When replacing Mines' End or top of Sokoban, the achievement for
finding the special prize there is reset. But the code to do so was
accidentally inside the monster processing loop and got repeated for
each monster on the old instance of the level (essentially a no-op
after the first one though). That code had been updated more than
once without noticing. Move it out of the loop.
Relax the count limit from 255 to ROWNO*(COLNO-1) so that it can
be big enough to fill an entire level yet remain small enough to
not churn away seemingly forever if an absurd amount is specified
for 'random' or for a class rather than a type. (By-type already
gives up as soon as failure occurs, so is implicitly limited to a
count matching the available space on the level.) Also impose the
same limit on 'count ^G monster' as '^G count monster'.
If a special level explicitly requests eg. a statue with a genocided
monster class, allow generating it.
Rationale is that those objects were generated before the monsters
became extinct. Also fixes a lua error.
Get rid of a couple of variables that were file scope and then
incorporated into 'g'. This should prevent the situation where
attacking a shade gave bogus feedback about glorkum although I
never did reproduce that.
This eliminates g.otmp and g.dieroll but leaves a couple of others.
g.vis really should go away....
I couldn't reproduce the problem; it appears to depend upon whether
the file-scope variable 'otmp' has a stale value, and that might
happen after a monster has tried to steal mon's saddle. However,
the code pointed out in the report is clearly wrong. This prevents
feedback of "glorkum" (with plural verb since quantity of 0 isn't 1),
but the potential stale value hasn't been dealt with.
If the stair or ladder x coord is 0, it doesn't exist.
Flipping it, caused "u_on_newpos: trying to place hero off map <80,0>"
when coming back up from the valley into the castle, and the
castle was flipped.
Fixes#331
Reported 3.5 years ago. Specifying a count for pickup to pick up a
subset of a stack was processed after scrolls of scare monster were
handled, so a whole stack of those got picked up or crumbled to dust
whether you gave a count or not. Also, if you were too encumbered to
pick up the full stack, they still all crumbled or all changed state
to crumble next time, then for the latter case you picked up as big a
subset as you could handle.
If either rumors.tru or rumors.fal was empty when makedefs made
'rumors', init_rumors() will set true_rumor_size to -1 to indicate
that rumors aren't available. It also closes the input file, and
then #wizrumorcheck closed that again, triggering a crash in the
dlb code.
Fortune cookies and oracles work ok (just not very interesting)
when rumors aren't available. Only the check command had trouble
with that.
The traceback points directly to the problem: divide by 0 happens
if the 'bogusmon' file only contains the "do not edit" line, which
would happen if 'bogusmon.txt' is empty. makedefs probably ought to
complain about that.
There is now one hardcoded bogus monster to fall back to: 'bogon'.
Random tombstone epitaphs report divide by 0 if their text source is
empty, but it is done by rn2() rather than rn2_for_display_rng() so
is just a warning for pre-release code. It would crash for release
version though.
I tried placing an empty engravings file and expected similar results
but didn't see any response. Not sure what that means.
After the fix, empty epitaph file yields blank result so graves that
want a random epitaph won't have any epitaph.
Fixes#302
While polymorphed and underwater, an eel bite killed the hero who
rehumaized and crawled out of the water, then the eel continued with
its second attack and "wrapped itself around you" even though no
longer adjacent. That's a long reach....
| ...@.
| .;}..
| .}}..
Make any additional attacks silently miss if hero changes location
during the attack sequence of a monster who has pinpointed the hero.
If 'sinking-into-lava' is disabled as a displayed status condition
but general 'trapped' is enabled, then display 'trapped' when in lava.
Similarly, if 'grabbed-by-eel' is disabled but more general 'held' is
enabled, display 'held' when grabbed.
Try to make the tracebacks generated via PANICTRACE's libc method
be more readable. There tends to be a lot of blank space in the
middle of lines, depending on the length of the program's or
libraries' file names, which can make lines wrap and the whole
thing sometimes be harder to make out. Narrow it by squeezing out
some spaces from each line before writing to reduce the chance of
line wrapping. That isn't guaranteed to make things ledgible but
seems to help quite a bit. Also, force the line number inserted
by nethack to be two digits so when spanning from line 9 to line 10
it doesn't cause the field positions to shift by a column. (Using
'#panic' doesn't have enough stack frames to illustrate that.)
I think the original libc formatting was designed for the address
column to hold a 32-bit value ('0x' prefix and 8 hex digits), and
eventually extending that to handle 64 bits ('0x' prefix and 16 hex
digits) made things wider than intended. But the gradual increase
in the length of function names we use is also a factor.
The old parseoptions() would get a FALSE return from parse_role_opts() and
then exit FALSE.
The new parseoptions() was printing an error message due to the FALSE return
value, and then exiting FALSE.
Have it behave the original way following parse_role_opts().
Role selection is insanely complex. I had to use a debugger to force
the relevant routine to be executed.
The analysis was correct: it could use rn2(14) to pick a role (valid
values 0 through 12) and randomly getting 13 would lead to a crash.
The terminating element of roles[] passes all the ok_role(), ok_race(),
etc tests. Explicitly exclude that element when collecting the roles
to choose from.
