Re: BUG: Blocking read after SIGWINCH with nodelay set.

[Thomas Dickey]

On Tue, Aug 20, 2002 at 07:25:01PM -0400, Sam Varshavchik wrote:
> 
> I'm seeing this behavior with the Linux 2.4 kernel:

...but what version of ncurses?  I had introduced a related bug, but fixed
last fall:

20011229
        + restore special case from 20010922 changes to omit SA_RESTART when
          setting up SIGWINCH handler, which is needed to allow wgetch() to be
          interrupted by that signal.

> Even when nodelay is enabled, ncurses still does a blocking read call
> after it receives a SIGWINCH.  What I could figure out was that the first
> call to wgetch() after a SIGWINCH eventually makes its way to kgetch():
> 
>     for (;;) {
>         if (!raw_key_in_fifo()) {
>             if (fifo_push() == ERR) {
>                 peek = head;    /* the keys stay uninterpreted */
>                 return ERR;
> 
> raw_key_in_fifo() will return false, and kgetch() will then end up
> blocking by calling read() in fifo_push(), even though the nodelay mode is
> set.
> 
> It looks like at this point the fifo buffer contains a single KEY_RESIZE, 
> for which fifo_push() returns false.
> 
> -- 
> Sam
> 
> 
> 
> _______________________________________________
> Bug-ncurses mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/bug-ncurses

-- 
Thomas E. Dickey <address@hidden>
http://invisible-island.net
ftp://invisible-island.net