2023 06 27 meeting

2023 06 27 Meeting

Current challenges

correctness of loading types from types database

There are a couple of predictable kinds of errors:

• types defined in the test suite (labeled with the Main module)
• derived from previous (e.g. Vector{Animal} if Animal comes from tests)

state space blow out if several Any-params are around

Some sort of streaming could help here but I failed to implement anything in reasonable time.

Relational.AI code: we seem to miss ~25% methods

In discussions with Nathan how to fix it. Jan suggests it may be not a big deal.

Going forward

Get database from RAICode

This entails running github.com:prl-julia/julia-type-stability on their package, which is not trivial.

Exhaustive enumeration form types DB is only first step

Some sort of bias would be preferable. E.g. bias on the ground of what we saw in the tests. It’s still not clear how to bias exactly.