[flang-commits] [flang] [flang][OpenACC] Reland data-clause designator path handling (PR #220781)

[Andre Kuhlenschmidt via flang-commits]

akuhlens wrote:

> Dropped diagnostics?

No, just me getting better at understanding what is and is not allowed. I am pretty sure this should be allowed. At least private does not conflict with copy.

>  !$acc parallel private(a(i)) private(a(i+1-1))

This is a shortfall of `evaluate::Fold()` not this code. For instance, if the second clause was `(private(1-1+i))` it would be recognized, deduplicated, and eliminate this error. The right place to fix this example in in constant folding.

>   !$acc parallel private(a) private(a(1:5))

I conceptually agree with you. I think this could be made to work. I am interested in seeing if @razvanlupusoru wants this case. 



https://github.com/llvm/llvm-project/pull/220781