[cfe-dev] [RFC] Delayed target-specific diagnostic when compiling for the devices.

[David Greene via cfe-dev]

"Finkel, Hal J. via cfe-dev" <cfe-dev at lists.llvm.org> writes:

>> Or "let's add a hack in favor of user code".
>>
> I disagree. I believe the resulting model is more consistent both for us
> and for our users. The model is: types are always fine, but specific
> operations might not be allowed.

This is almost certainly what users will expect.  I'm strongly +1 on
everything Hal has said.  Making porting code as easy as possible should
be a high priority.

                               -David