On 06/06/2016 02:04 AM, Francisco Jerez wrote:
Vedran Miletić <ved...@miletic.net> writes:
Aside from working just like NVIDIA and AMD proprietary stacks, no.
However, what we do right now is most certainly broken. Consider the
following argument
-I"/foo bar/baz"
which is going to be sent to Clang as two arguments broken over space,
while it should be one.
But the OpenCL spec doesn't impose any explicit requirements on the
formatting of the option string (unless I'm missing something), so
strictly speaking both interpretations are equally broken. It wasn't my
intention to argue against implementing the same behaviour as the
proprietary CL stacks as temporary solution, but the fact that this is
unspecified is certainly an OpenCL spec bug.
Agree that the spec is incomplete. Additional problem is that Clang is
unable to handle the incorrectly split arguments, in which case Clang
IMHO does the correct thing. Mesa is right now breaking the user/app
supplied compiler args into an array before passing it to Clang and that
breaking is done in a wrong way.
I am not claiming that it's Clover's job to unquote args, even though it
would be convenient for Clang if it was.
Regards,
Vedran
We should split arguments correctly, and whether or not we remove
quotes in the process or we expect Clang to handle them is the easier
part of the job.
--
Vedran Miletić
vedran.miletic.net
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
