r211411 - Driver: Record that we're in crashdump and push flags to ConstructJob

[Justin Bogner]

Jordan Rose <jordan_rose at apple.com> writes:
> On Jun 23, 2014, at 15:18 , Justin Bogner <mail at justinbogner.com> wrote:
>
>> Jordan Rose <jordan_rose at apple.com> writes:
>>> Hm. .i isn't really correct for something that still has macros in
>>> it. IIRC Clang happens to expand them again, but if we've merely
>>> rewritten includes I would expect the resulting file to have a .c
>>> extension.
>> 
>> It's a bit tricky to get this right consistently in a crashdump context,
>> due to the way the logic is split between the driver and cc1. Notably, I
>> can fairly easily special case the driver here to use a .c extension for
>> crashdumps, but as of r211421 that isn't always right. We don't really
>> know if we're going to specify rewrite-includes or not at the time that
>> we decide the filename.
>> 
>> That said, the trade off is a matter of which is worse:
>> 
>> 1. Crashdump files named .i/.mi even though they still have macros when
>>   -fmodules isn't present, or
>> 
>> 2. Crashdump files named .c/.m even though they're fully preprocessed
>>   when -fmodules is present.
>> 
>> I suppose that, given we'd like to make rewrite-includes and module
>> crashdumps work together in the long term, (2) is a better trade off.
>> It also avoids changing the status quo. Agree?
>
> I think I agree as well. "Fully preprocessed" is a little weird when
> modules are involved anyway.

Sounds good. r211572.