Re: LYNX-DEV Control-Z and SIGSTOP mysteries

[Kari E. Hurtta]

Bela Lubkin:
> Klaus Weide wrote:
> 
> > Ok, that encouraged me to look at the suspend issues more closely.
> > I am sure some unix guru (from the east or west coast or anywhere else)
> > would be more qualified...
> > 
> > Here's one problem:  do something that invoke a child process which
> > can be suspended with ^Z.  For example, I have a PRINTER: which displays
> > the file with a shell command which invokes `less'.  While in less,
> > I can suspend with ^Z.  When I fg, both processes come alive (less and
> > lynx) and compete for the display and keyboard, which is a bit confusing
> > to say the least...  This happens with curses, ncurses, and slang.
> > 
> > I am experimenting with a LYSystem() wrapper to be used instead of
> > system(), which changes signal(SIGTSTP,...) before calling system().
> > This is of course only for unix - there already is something similar
> > for VMS.  But there should be a better way than this - how do other
> > programs which use system() deal with it?
> 
> Unix gurus from any coast agree: system() does not give the programmer
> enough control, and should only be used in one-off or prototype
> programs.  For real use you should construct your own subprocess
> handling code out of the underlying fork(), exec(), wait() functions.
> One system() problem which is especially relevant to Lynx is that its
> underlying implementation varies from one Unix to another.  These
> variations will show up particularly in areas like the signal handling
> anomalies you're experiencing.
> 
> Any time Lynx is waiting on a subprocess, I believe it should be
> completely ignoring job control signals (SIGSTOP, SIGTSTP, SIGTTIN,
> SIGTTOU, SIGCONT).  It should neither stop nor continue; it should

No. If it does not stop and you are on running some subprocess:
        user hits Ctrl-Z
                terminal driver deriver SIGTSTP foreground process group
                subprocess and lynx gets SIGTSTP
                subprocess stops
                        lynx continues waits to end subprocess
                        > shell does not see that lynx stops
                          it is still waiting that lynx either
                                stops or terminates
                -> there is no process left which takes user input!
                > That is hang from user point of view

> tenaciously wait for its child, nothing else.  (The rules are very
> different for asynchronous subprocesses; but I can't think of any cases
> where Lynx uses one.  Mailcap items such as "xv %s &" in mailcap do not
> count: as far as Lynx is concerned, they are synchronous; they detach
> *themselves* and live on after Lynx has finished launching them.)

;
; To UNSUBSCRIBE:  Send a mail message to address@hidden
;                  with "unsubscribe lynx-dev" (without the
;                  quotation marks) on a line by itself.
;