AST-53: Pattern-matching objects

[ggreif]

https://dfinity.atlassian.net/browse/AST-53

In coverage.ml I am cutting corners, and reusing the infrastructure of plain products for labelled products, filling non-appearing field labels with wildcard patterns. I am interested to know whether you find this disgusting, I actually think it is cool. (I have changed nomenclature a bit to analyse products in general.)

Laundry list:

• refactor code parts which deal with pattern lists
• check whether there are simpler, more direct ways
• rename selectors, to avoid the necessity of disambiguation
• codegen
• suppress actor matches (for now); checking mode: done, inferring: not possible?
• suppress text, array matches this works now in frontend, codegen: done
• fix sort inference for shared func (other PR?, later)
• reject matching on mutable fields
• disable repeated pattern fields (dubious feature, no coverage yet); added better error message.

[nomeata]

In coverage.ml I am cutting corners, and reusing the infrastructure of plain products for labelled products, filling non-appearing field labels with wildcard patterns. I am interested to know whether you find this disgusting, I actually think it is cool.

It's a good way to think about it, but it might be cleaner and easier to understand in the future if you’d factor out the common logic a helper (match_product?) and use that for both tuples and objects.

[ggreif]

It's a good way to think about it, but it might be cleaner and easier to understand in the future if you’d factor out the common logic a helper (match_product?) and use that for both tuples and objects.

@nomeata I have now Product in the coverage checker and both TupP and ObjP dispatch to that.

[nomeata]

Do we want to allow record pattern pattern-matching syntax on Actors (rather than just local and shared objects)?

[nomeata]

This only works for real objects, not actors. Better assert the sort somewhere here.

[ggreif]

That would be easy if the ObjP had a sort enclosed. Also there is no type of scrutinee around, so it is hard to get hold of the sort. On the expression side we have ActorDotE and DotE, so it is easier to distinguish. Might be necessary to have these with object patterns too.

Let's look for a place where ActorDotE is created and see whether we can suppress ObjP for actor scrutinees.