[cfe-commits] [Patch] Produce errors for 'static' and type qualifiers in array declarators

[John McCall]

On Aug 14, 2012, at 6:16 AM, Hans Wennborg wrote:
> On Sun, Aug 12, 2012 at 4:20 PM, Hans Wennborg <hans at chromium.org> wrote:
>> Thanks for the review! New patch attached.
>> 
>> On Sat, Aug 11, 2012 at 8:14 PM, John McCall <rjmccall at apple.com> wrote:
>>> +        unsigned x = chunkIndex;
>>> +        for (;;) {
>>> +          // Walk outwards along the declarator chunks.
>>> +          if (x == 0)
>>> +            break;
>>> +          x--;
>>> 
>>> This can just be
>>>  while (x != 0) {
>>>    x--;
>> 
>> Yeah, that's better.
>> 
>>> +          switch (DC.Kind) {
>>> +          case DeclaratorChunk::Paren:
>>> +            continue;
>>> +          case DeclaratorChunk::Array:
>>> +          case DeclaratorChunk::Pointer:
>>> 
>>> We try to avoid 'default' cases, and this is a good example why.  I claim
>>> that there are no declarator chunks which are actually valid here except
>>> Paren.
>> 
>> Right.
>> 
>>> Suppressing the error when the chunk is a Function or BlockPointer
>>> is fine, since we'll still get the more-important diagnostic that the type is
>>> invalid.  We should not suppress the error when the chunk is Reference
>>> or MemberPointer, though.
>> 
>> I was thinking that in these cases we would get errors about trying to
>> use a C99 feature in C++, but maybe it's better not to ignore it.
>> 
>>> Also, if someone adds a new declarator chunk
>>> (not entirely impossible), they should have to consider what to do in this
>>> case.
>> 
>> That's a good point.
>> 
>> Thanks,
>> Hans
> 
> Ping?
> <static_in_array_decls3.patch>

Looks great, thanks!  Might as well add a test for the function and block-pointer
cases (to verify that an error is emitted, but not this one), but feel free to commit.

John.