(Towards #2165) add support for associate constructs.

[arporter]

This is an initial cut that we can improve upon if we need to. In particular, the solution chosen here replaces instances of associate-name with the corresponding expression. However, since that expression must be evaluated at the start of the construct, if any of the variables it references are written to within the construct then this substitution cannot be performed. We could refine this by checking the precise location at which any such writes are performed - if they happen to only occur after lines where we need to perform the substitution then this would still be OK.

Alternatively, we could attempt to determine the type of the expression and then store its value in a new variable at the start of the construct.

[codecov]

Codecov Report

Patch coverage: 99.50% and project coverage change: -0.01 ⚠️

Comparison is base (badc7f5) 99.84% compared to head (14c8bd5) 99.84%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2167      +/-   ##
==========================================
- Coverage   99.84%   99.84%   -0.01%     
==========================================
  Files         334      336       +2     
  Lines       45137    45239     +102     
==========================================
+ Hits        45068    45169     +101     
- Misses         69       70       +1     

Impacted Files	Coverage Δ
src/psyclone/psyir/frontend/fparser2.py	99.93% <98.18%> (-0.07%)	⬇️
src/psyclone/psyir/tools/__init__.py	100.00% <100.00%> (ø)
src/psyclone/psyir/tools/dependency_tools.py	100.00% <100.00%> (ø)
src/psyclone/psyir/tools/substitution_tool.py	100.00% <100.00%> (ø)
src/psyclone/psyir/transformations/inline_trans.py	100.00% <100.00%> (ø)
.../psyir/frontend/fparser2_associate_handler_test.py	100.00% <100.00%> (ø)

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[arporter]

I need to test this branch with some real NEMOVAR code before putting it as ready for review.

[arporter]

Of course, the very first NEMOVAR I tried this on had some WRITE stmts inside the associate construct so we're no further forward. I can refine by checking which symbols are referenced by the resulting CodeBlock but that may not help. If that's so then I think the only option is to remove the WRITES?

[arporter]

In the telco this morning @sergisiso reminded me that we can't handle Associate constructs that contain calls to impure routines either. I need to catch this case.

[arporter]

I could refine what I have to check that any Call or Write is not the last statement in the construct. Beyond that, I could check whether there are any reads of any associate names after a Call or write.

[arporter]

The ASSOCIATE constructs in the NV code we looked at today all contained calls and so this approach won't work. We could allow a user to permit calls within associate blocks but that feels dangerous and very hard to test. I think therefore that the only remaining option is to create temporary variables and that means we have to be able to determine their type.