Add transformation generated imports to driver or after inlining

[awnawab]

This PR provides the following fixes:

1. TemporariesPoolAllocator transformation now adds global variable imports to the driver for variables used to define array sizes.
2. HoistArraysTransformation adds imports used to define the sizes of hoisted variables.
3. The addition of module imports from inlined children in the InlineTransformation is fixed.
4. InlineTransformation now also adds any explicit interfaces from inlined children.

[github-actions]

Documentation for this branch can be viewed at https://sites.ecmwf.int/docs/loki/321/index.html

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 95.16%. Comparing base (eb793e2) to head (78d9f45).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #321   +/-   ##
=======================================
  Coverage   95.15%   95.16%           
=======================================
  Files         168      168           
  Lines       35454    35503   +49     
=======================================
+ Hits        33737    33786   +49     
  Misses       1717     1717           

Flag	Coverage Δ
lint_rules	96.38% <ø> (ø)
loki	95.14% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[reuterbal]

Thanks, looks great! Some small implementation pointers but otherwise all good.

[reuterbal]

This rebuilds the import_map in every iteration. Better to cache it before the list comprehension.

Suggested change

	item.trafo_data[self._key]["imported_sizes"] = [(d.type.module, d) for d in dims
	if str(d) in routine.import_map]
	import_map = routine.import_map
	item.trafo_data[self._key]["imported_sizes"] = [(d.type.module, d) for d in dims
	if d.name in import_map]

[reuterbal]

Same comment as above about caching import_map

[reuterbal]

Suggested change

	Resolve sequence association in calls to all member procedures (if `inline_internals = True`)
	or in calls to procedures that have been marked with an inline pragma (if `inline_marked = True`).
	Resolve sequence association in calls to all member procedures (if ``inline_internals = True``)
	or in calls to procedures that have been marked with an inline pragma (if ``inline_marked = True``).

RST syntax...

[awnawab]

Oops, not sure how that ended up here, thanks for spotting!

[reuterbal]

Can we not use intf.symbols here?

[awnawab]

Interface.body comprises of Subroutine objects and not just the symbols, so I did it like this to capture the whole Subroutine object. If using intf.symbols, I would have to add symbol.type.dtype.procedure to the body rather than the symbol. Either is fine by me, please let me know if you would like me to update to the latter.

[reuterbal]

Yes, I'm aware - but you're only inspecting the procedure symbols of these subroutines not the objects themselves, which is what intf.symbols should give you. Have a look at the implementation of Interface.symbols, which should be essentially the same as what you do here, but also includes the case of named interfaces etc:

loki/loki/ir/nodes.py

Lines 700 to 710 in 77114a9

	def symbols(self):
	"""
	The list of symbol names declared by this interface
	"""
	symbols = as_tuple(flatten(
	getattr(node, 'procedure_symbol', getattr(node, 'symbols', ()))
	for node in self.body # pylint: disable=not-an-iterable
	))
	if self.spec:
	return (self.spec,) + symbols
	return symbols

[awnawab]

Many thanks @reuterbal for the super prompt review and feedback 🙏 I've addressed all your comments, please have another look whenever you can!

[reuterbal]

Thanks for the quick implementation of changes! Looks good to me

[reuterbal]

Apologies, this has accrued conflicts now - @awnawab, would you mind rebasing this over main?