[cfe-commits] r126134 - in /cfe/trunk: lib/Parse/ParseDecl.cpp test/SemaObjC/auto-objective-c.m

[Richard Smith]

On Tue, February 22, 2011 01:19, Douglas Gregor wrote:
> On Feb 21, 2011, at 5:09 PM, Richard Smith wrote:
>> On Tue, February 22, 2011 00:50, Douglas Gregor wrote:
>>> On Feb 21, 2011, at 4:44 PM, Richard Smith wrote:
>>>> On Mon, February 21, 2011 18:37, Fariborz Jahanian wrote:
>>>>
>>>>> Author: fjahanian
>>>>> Date: Mon Feb 21 12:37:13 2011
>>>>> New Revision: 126134
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> URL: http://llvm.org/viewvc/llvm-project?rev=126134&view=rev
>>>>> Log:
>>>>> Turn on 'auto' in plain  objc mode.
>>>>>
>>>>>
>>>>
>>>> Without trailing return type support, this produces unhelpful
>>>> diagnostics:
>>>>
>>>>
>>>>
>>>> $ echo -e 'auto a();' | clang++ -xobjective-c -
>>>> <stdin>:1:1: error: 'auto' return without trailing return type
>>>> auto a(); ^ 1 error generated.
>>>> $ echo -e 'auto a() -> void;' | clang++ -xobjective-c -
>>>> <stdin>:1:10: error: expected function body after function
>>>> declarator auto a() -> void; ^ 1 error generated.
>>>>
>>>>
>>>>
>>>> It should be straightforward to teach GetTypeForDeclarator that
>>>> objective-c (or perhaps anything that's not C++) is a special case,
>>>> if you don't want trailing return types there too.
>>>
>>> Good point! We don't want trailing return types in Objective-C.
>>>
>>
>> This should sort it out.
>>
>
> Hey, thanks!
>
>
> -      !D.isFunctionDeclarator()) {
> +      (!getLangOptions().CPlusPlus0x || !D.isFunctionDeclarator())) {
>
>
> Maybe this should just be getLangOptions().CPlusPlus, in case we do the
> previously-threatened trick of catching some uses of 'auto' as a
> type-specifier in C++?

r126174. The right thing for this check will depend on whether we want to
support trailing return types in C++03, or just deduced auto; this'll need
revisiting if we want trailing return types as an extension too.

My apologies to fjahanian for any duplicated effort!

Richard