handle_exit called with missing starter causes a crash

[tomalexhughes]

Creating an issue for this problem discussed in the Discord.

It's possible for the handle_exit function in the supervisor module to be called with state that is missing the starter.exec function.

This causes the following error due to this let assert

Runtime terminating during boot ({#{function=><<"handle_exit">>,line=>303,message=><<"Assertion pattern match failed">>,module=><<"gleam/otp/supervisor">>,value=>none,gleam_error=>let_assert},[{'gleam@otp@supervisor',handle_exit,2,[{file,"/[....]/build/dev/erlang/gleam_otp/_gleam_artefacts/gleam@[email protected]"},{line,235}]},{'gleam@otp@actor',loop,1,[{file,"/[....]/build/dev/erlang/gleam_otp/_gleam_artefacts/gleam@[email protected]"},{line,150}]}]})

Unfortunately I do not have a way to reproduce at the moment and I'm unsure as to the exact details that cause the none value.

[lpil]

Thank you.

[ghivert]

FYI, I have a regular error on Gloogle due to this error, because I'm exiting normally some children. That triggers server to restart. It's not a big deal at the moment, but it's bothering enough I'd like to fix it. Are you interested in some PR @lpil ? In case you are, do you have some opinions or some hints to get started on the subject?

[lpil]

PRs welcome but I've not investigated this.

[ghivert]

I think I understood why it's happening: I'm using a supervisor to start fire & forget tasks. In Gloogle, when I want to extract signatures, I'm spawning a process to do it in an async way, but I'm starting it within a supervisor. So when it shutdown, there's no other children to start, because it's supposed to stop. Am I right to say that every supervisor children processes should be thought as long running processes, that restart automatically?

[lpil]

Am I right to say that every supervisor children processes should be thought as long running processes, that restart automatically?

No, it would depend on what supervisor you are using and how it has been configured.