r173952 - [preprocessor] Don't warn about "disabled expansion of recursive macro"

[Richard Smith]

On Wed, Jan 30, 2013 at 2:16 PM, Richard Smith <richard at metafoo.co.uk> wrote:
> On Wed, Jan 30, 2013 at 1:39 PM, Douglas Gregor <dgregor at apple.com> wrote:
>>
>> On Jan 30, 2013, at 1:23 PM, Richard Smith <richard at metafoo.co.uk> wrote:
>>
>>> On Wed, Jan 30, 2013 at 12:44 PM, Argyrios Kyrtzidis <akyrtzi at gmail.com> wrote:
>>>> On Jan 30, 2013, at 11:54 AM, Abramo Bagnara <abramo.bagnara at gmail.com> wrote:
>>>>
>>>>> Il 30/01/2013 19:55, Argyrios Kyrtzidis ha scritto:
>>>>>> Author: akirtzidis
>>>>>> Date: Wed Jan 30 12:55:52 2013
>>>>>> New Revision: 173952
>>>>>>
>>>>>> URL: http://llvm.org/viewvc/llvm-project?rev=173952&view=rev
>>>>>> Log:
>>>>>> [preprocessor] Don't warn about "disabled expansion of recursive macro"
>>>>>> for "#define X X".
>>>>>>
>>>>>> This is a pattern that, for example, stdbool.h uses.
>>>>>> rdar://12435773
>>>>>
>>>>> Please revert this commit: this has already been discussed in detail in
>>>>> past and we decided otherwise. Last time it has been discussed in this
>>>>> thread:
>>>>>
>>>>> http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20121029/067098.html
>>>>
>>>> Reverted in r173970.
>>>>
>>>>>
>>>>>>
>>>>>> Modified:
>>>>>>   cfe/trunk/lib/Lex/PPMacroExpansion.cpp
>>>>>>   cfe/trunk/test/Headers/stdbool.cpp
>>>>>>   cfe/trunk/test/Preprocessor/warn-disabled-macro-expansion.c
>>>>>>
>>>>>> Modified: cfe/trunk/lib/Lex/PPMacroExpansion.cpp
>>>>>> URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Lex/PPMacroExpansion.cpp?rev=173952&r1=173951&r2=173952&view=diff
>>>>>> ==============================================================================
>>>>>> --- cfe/trunk/lib/Lex/PPMacroExpansion.cpp (original)
>>>>>> +++ cfe/trunk/lib/Lex/PPMacroExpansion.cpp Wed Jan 30 12:55:52 2013
>>>>>> @@ -459,7 +459,10 @@ bool Preprocessor::HandleMacroExpandedId
>>>>>>      if (MacroInfo *NewMI = getMacroInfo(NewII))
>>>>>>        if (!NewMI->isEnabled() || NewMI == MI) {
>>>>>>          Identifier.setFlag(Token::DisableExpand);
>>>>>> -          Diag(Identifier, diag::pp_disabled_macro_expansion);
>>>>>> +          // Don't warn for "#define X X" like "#define bool bool" from
>>>>>> +          // stdbool.h.
>>>>>
>>>>> Would you be so kind to add a comment here to remember that the choice
>>>>> to have this warning for thar case (although disabled by default) is
>>>>> deliberate?
>>>>
>>>> Just to make sure, are you of the opinion that we should warn when <stdbool.h> is included and its macros are used ?
>>>
>>> As discussed on that thread, we should modify <stdbool.h> to suppress
>>> the warning.
>>
>> How do you propose to do that? #pragmas don't help, because the diagnostic comes from the expansion location, not the definition location. Changing #define bool bool to something that involves _Pragma is very likely to break code elsewhere.
>
> To my reading of C11 6.10.3.4, that's a bug in Clang's preprocessor.
> _Pragmas are supposed to take effect (and, per 6.10.9/1, be removed
> from the token stream) during macro rescanning, so I think this should
> work:
>
> #define true _Pragma("clang diagnostic push") _Pragma("clang
> diagnostic ignored \"-Wdisabled-macro-expansion\"") true
> _Pragma("clang diagnostic pop")
>
> #define CAT2(x,y) x##y
> #define CAT(x,y) CAT2(x,y)
> bool b = CAT(true,n);
>
> ... because the pragma should take effect during the rescan of the
> expansion of 'true', before we perform the token paste.
>
> Another example we seem to get wrong:
>
> #define foo _Pragma("GCC diagnostic ignored \"-Wuninitialized\"")
> #define str(x) #x
> #define STR(x) str(x)
> const char *p = STR(foo);
> int n() { int k; return k; }
>
> This should produce an empty string and no warning, as far as I can tell.

FWIW, gcc gets this right, producing an empty string and suppressing
the warning here:

#define foo _Pragma("GCC diagnostic ignored \"-Wuninitialized\"")
#define str(x) #x
#define STR(x) str(x)
void f() {
const char *p = ({
STR(foo);
});
}
int n() { int k; return k; }

(extra gunk to get around gcc's restrictions on where pragmas can appear).