Andreas Rumpf

How does this affect compile times?

Sure but it's easy to measure: bootstrap time with type bound ops disabled: 5.2s, with enabled: 5.3s (for example!)

This has nothing to do with `defer` as the following program shows the same wrong behavior: ```nim import std/asyncdispatch proc doStuff {.async.} = try: for i in 0..

Workaround: Avoid complex finally sections, move logic to an async helper proc: ```nim import std/asyncdispatch proc action() {.async.} = for i in 0..

> Or have I missed something, and the capability already exists? There just isn't a need for it at all. Instead of "atlas, use path/to/package/in/development" you would simply patch the...

Has been implemented.

Apart from my code review remarks, seems fine.

Nim's `cast` is not C's cast, make the codegen use a memcpy.

But it seems to already use a `union` so I don't get the problem.

Good points on how `union` wouldn't solve the problem but I'm still of the opinion the codegen should be fixed to make the code valid.