Michael J. Sullivan

Use generic functions as `PyGetSet`s instead of generating so much code

1

We currently generate a *lot* of code to provide python level getters and setters for native class attributes. I think we could avoid generating almost all of it and instead...

There's a lot of boilerplate between all of the many different places that we generate different kinds of glue methods, and I bet it could be cleaned up

The error handling code for unboxing tuples was busted and would cause crashes when it failed. As an expedient during bootstrapping, where this was making it hard to debug, I...

Avoid pointless boxing and unboxing

3

We sometimes generate pointless unboxing and boxing. For example: ``` [case testListCopy] from typing import List def copy(a: List[int], b: List[int]) -> None: a[0] = b[0] [out] def copy(a, b):...

Add Sequence/Mapping like types that are specialized to List/Dict?

1

It is often nice to use Sequence and Mapping types for function arguments and the like because they are covariant and because it indicates that they will not be mutated....

START TRANSACTION and COMMIT don't seem to ever get serviced by the query cache. This doesn't seem like it should be necessary. Those are fast to compile, but it's still...

The core thing here is that when we move a compiled IR expression into a FOR loop, we need to adjust the scope tree so that it is inside the...

https://github.com/edgedb/edgedb/pull/5949 did the bulk of the work for extension-defined configs and nested database configs, but there are some loose ends: - [x] Support extension-defined SESSION configs - [x] Support redacting...

Collect sharded CI test failures into one list

1

Bonus points if that one list gets rendered nicely by github somehow (I think there are existing things that integrate with junit xml output??) but that isn't necessary for the...