goodboy

Tried the `pexpect.popen_pawn.PopenSpawn` approach and it seems [our run shows that the patterns we're trying to match are no good?](https://github.com/goodboy/tractor/runs/1250912871?check_suite_focus=true). Some other options for someone on a windows system to...

Commit above is where someone can take over.

Ok, so figured out what I think is the main bug to do with the #129 stuff. - run a program with a `await tractor.breakpoint()` in it - execute the...

Alright dug slightly into using `deliver_cancel`: - it's only available when using [`run_process()`](https://github.com/python-trio/trio/blob/master/trio/_subprocess.py#L426) (which we're not). - if we wanted to implement something similar ourselves [here are the impls](https://github.com/python-trio/trio/blob/master/trio/_subprocess.py#L401)

Disabled `SIGINT` in children and all other machinery seems to still work in the test suite; so I guess that's good? It's still not solving the main issue with debugging...

I *think* the issue has been resolved by [allocating a separate cancel scope](https://github.com/goodboy/tractor/blob/multiproc_debug/tractor/_debug.py#L191) for the service nursery scheduled in [`wait_for_parent_stdin_hijack`](https://github.com/goodboy/tractor/blob/multiproc_debug/tractor/_debug.py#L187). By simply [cancelling this scope in `Actor.cancel()`](https://github.com/goodboy/tractor/blob/multiproc_debug/tractor/_actor.py#L838) deadlocks are avoided...

Hmm, the number of failures on that makes me think maybe we have something simple, maybe at syntax or import level that's broken. Not sure what offhand since the runs...

@guilledk for fun you should bump this to latest master to see if the deterministic cancel stuff improves anything. I got my doubts though.

think this is done now yah?

Oof 😂