Re: Why `make` could hang

[Paul Smith]

On Thu, 2021-09-30 at 18:56 +0300, uladzimir.bely@promwad.com wrote:
> We haven't stop the process yet, so the hanging `make` process 20421
> is still running on the server. So, we could attach again with gdb to
> it and extract any useful information that could help to understand
> what had happened.

I won't have time to look at this until the weekend at the earliest.

However, it seems that make believes that one of the jobs it started is
still running, since it goes through the:

54           DBF (DB_VERBOSE, _("Still updating file '%s'.\n"));

over and over.

You would probably find some interesting information by examining (via
the debugger) _which_ file this is; when you hit this line, print out
the value if file->name, or even the entire *file structure.  Do that a
few times in a row and see if the file changes or if it's always the
same one.

You should probably also take a gcore of the running process before you
kill it; in some ways it's simpler to debug a running process but
having a coredump of the process is the next best thing.

FYI, I haven't seen anything like this.  However I know there are
issues with make's handling of fatal signals such as SIGINT, etc.
especially if >1 of them is received by the same make process.  That's
a difficult problem to solve; I have a kind of sample solution that
doesn't quite work but have not had a chance to complete it.