[cfe-commits] [PATCH] Rejecting forbidden storage class specifiers in OpenCL

[George Russell]

Hi,

I have attached an updated patch addressing each of the issues mentioned.

Regards,
George Russell

On 6 December 2010 21:00, Peter Collingbourne <peter at pcc.me.uk> wrote:
> Hi George,
>
> On Mon, Nov 29, 2010 at 11:15:26PM +0100, George Russell wrote:
>> The OpenCL specification states that the storage class specifiers
>> "extern","auto","static", and "register" are prohibited (in section
>> 6.8.g, Restrictions).
>>
>> This patch adds diagnostics for each of these, when the OpenCL
>> language is selected, and a test case.
>>
>> Feedback?
>
> A few points:
>
> The test should be moved to Sema (in DeclSpec::SetStorageClassSpec).
> This will decrease code repetition, and Sema seems a more logical
> place to perform the test.
>
> We should forbid __private_extern__ also, I think.  While this storage
> class is not mentioned in the OpenCL spec, it isn't in C99 either,
> and doesn't make sense for OpenCL.
>
> You should add the .cl extension to test/lit.cfg or the test case
> will never run.  Also, our parser tests tend to use -fsyntax-only
> (which causes only the parser and semantic analysis to run).
>
> A single parametrised diagnostic would be better than one for each
> storage class, as this decreases repetition as well as burden on
> translators (if/when we get i18n support).
>
> Less important: your code applies the storage-class specifier even
> if it is prohibited, which doesn't seem right to me -- if we were
> to downgrade this diagnostic to a warning, the resulting AST should
> still be valid.
>
> Thanks,
> --
> Peter
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: cl_storage_class_spec.diff
Type: application/octet-stream
Size: 7256 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20101212/3268d599/attachment.obj>