[patch] Fix pr16247

[Douglas Gregor]

On Jun 17, 2013, at 11:02 AM, Richard Smith <richard at metafoo.co.uk> wrote:

> On Mon, Jun 17, 2013 at 9:21 AM, Douglas Gregor <dgregor at apple.com> wrote:
>> 
>> On Jun 11, 2013, at 6:02 AM, Rafael Espíndola <rafael.espindola at gmail.com>
>> wrote:
>> 
>> On 10 June 2013 19:08, Rafael Espíndola <rafael.espindola at gmail.com> wrote:
>> 
>> Richard pointed out on IRC that the patch should look at the
>> redeclaration context to avoid problems with one of the functions
>> being in a extern "C++". Implementing that found some interesting
>> problems. Consider
>> 
>> extern "C" {
>> static void foo(int x);
>> static void foo() {
>> }
>> void foo(int x) {
>> }
>> }
>> 
>> This should be valid, since both functions have internal linkage and
>> therefore none of them has C language linkage. Commenting the first
>> declaration makes the code invalid as now the last one is extern C and
>> [dcl.link] p6 kicks in. This means we have to do "normal" overload
>> resolution first to see if the last decl we are looking at has C
>> language linkage or not.
>> 
>> 
>> BTW, is this sufficient evidence that we should just give static
>> functions C language linkage? I just checked http://gcc.godbolt.org/
>> and gcc 4.8 and icc 13 (which is edg based, no?), reject
>> 
>> extern "C" {
>> static void foo() {
>> }
>> static void foo(int x) {
>> }
>> }
>> 
>> We are putting quiet a bit of effort to make sure we accept it and I
>> still can't see the value.
>> 
>> 
>> I, too, find this behavior strange. Changing between static and non-static
>> shouldn’t affect whether a function can be overloaded; it’s completely
>> non-intuitive and apparently at odds with existing practice in GCC/EDG.
> 
> The latter, at least, is not the case; EDG accepts this code in both
> its compliant mode and in its g++-compatible mode. I'm surprised that
> ICC rejects this.

Ah, interesting. Might be one of the many compatibility knobs that EDG provides.

> The former makes some degree of sense to me -- a
> 'static' declaration doesn't have external linkage and thus doesn't
> have C linkage -- although I agree that it's weird.


So existing practice is mixed, with a bias toward rejecting the code above (apparently, MSVC also rejects the code in question). The standard is clear that the code is well-formed, but we find that result to be weird. This is starting to feel like a case where we badger the committee into fixing the standard :)

	- Doug

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130617/a4221f1d/attachment.html>