[flang] gfortran expects semantic errors in some assumed-rank tests.

[jeanPerier]

Need to triage if semantic errors expected by gfortran test suite are legit:

regression:

• c_sizeof_7.f90

regression/c-interop:

• c535b-3.f90
• c535c-1.f90
• c535c-2.f90
• c535c-3.f90
• c535c-4.f90

[llvmbot]

@llvm/issue-subscribers-flang-frontend

Author: None (jeanPerier)

Need to triage if semantic errors expected by gfortran test suite are legit:

regression:

• c_sizeof_7.f90

regression/c-interop:

• c535b-3.f90
• c535c-1.f90
• c535c-2.f90
• c535c-3.f90
• c535c-4.f90

[klausler]

For c_sizeof_7.f90: The F'2023 definition of C_SIZEOF doesn't require a whole array object. 18.2.3.8: "X shall be a data entity with interoperable type and type parameters, and shall not be an assumed- size array, an assumed-rank array that is associated with an assumed-size array, an unallocated allocatable variable, or a pointer that is not associated." and
"If X is an array, the result value is the value that the companion processor returns as the result of applying the sizeof operator to an object of a type that interoperates with the type and type parameters of X, multiplied by the number of elements in X."

For c535b-3.f90, we allow the TARGET argument to be assumed-rank if the POINTER argument also is assumed-rank, since we allow pointer assignment in that case as well.

For c535c-1.f90 through c535c-4.f90, errors should be emitted for actual arguments corresponding to INTENT(OUT) assumed-rank dummy arguments when the actual argument size cannot be known (e.g., assumed-size). Will fix. I think that some of the cases in those tests are conforming, though.

[klausler]

#111204