This kind of cross-process tracing is so difficult! It reminds me of a “stray signal” bug that persisted for years in the https://www.oilshell.org/ test suite

I use strace, and we’ve also started using bpftrace, thanks to a contribution

I also learned there is this crazy netlink interface to monitor processes on Linux, I haven’t tried it yet - https://fabiensanglard.net/st/index.html

FWIW I ended up a little confused by the bug - a summary at the end of the post may have helped. I saw the patch

https://github.com/dlang/phobos/pull/10586/files

But it’s not clear why the test was like that in the first place. The commit description doesn’t really explain it either

Also I wasn’t clear on the perl-init vs. dumb-init part