[PSyData] Kernel extraction and derived-types as arguments or with any other write access

[hiker]

While #2706 adds support for derived type, this actually only works if components of a derived types are used, e.g.:

        DO j = tdims % j_start, tdims % j_end

This would add tdims%jstart and j_end to the kernel-data-file as expected. But if a whole derived type is passed on as an argument, the extraction/driver creation fails. Example:

TYPE(array_dims), SAVE :: tdims
...
CALL pc2_checks(...tdims % i_len, tdims % j_len, tdims % k_end, ...)     ! This works

CALL wtrac_pc2_phase_chg(tdims, q, qcl, 'pc2_initiate', wtrac = wtrac)   ! Doesn't work

In this case the driver creation fails because tdims has no intrinsic type. (which happens when the driver tries to create the corresponding _post variable)

As far as I can see, the only option here is to add code that can write the all individual components of a derived type, and make sure that the two different usages (tdims%i_len and tdims) are stored only once for the same object.

Suggested approach:

• If a structure type is detected in the access analysis, replace it with access to each individual element (recursively in case of nested derived types)
  • Not sure if this should become an option of the variable access collection, or a post-processing step
• Update the driver extraction (if still required) to read the individual components
• Verify the extraction. At this stage, I would expect the extraction to fail to compile (since there won't be a generic interface that can write the derived type