test: reproduce dune exec -w crash with pkg management

[anmonteiro]

reproduces the crash in #10959

[Alizter]

  let update_build_progress_exn ~f =

is initialising when this happens so I need to find who is updating the build progress this early.

[Alizter]

For some reason this stems from the let* binding of env in run_eagar_watch. Not sure how it is being triggered yet however.

[anmonteiro]

I just pushed a commit that delays evaluating Env.t Memo.t until get_path_and_build_if_necessary

[Alizter]

2 seems long, what about 0.1? The other exec-watch tests appear to pick something like that. Although killing processes this was seems very fragile, there ought to be a more principled way to test these. Not this PRs problem however.

[Alizter]

The watching/ tests use some better helpers. Up to you if you want to improve things. Also I would move this from pkg to exec-watch.

[anmonteiro]

could you point me to a relevant example?

[Alizter]

They have these start_dune stop_dune helpers from a shared script (see the dune file for where it is). A test example is: https://github.com/ocaml/dune/blob/main/test/blackbox-tests/test-cases/watching/copy-rules.t

The logic in the script handles stopping the watch mode build and the output. Presumably something like that could be adapted for exec. But if this is too much work that is also fine, since the other exec-watch/ tests are written this way.

[Alizter]

Perhaps call this get_env_and_build_if_necessary.

[anmonteiro]

before addressing the outstanding code review items, I'd like some confirmation that the fix is in the right direction, or even correct.