r209172 - Get ARCMT/GC-check-warn-nsalloc.m working

[Alp Toker]

On 20/05/2014 02:38, Jordan Rose wrote:
>> On May 19, 2014, at 16:36, Alp Toker <alp at nuanti.com> wrote:
>>
>>
>> On 20/05/2014 02:10, Jordan Rose wrote:
>>>> On May 19, 2014, at 15:51, Alp Toker <alp at nuanti.com <mailto:alp at nuanti.com>> wrote:
>>>>
>>>> -  Diags->setDiagnosticMapping(diag::err_arcmt_nsalloc_realloc,
>>>> -                              NoNSAllocReallocError ? diag::MAP_WARNING
>>>> -                                                    : diag::MAP_ERROR,
>>>> -                              SourceLocation());
>>>> +  if (NoNSAllocReallocError)
>>>> +    Diags->setDiagnosticMapping(diag::warn_arcmt_nsalloc_realloc,
>>>> +                                diag::MAP_IGNORE, SourceLocation());
>>>>
>>> This changes the behavior of the existing flag from (effectively) -Wno-error= to -Wno-. I don't think we should do that—any existing tools of ours that rely on the frontend flag probably still want the warning, and new tools should just use -W*.
>> You'll have to choose -- we can't support both.
>>
>> If we give -no-ns-alloc-error control over the diagnostic mapping, the -W* mappings will no longer take precedence and the diagnostic won't need a named group (so it'll end up the way it was before, but actually working now). That seems fine by me -- ready to commit if it sounds good to you.
> Fariborz is the one dealing with the migrators these days. Argyrios may also have an opinion, since he put a lot of work in in the past.
>
> I don't know myself who's actually using this flag (even internally).

I've gone ahead and minimised the changes in r209184 so the diagnostic 
works similarly to the original intention but with the bug fixed, and 
it's again no longer controlled by a -W* flag.

Signing off and handing over to Fariborz for any fine tuning now that 
the tests and feature are fixed.

Cheers,
Alp.

-- 
http://www.nuanti.com
the browser experts