[PATCH] D53705: [OpenCL] Postpone PSV address space diagnostic

[John McCall via Phabricator via cfe-commits]

rjmccall added a comment.

In https://reviews.llvm.org/D53705#1280264, @Anastasia wrote:

> In https://reviews.llvm.org/D53705#1279086, @rjmccall wrote:
>
> > In https://reviews.llvm.org/D53705#1279068, @svenvh wrote:
> >
> > > Unlikely, since address spaces are provided in a different way in OpenCL C++ vs OpenCL C.
> > >
> > > OpenCL C provides qualifiers such as `global` as part of the language.  OpenCL C++ provides template classes such as `cl::global<T>` through a header file.
> >
> >
> > So OpenCL C code has to be completely rewritten if it needs to be used as part of an OpenCL C++ program?  And it doesn't really compose like a type if it's supposed to change how a variable is stored.  What a terrible little mess they've made for themselves.
>
>
> Fair point. I would like to allow regular OpenCL address space qualifiers as a Clang feature at least, in case we won't be able to alter the spec. But one problem is that the `private` address space qualifier keyword conflicts with the `private` class access modifier. I guess we can only allow the address space qualifiers with the `__` prefix. So some existing code would have to be changed when ported to OpenCL C++, but it should be easier than rewriting using classes.


That's actually a really easy ambiguity to handle given the current uses of the access-modifier keywords.  You just don't parse `private` as an access modifier if it's not followed by a colon.


Repository:
  rC Clang

https://reviews.llvm.org/D53705