(Closes #2515) fix bugs in symbol-table merge

[arporter]

As it says. Primarily this is about making sure we don't rename symbols that are being imported into scope from elsewhere. Broken out of #924 as that is getting big.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 99.86%. Comparing base (ed742fd) to head (4c86703).

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2517   +/-   ##
=======================================
  Coverage   99.86%   99.86%           
=======================================
  Files         357      357           
  Lines       48234    48300   +66     
=======================================
+ Hits        48168    48234   +66     
  Misses         66       66           

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[arporter]

Ready for a first look. Probably one for @sergisiso as it adds some SymbolTable functionality, as well as fixing some bugs. I've triggered the integration tests. Note that the sphinx-link-check in the CI has mysteriously cancelled itself. Also, CodeCov is having one of its moments and is not reporting on coverage.

[arporter]

Taking this back as the LFRic and NEMO integration tests failed :-(

[arporter]

Allowing for intrinsics has fixed the LFRic integration tests. However, the NEMO MO passthrough still fails:

Executing:psyclone --limit output -api nemo -oalg /dev/null -opsy diatmb.f90 cfgs/MO_GO8_gnu/BLD/ppsrc/nemo/diatmb.f90
...
psyclone.psyir.symbols.symbol.SymbolError: PSyclone SymbolTable error: A symbol named 'iom_put' is present but unresolved in both tables and they do not share a wildcard import that could be bringing it into scope.

This is in a routine that has no imports. 'iom_put' is brought into scope by a wildcard import from iom at module level.

[arporter]

The problem is in the Routine handler in the Fortran backend: we flatten all tables by populating a brand-new one. However, the brand-new one is not (initially) connected into the node hierarchy and therefore the search for a common wildcard import returns nothing.

[arporter]

Finally got all of the integration tests to pass. Really ready for review now!

[arporter]

CodeCov really isn't happy so I've checked the coverage for symbol_table.py locally and all is fine. I've tried firing off the CI again so maybe CodeCov will sort itself out. Anyway, ready for another look now.

[arporter]

Should be ready for another look now, integration tests permitting.