Re: Weird bug / missing feature with gvim interaction

[Felipe Contreras <felipe.contreras@xxxxxxxxx>]

On Mon, Mar 6, 2023 at 1:35 PM Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx> wrote:
>
> On Mon, Mar 6, 2023 at 6:31 AM Felipe Contreras
> <felipe.contreras@xxxxxxxxx> wrote:
> >
> > On Sun, Mar 5, 2023 at 10:28 PM Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx> wrote:
> > >
> > > What's not clear is where the HUP is coming from
> > >
> > > Puzzlingly, using "nohup xdg-open ..." doesn't help.
> >
> > That doesn't help, but "setsid nohup xdg-open ..." does help, both in
> > bash and zsh. But I believe that's what the vim code essentially tried
> > to do.
>
> Sure, but vim is doing it before zsh starts up, whereas when you use
> the setsid command it creates the new process group after zsh runs it.
> E.g., without the setsid command, I believe the thread would be that
> zsh execs nohup which execs xdg-open, all with no forking involved,
> but if you add setsid in front then zsh execs setsid which forks
> before exec of nohup.
>
> Zsh doesn't use setsid(2) and doesn't mess with the process group by
> any other means unless there's a controlling tty to attach to, which
> there presumably is not in this case.
>
> Still doesn't explain the source of the HUP signal, or why nohup fails
> to ignore it ... unless xdg-open is deliberately unblocking it again.

I believe that's what happens because GLib's g_spawn_async does
something like that.

When I hunted the code down I believe it does something like this:

#include <spawn.h>
#include <signal.h>

extern char **environ;

int main(void)
{
char *const argv[] = { "/usr/bin/chromium", "http://google.com";, NULL };
posix_spawnattr_t attr;
sigset_t mask;

posix_spawnattr_init(&attr);
sigemptyset(&mask);
sigaddset(&mask, SIGHUP);

posix_spawnattr_setsigdefault(&attr, &mask);
posix_spawnattr_setflags(&attr, POSIX_SPAWN_SETSIGDEF);

return posix_spawn(NULL, argv[0], NULL, &attr, argv, environ);
}

-- 
Felipe Contreras