Stefan Monnier <monnier@iro.umontreal.ca> writes:

>> In fact, there is already a configuration that would lead to the non
>> deletion of the PID from the list:
>
>> - Call to delete_process,
>> - Interrupted by a signal before inserting  in deleted_processes_list ,
>> - Deletion from  process_alist,
>
> How can this happen?  Are you saying that a signal handler may delete the
> process from process_alist?  Which signal handler can do that?

You're right, it's not its job.

> I do think there's a leak but it goes like this:
>
> - catch the SIGCHLD signal
> - start processing delete-process, which will add the PID to
>   deleted_pid_list even though the signal has already been caught so the pid
>   won't get a chance to be removed.

> I guess we could check the process's status after adding it to
> deleted_pid_list, to catch this case.

I think it would be OK. Here's the new (new new) version of the patch.
The kill_process part is not called if the process is already
terminated.