#3595: problems suspending mutt when called from shell script
---------------------------+----------------------
Reporter: blmassingill | Owner: mutt-dev
Type: defect | Status: new
Priority: minor | Milestone:
Component: mutt | Version: 1.5.21
Resolution: | Keywords:
---------------------------+----------------------
Comment (by blmassingill):
Replying to [comment:1 zsdzsd]:
> Are you sure it works in 1.4.2? I tried compiling 1.4.2.3 and I still
> get the same problem.
Yes -- but what I'm using is a slightly earlier version. First line of
output of "mutt -v" is
Mutt 1.4.2.1i (2004-02-12)
>
> The problem, which shows up in programs such as grepm (I add that
> mainly for people searching the bug list), may be related to
> the following code in signal.c:
> {{{
> RETSIGTYPE sighandler (int sig)
> {
> int save_errno = errno;
> struct sigaction act;
>
> switch (sig)
> {
> case SIGTSTP: /* user requested a suspend */
> if (!option (OPTSUSPEND))
> break;
> IsEndwin = isendwin ();
> curs_set (1);
> if (!IsEndwin)
> endwin ();
> kill (0, SIGSTOP);
> }}}
> This codes assumes that a SIGSTOP is just as good as a SIGTSTP, but
> (perhaps) when mutt is running inside a subshell (a la) grepm, the two
> signals are not handled analogously. Or perhaps the issue is as I
> speculate below (*).
>
>
> I tried making the following change to signal.c in 1.4.2.3:
> {{{
> RETSIGTYPE sighandler (int sig)
> {
> int save_errno = errno;
> sigset_t mask; /* NEW */
> struct sigaction act;
>
> switch (sig)
> {
> case SIGTSTP: /* user requested a suspend */
> if (!option (OPTSUSPEND))
> break;
> IsEndwin = isendwin ();
> curs_set (1);
> if (!IsEndwin)
> endwin ();
> /* OLD kill (0, SIGSTOP); */
> /* REST OF THIS TO "HERE" IS NEW: */
> sigemptyset(&mask);
> sigaddset(&mask, SIGTSTP);
> sigprocmask(SIG_UNBLOCK, &mask, NULL);
> signal(SIGTSTP, SIG_DFL); /* reset disposition to default */
> sigemptyset(&act.sa_mask);
> sigaddset(&act.sa_mask, SIGTSTP);
> act.sa_flags = 0;
> act.sa_handler = sighandler;
> kill(0, SIGTSTP);
> /* wait to be CONTinued... */
> sigprocmask(SIG_BLOCK, &mask, NULL); /* DO I WANT THIS?? */
> sigaction(SIGTSTP, &act, NULL);
> /* HERE */
> }}}
>
> and the situation is much improved, but (annoyingly) there is the
> occasional situation in which the problem occurs. (See (*) below.)
>
>
> I tried replacing
> {{{
> kill(0, SIGTSTP);
> }}}
> with
> {{{
> kill(getpid(), SIGTSTP);
> }}}
> and that didn't work any better. (The thought being that the subshell
which called mutt has
> already received SIGSTSP, so it doesn't need another signal.)
>
> (*) I speculate that there is some race condition between when mutt
> and the subshell process the SIGTSTP. But even if this is true, I'm
> not immediately sure how to deal with this.
>
> Are there any signal handling / process groups / terminal emulator
> wizards here who could comment?
>
--
Ticket URL: <http://dev.mutt.org/trac/ticket/3595#comment:3>
Mutt <http://www.mutt.org/>
The Mutt mail user agent
