[interpreter] Fix symbolic local resolution with typeuse

[rossberg]

Fixes #1212.

There is a complication here. In order to maintain the malformed/invalid distinction properly, we need to make sure that type definitions are only looked up during parsing when actually needed for computing symbolic indices. Otherwise, we'd turn some syntactically valid, fully desugared programs like the following:

(module (func (type 0) (local i32)))

into triggering a syntax error instead of a validation error (due to an undefined type index). My first naive attempt at a fix actually did that and regressed one test expectation from assert_invalid to assert_malformed, which e.g. means that it could not be converted to binary anymore.

Unfortunately, we generally don't know whether the type will be needed until we have seen the full function body and know whether it contains any symbolic local (consider let in particular).

To address that, the parser now uses a work list to defer looking up the function type (and consecutive index shifts for anonymous binders) until actually needed for computing symbolic indices. In a fully desugared program it is never triggered.

This is a bit ugly, but most implementations probably don't need to care about this distinction. But for the reference interpreter it seems relevant.

[binji]

Is this an unrelated change?

[rossberg]

Ha, yeah, I didn't notice I still had that in my tree. One could argue it's somewhat related to typeuse, though, since it expands type uses in block types. :)

[binji]

Strange, the IPR bot says that you haven't joined the community group. Did your email address change?

[rossberg]

@binji, yeah, I have had this problem for a while, ever since I merged my two GH accounts. I couldn't figure out how to fix it. It tells me I need to join the CG (with some other account?), but when I try by clicking the link it tells me that I've already joined.