittedly
>>>> this isn't important - just an aesthetic issue for me)
>>>
>>> I know. I tried to make it work, but couldn't. The thing is that clang is built with MAKECMDGOALS == all, and not 'clang-only', which I would prefer.
>>
>> Where's that MAKECMDGOALS value coming from? just the LLVM build
>> invoking the Clang build?
>>
> It's set automatically by the 'make' program. So if you do this:
>
>         make fnord
>
> the MAKECMDGOALS will be 'fnord'. When we build clang it's set to 'all' because that's how the whole build system seems to work...
>
>>> So I had to default to making it a variable. :-(
>>>
>>>> - what's the
>>>> use case for this? It just seems like this should naturally fall out
>>>> from any build command that wants to just build clang. If it doesn't
>>>> then someone may get sub-optimal builds because they don't know about
>>>> this environment variable they need to use.
>>>
>>> This is for building clang with LTO and wanting only the clang and clang++ executables and libraries (no tools, etc.). The problem here is that LTO takes a *long* time to link things because of how much memory it takes. Building anything extraneous is a waste here.
>>
>> Yeah, I figured it was something like that - my point was that while
>> it's a big waste for this case, it's a small waste for every other
>> user that wants to just build the Clang installation binaries &
>> doesn't use this flag.
>>
> True. But they don't need to know about the flag. :) Here's the code in the top-level Makefile:
>
> ifeq ($(MAKECMDGOALS),clang-only)
>   BUILD_CLANG_ONLY := YES
>   DIRS := $(filter-out tools docs unittests, $(DIRS)) \
>           tools/clang tools/lto
>   OPTIONAL_DIRS :=
> endif
>
>>> What I did is link this environment variable with the 'clang-only' flag (see equivalent patch on llvm-commits). When someone specifies 'make clang-only', they won't get the other stuff.
>>
>> Hmm, OK - I'm not quite sure how that works just looking at the LLVM
>> Makefile, but I'll keep an eye out for the patch as I work through my
>> email backlog. If it does the right thing for any LLVM make target
>> then I can't complain too much - just a minor implementation wart at
>> that point.
>>
> I forgot to check the code in. (>.<) It's there now.

;) Makes sense now, thanks.

(idly: I wonder if the CMake build does this any better/more
finely-grained so that the outer build can build just the bits it
needs from clang, depending on the target.)

- David