Re: [PATCH] [RFC] Move SetConsoleCtrlHandler calls to async thread

[Tom Tromey <tromey@adacore.com>]

Tom> Another problem I thought of is what should happen with these interrupt
Tom> sequences when the user puts the inferior in the background, like with
Tom> "continue &".  Having them stop the inferior seems wrong, the user might
Tom> be trying to interrupt some gdb command instead.

Tom> I don't know how this area is supposed to work.

I looked into how Linux handles this.  Of course, it turns out to be
super complicated.  I did learn some things, though.

First, windows_nat_target doesn't override pass_ctrlc.  However, the one
it inherits calls child_pass_ctrlc, which does:

#ifndef _WIN32
	  kill (inf->pid, SIGINT);
#endif

... oops for this case.

However, this function is pretty interesting because it looks for a
foreground process to interrupt, and if it doesn't find one:

  gdb_assert_not_reached ("no inferior resumed in the fg found");

This leads me to think that if we can integrate windows_nat_target into
the pass_ctrlc codepath, things should work.

I have some patches to attempt this, but of course they don't really
work.  The main mystery right now is that in some scenarios (like the
new-console scenario), when I type C-c, default_quit_handler doesn't
seem to be called at all.

C-break does still work.  (I have this working on my branch by
installing a signal handler for SIGBREAK.)  So I thought maybe SIGINT is
being redirected somewhere, but I couldn't find that.

So once again, not sure what's going on.  Maybe closer though.

Tom