setpgid() fails with EPERM if pgid is exited but not reaped (it should succeed)

[ridiculousfish]

This is a reduced test case for some behavior that we're seeing in the fish shell. One issue in fish, previous WSL issue which was considered fixed but is not.

fish handles pipeline jobs by making the first process the group leader for all processes in the job, via setpgid. However if the group leader has exited (but not yet been reaped), the setpgid() call fails with EPERM on WSL, where AFAICT it ought to succeed (it does on macOS and Linux).

Tested on Win10 build 16299.rs3_release.170928-1534.

To reproduce: compile and run run.c as cc run.c; ./a.out On WSL this will output some errors and hang. On macOS and Linux it succeeds.

You can simulate fish's ugly workaround by running ./a.out keepalive. This keeps the parent process alive until the child has joined it, which prevents the error.

[therealkenc]

I call this the Rick Grimes problem. Like #1353 (message). From ps on Real Linux:

11006 pts/0    S+     0:00 ./a.out
11007 pts/0    Z      0:00 [a.out] <defunct>

On WSL, zombies die right away.

[ridiculousfish]

[Brian-Perkins]

This should work better starting in Insider Build 17713