Re: Failure of resizeterm() causes subsequent call with stdscr to crash

[Thomas Dickey]

On Wed, Sep 25, 2024 at 10:35:24AM -0400, Bill Gray wrote:
> On 9/25/24 08:29, Peter Bierma wrote:> This was reported downstream as a bug
> in Python's curses module. See the
> > issue there: https://github.com/python/cpython/issues/120378
> > <https://github.com/python/cpython/issues/120378>
> > 
> > It seems that calling resizeterm() with some ridiculous resolution
> > (35000x1 in the CPython issue) causes stdscr to break (i.e., a crash). A
> > small reproducer:
> > 
> >      #include <curses.h> // or ncurses.h
> > 
> >      int
> >      main(void)
> >      {
> >          initscr();
> >          if (resizeterm(35000, 1) < 0) {
> >              puts("resizeterm() failed, trying to reset");
> >          }
> >          wrefresh(stdscr); // Segfault!
> >          return 0;
> >      }
> 
>    The limit appears to be 32767.  32768 (which,  as a signed short, is
> really -32768) fails.  Similarly,
> 
> win = newwin( 32767, 0, 0, 0);
> 
>    returns a non-NULL pointer.  Go up by one,  and win == NULL.

for newwin, I have this crude check for overflow:

        static bool
        dimension_limit(int value)
        {
            NCURSES_SIZE_T test = (NCURSES_SIZE_T) value;
            return (test == value && value > 0);
        }

though a hard limit would suffice (agreeing that terminals with 32k lines
and columns are hard to come by - but pads should be less limited).

>    I can't come up with a reason why you'd have such a large window. But
> then again,  I also can't come up with a reason why it should fail.

overlooked :-(

Unless specially configured, ncurses's using short internally -- as well as in
the WINDOW struct, because that was from the original curses implementation,
based on the terminfo format.

I made the sizes configurable to an int long ago (2000), but no packager
followed up on that, and changing it would have changed ABI.  Later (2007),
I provided for making WINDOW opaque (which would have allowed changing its
internals), and it got little attention as well.

Changing the default to opaque in ncurses 6.5 has limited effect - I recall
seeing a packager comment that they would override that default.

Now I could have added a limit-check to wresize (looks overdue), but
it's very rare for programmers to do this, and none of those had
reported breakage -- as I recall it, all of those had tried creating
a too-large window with newwin.

-- 
Thomas E. Dickey <dickey@invisible-island.net>
https://invisible-island.net

signature.asc
Description: PGP signature