Re: NPTL work committed to jimb-rda-nptl-branch

[Jim Blandy <jimb at redhat dot com>]

[dropped list CC]

Daniel Jacobowitz <drow@false.org> writes:
> By "event" I meant an exit from waitpid.  If waitpid is really letting
> an attached thread disappear without reporting exit status to the
> debugger, then something is wrong.

No kidding.  That's kind of a relief, actually.

I'm pretty I was accurately seeing what was going on.  Aside from the
code that forks the initial process, there is only one call to waitpid
being used for threaded Linux programs in all of RDA now: the one in
lwp-pool.c:wait_and_handle.  And that one's got a tracing fprintf
directly after it.  And after I'd made it drop LWP's that we'd gotten
TD_DEATH events for and that had completed their notifications, there
are consistency checks in lwp-pool.c:hash_find_known that would have
complained if we'd gotten wait statuses for them anyway.

For posterity, here is the kernel that does this:

[jimb@alligator testsuite]$ uname -r
2.4.22-1.2115.nptl

However, that isn't the kernel we'll actually be delivering on.  The
vanishing thread hair should be pretty easy to ditch if the real
kernel works better.  Thanks for the heads up!