Add support for WHERE with imported symbols

[sergisiso]

WHEREs with imported symbols create a CodeBlock with:
PSyclone doesn't yet support reference to imported symbols inside WHERE clauses.

Now that we can resolve modules in the frontend before iterating down to the execution section we can potentially fix this. Or improve the message to provide the instructions to fix it.

[RobWatersMet]

Just to second this as we have a number of these in UKCA.

For example:

https://code.metoffice.gov.uk/trac/ukca/browser/main/trunk/src/science/core/chemistry/asad/asad_hetero.F90#L283
https://code.metoffice.gov.uk/trac/ukca/browser/main/trunk/src/science/core/chemistry/asad/asad_hetero.F90#L322

[JonElsey]

Are there any plans to add this feature? There are a lot of these within UKCA aerosol also. Or is there a way of ignoring this error specifically within a transformer script? I'd like to avoid using the force argument wholesale when applying OpenMP transformations, but being able to bypass this error would increase coverage a fair bit.

Thanks!

[LonelyCat124]

We would like to add this at some stage, its just difficult to prioritise right now. If its fairly small I might be able to have a look.

W.r.t workaround in scripts, there is something you could try (though its not the most resilient)

If you have:

try:
   mytrans.apply(nodes)
except TransformationError:
   pass

you could inspect the transformation error and then go a bit further, e.g.:

try:
    mytrans.apply(nodes)
except TransformationError as err:
    if "Nodes of type 'CodeBlock' cannot be" in str(err.value):
        cbs = nodes.walk(CodeBlock)
        for block in cbs:
             if "PSyclone doesn't yet support reference to imported symbols inside WHERE clauses." not in cbs.preceding_comment:
                 break
        else:
            try:
                mytrans.apply(nodes, force=True) # Or opts={force: True} depending on your PSyclone version and the transformation
                # Instead of force you can also try `node_type_check=False`
            except:
                 pass

NB. I've not tested this and some of the messsage might not quite be correct, and you may end up with some spurious comments in the final code, but it may work as a workaround for now. Its not resilient as it will stop working if any of the comments/error messages change - you could check whats inside the CodeBlock explicitly but thats also a bit complicated since it involves fparser.

For purely seeing how performance is though it might work - you could always add a TODO referencing this issue.

[sergisiso]

This is a problem in NEMOv5 icethd_sal.f90. I don't think this should be a limitation now that we have RESOLVE_IMPORTS, so I am going to have another look at this.