Nethack crashes when you are riding a flying monster over a
pool/moat and some engulfing monster plucks you off your saddle.
After falling into the water you'll get the normal message
sequence (sink like rock ... phew, that was close; you also get
chance to teleport if you can). After the last message the game
will crash with a segmentation fault.
- reproducible; null pointer dereference in swallowed()
The crash results because dismount_steed() calls float_down(), which
calls drown(), which calls teleds(), which clears u.ustuck. So when
gulpmu calls swallowed after dismount_steed(), this line attempts to
derefernce a null pointer:
swallower = monsndx(u.ustuck->data);
This patch bypasses the float_down() in dismount steed() altogether.
That routine is meant to return the hero to the floor, and that
isn't appropriate if a purple worm just plucked you off the steed
anyway.
While this fixes the crash, a problem still exists. The
way swallowing works, the swallowing monster's location
switches to that of the hero. Since that location is
over water, the purple worm ends up drowning almost
immediately after you are swallowed, while you are
swallowed. The purple worm's death is not revealed
to you since the "The purple worm drowns." message is
conditional. This patch also adds a message when
the purple worm dies, but should this guaranteed
drowing take place?
"Force the mount to succeed?" doesn't necessarily do so;
it's worse than useless to suppress the message which indicates
why a mount attempt is failing when a failure occurs.
From the newsgroup:
> Apparent bug or weird-seeming feature:
> If you try to mount a horse that is already in a pit,
> you get the message "You lead the poor <foo> into a pit!"
I checked into fixing this properly (you mount the steed and
then shared its trapped status without actively retriggering the
trap in the process), but the trap code is too convoluted to do
that cleanly. So this "fix" just makes mounting attempts fail if
the steed is trapped.
(This diff is mostly whitespace cleanup.)
Fixing some iron ball/teleds stuff:
-- If the player can pass through walls, ignore all checks for walls, or else
things will behave weirdly.
-- Instead of using the kludge "if the distance is >2 it must be a teleport",
pass a parameter indicating whether they crawled or teleported onto the new
space. This fixes a special case, where the player moved one space and the
ball didn't move, but the chain moved through solid rock. This is acceptable
if teleporting and unacceptable if dragging.
This also required some rearrangement of teleds() so that u.ux,u.uy
are set after placing the ball, not before. I'm still not sure the pit
filling line is in the right place; anyone know?
-- add some comments so I can look at the code in a month and still know what
I did.
Most of this patch is just adding the new parameter.
Add absent prototypes to some core routines.
Also add some port function() to function(void) in some win32 routines.
Also updates the Borland C Makefile for win32.
Patch from <Someone> to the list on 10/01/2001. The changes cause riding a steed
into water to result in a dismount, and, if the steed can't survive in
water, the death of the steed, with the usual pet-death penalties.
