[cfe-dev] How to deal with mangled names matching with extern "C" ones?

[Andrey Bokhanko]

Fix prepared, submitted for review: http://reviews.llvm.org/D11297.

It fixes the problem described in my initial message and PR17829. It
doesn't cover cases from Richard Smith's message, which are a subject
of a separate fix.

Andrey

On Wed, Jul 8, 2015 at 5:04 PM, Andrey Bokhanko
<andreybokhanko at gmail.com> wrote:
> OK, I will add checks in CodeGen and will send a patch for review.
>
> Yours,
> Andrey
>
>
> On Wed, Jul 8, 2015 at 12:08 AM, Richard Smith <richard at metafoo.co.uk> wrote:
>> On Tue, Jul 7, 2015 at 8:18 AM, Reid Kleckner <rnk at google.com> wrote:
>>>
>>> I think we should just fix the assert by erroring out from CodeGen when it
>>> detects that two Decls have the same mangled name. This might be tricky
>>> because we usually assume that decls have a unique mangled name.
>>
>>
>> That is both reasonable and what we already attempt to do, but we don't do
>> so particularly thoroughly or consistently. For instance:
>>
>>   int _Z1fv;
>>   void f() {} // gives an error "definition with same mangled name as
>> another definition"
>>
>> but:
>>
>>   void f() {}
>>   int _Z1fv; // no error, variable silently replaces function
>>
>>>
>>> If people want to compile C++ transpiled to C, I don't see why clang
>>> should forbid that. The Itanium ABI restricts itself to valid C identifiers
>>> to support this exact use case.
>>
>>
>> I agree. While it may be reasonable to warn on reserved identifiers outside
>> system headers, we should still accept them when they don't conflict with
>> some other meaning we've assigned to them. CodeGen needs to be able to deal
>> with name conflicts, and should avoid crashing when they occur. And this
>> problem is not limited to C++ name mangling; we get a conflict here too:
>>
>>   int n = 1;
>>   int m __asm__("n") = 2;
>>
>>>
>>> On Tue, Jul 7, 2015 at 7:44 AM, Andrey Bokhanko <andreybokhanko at gmail.com>
>>> wrote:
>>>>
>>>> Hmmm, maybe we should just check that identifiers don't start with
>>>> _[A-Z] and print an error if they are?
>>>>
>>>> Richard [Smith], what do you think?
>>>>
>>>> Yours,
>>>> Andrey
>>>>
>>>>
>>>> On Tue, Jul 7, 2015 at 2:39 PM, Renato Golin <renato.golin at linaro.org>
>>>> wrote:
>>>> > On 7 July 2015 at 12:20, Stephan Bergmann <sbergman at redhat.com> wrote:
>>>> >> [lex.name]/3.1: "Each identifier that contains a double underscore __
>>>> >> or
>>>> >> begins with an underscore followed by an
>>>> >> uppercase letter is reserved to the implementation for any use."
>>>> >
>>>> > Right, I always forget the _[A-Z] case. I guess how we deal with this
>>>> > is up to interpretation.
>>>> >
>>>> > cheers,
>>>> > --renato
>>>> > _______________________________________________
>>>> > cfe-dev mailing list
>>>> > cfe-dev at cs.uiuc.edu
>>>> > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>>> _______________________________________________
>>>> cfe-dev mailing list
>>>> cfe-dev at cs.uiuc.edu
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>>
>>>
>>