Re: Opportunistic GC

[Pip Cet]

On Mon, Mar 8, 2021 at 5:13 PM Eli Zaretskii <eliz@gnu.org> wrote:
> > From: Stefan Monnier <monnier@iro.umontreal.ca>
> > Cc: Pip Cet <pipcet@gmail.com>,  eliz@gnu.org,  emacs-devel@gnu.org
> > Date: Mon, 08 Mar 2021 11:44:05 -0500
> >
> > > I also expect fork will work, I'm more wondering about interaction with
> > > sockets file descriptors or anything else in our codebase and/or in
> > > libraries we have loaded.
> >
> > I suspect we wouldn't want to use `fork` literally, but rather something
> > like Linux's `clone` so as to have more control about what is
> > really shared.  Basically, we just want to spawn a thread, except it runs
> > in a snapshot of the heap.
>
> There are other potential complications with this, related to
> threads.  Posix says 'fork' only copies a single thread, the one that
> called it, but what if that single thread is not the main thread?

Why would that be a problem? I think everything would work just fine
in that case, but maybe I'm missing a problem that should be obvious
to me.

> In
> Emacs nowadays a non-main Lisp thread can trigger GC.  What happens to
> the mutexes we use in the threads machinery when we fork like that?

In the child, they're irrelevant, because we'll never return from the
function that triggered GC, or allow signal handlers to run; in the
parent, they remain valid. Everything should be fine...

Pip