This takes care of a lot of the leaked memory in the curses interface.
It still needs to free memory allocated for status fields when the
status window is destroyed at game end; likewise for message history
when the message window is destroyed.
Support <delete> (aka <rubout>) during getline(). It doesn't actually
honor the current erase_char value set up for the terminal, just
treats DEL the same as ^H. (The previous lack of support had nothing
to do with terminfo specifying ^H; the handling is hard-coded.)
tty treats escape while there's already some input as kill_char (erase
the input but get more from scratch) and returns ESC if there isn't.
curses was doing the first half but not the second, so not providing
any way to communicate "cancel" back to the core. Fix is simple.
Other getline() bug fixes:
1] there was a wprintw("%*something") which was passing the value from
strlen (type 'size_t') to the "%*" argument (type 'int'). That's
always wrong (size_t is guaranteed to be unsigned) and could be severe
(if size_t is different width than int--as on current OSX systems--
depending upon the internals of argument passing).
2] strncpy() only supplies a terminating '\0' if the input is shorter
than the number of characters specified.
A lot of reformatting is warranted but I only did the getline routine
(manually, so might have missed stuff).
Three or four instances of one simple memory leak. Allocating a union
'anything' to pass to add_menu(), then not doing anything with it. The
value gets copied so there's no reason for the original to stick around.
[There are still lots of other memory leaks.]
There was no provision for malloc() potentially returning Null and it
wasn't integrated with nethack's MONITOR_HEAP. 'heaputil' shows that
the curses interface is leaking like a sieve. If some things are
actually being allocated separately and then freed from within curses,
those need to be thoroughly documented and maybe switched back to
malloc().
The curses interface already has a hack to keep 'Count: 12', 'Count:
123' intermediate multi-digit counts out of its message recall history
for ^P, but it was flushing real messages when getpos()'s 'autodescribe'
reported what the cursor moved over. Overload the count hack to support
putstr(WIN_MESSAGE, ATR_NOHISTORY, text)
(which is what custompline(SUPPRESS_HISTORY, ...) eventually calls).
The conditional logic for when to create the 'count_window' was pretty
convoluted. This simplification has the same semantics but I don't
have PDCURSES to actually verify that.
I don't see the #H8124 bug getting fixed any time soon, so add a
comment about it since the relevant place is unambiguous.
No change in code except removing a couple instances of 'register'.
Noticed while investigating the issue with DECgraphics characters in
msg_window:full/combination/reverse output for tty which got fixed
by the previous commit. There was a discrepancy in DUMPLOG because
the pager code bypasses pline() in order to use putmixed(). tty
puts strings from the latter into ^P recall history (although they'll
only render correctly if nothing after the first character needs
special handling), but nothing was putting that same info into
DUMPLOG. This fix is pretty clumsy but eliminates the discrepancy.
Noticed while testing the history suppression: if you have DECgraphics
enabled and look at a graphics character on the map, the topline shows
x description of x
where 'x' is displayed as it appears on the map (line drawing char).
^P for msg_window:single knows about that and reproduces the effect if
you recall such a line. But msg_window:full/combination/reverse didn't
know about that and dumped it as-is into text output, ending up with a
strange 8-bit character for 'x' instead of the line drawing one.
I think other rendering schemes will be unaffected by this. It's just
duplicating what is done for msg_window:single.
Extend 'putstr(WIN_MESSAGE, attribute, string)'s attribute so that
'custompline(SUPPRESS_HISTORY, ...)' can work with ^P's message
history like DUMPLOG history, in order to keep autodescribe feedback
and intermediate prompts for multi-digit count ('Count: 12', 'Count:
123') prompts out of recall history. The old autodescribe behavior
could easily push all real messages out of the recall buffer when
moving the cursor around for getpos, and the count behavior looked
silly for a four or five digit gold count if you set the msg_window
option to 'full' or 'combination' and viewed them all at once.
Other interfaces may want to follow suit, but this doesn't force them
to make any changes. I added a hook for "urgent messages" that might
be rendered in bold or red or some such and/or override the use of
ESC at --More-- from suppressing further messages, but there aren't
any custompline(URGENT_MESSAGE, ...) calls (potentially "You die...",
for instance) to exercise it. Other people have implemented similar
feature it different ways and I'm not sure whether this one is really
the way to go since the core needs to categorize each message that it
deems to be urgent. MSG_TYPE:stop may be sufficent, although MSG_TYPE
matching can entail a lot of regexp execution overhead at run-time.
Fixes#101
If you tell the vault guard your name, drop carried gold, wait one
turn, then pick up the gold again, the guard will move a step away
during the wait. If you teleport away, the guard will seal vault
walls and then park himself on the one-square (so far) temporary
corridor adjacent to the vault wall. Periodically he'll say "Move
along!" and the hero will hear it, regardless of location on the
level. Unless you dig a corridor to rescue him, or one or more of
the vault's walls get breached again, he will never move.
The report emphasized that you could use this to steal the vault's
gold, but it relies on being able to teleport beyond the gaurd's
reach and if you can do that, you might as well do so before the
guard comes. The stranded guard, and him saying "Move along!" when
no longer leading hero out of the vault, are more significant bugs.
Bonus fix: if the game ends and the guard seals up the vault while
the hero is in a spot that gets fixed up (vault wall or temporary
corridor) don't give the "You are encased in the rock" message if
game end was due to death rather than quit.
When ^I was changed to allow picking more than one item to make
its temporary identification become persistent, group accelators got
left out. So to pick all potions, you needed to select them letter
by letter (or via '.' to select everything, then deselect non-potions
letter by letter). Now you can use '!' to select all potions.
Make healing magic which cures blindness also cure deafness. So,
drinking non-cursed potion of healing or any extra healing or full
healing; breathing fumes from blessed potion of healing or non-cursed
potion of extra healing or any potion of full healing; prayer reward
to cure blindness as a minor trouble. (Doesn't affect unicorn horns
which already treat deafness and blindness as two distinct troubles
that are eligible to be cured.)
More of a missing feature than a bug fix, so I listed it in the new
features section of the fixes file.
When farlooking at a leash glyph on a map, the fake object should never
have leashmon set. This happened when a mimic was on the same spot
and was mimicing a corpse or statue, due to leashmon and corpsenm
using the same field.
Lock context wasn't being cleared if it was for a container and that
container got destroyed. Case discovered was forcelock() ->
breakchestlock() -> delobj() (sometimes the container is destroyed
rather than just breaking its lock) followed by #wizmakemap (replace
current level) and maybe_reset_pick() trying to check whether
xlock.box was being carried. But being interrupted, destroying the
container or dropping it down a hole to ship it to another level, then
attempting to resume picking the lock would also find a stale pointer.
Fuzzer feebdack. When turning a monster into a statue, monstone()
builds a linked list of mon->minvent items to put into that statue.
It doesn't use obj_extract_self() to take them off again, leaving
obj->nobj non-Null. Not noticed for the normal case where each item
gets linked into the container's contents, but triggers panic if an
item merges with something already put inside so gets removed.
Suddenly, the dungeon collapses.
dealloc_obj with nobj
[2] 0x01000c4193 panic + 995
[3] 0x0100155427 dealloc_obj + 71
[4] 0x010021d4de obfree + 686
[5] 0x01000f2f92 merged + 834
[6] 0x010015356e add_to_container + 126
[7] 0x01001628ac monstone + 636
I don't know why the petrified monster's mergeable inventory wasn't
already merged while in inventory.
../sys/unix/unixmain.c: In function ‘sys_random_seed’:
../sys/unix/unixmain.c:779:29: error: expected expression before ‘long’
fread(&seed, sizeof long, 1, fptr);
Compiler gave four diagnostics about 'seed' being used uninitialized
if 'no_seed' were false, but two of those were alternate suggestions
for how to suppress them.
The new code provoked several warnings; this fixes one of them.
Moving the declaration of 'rolecount' would have been sufficient,
but I've gone another way.
../win/Qt4/qt4inv.cpp:41:26: error: macro "obj_to_glyph" requires 2 arguments, but only 1 given
glyph=obj_to_glyph(nhobj);
^
../win/Qt4/qt4inv.cpp: In member function ‘void nethack_qt4::NetHackQtInvUsageWindow::drawWorn(QPainter&, obj*, int, int, bool)’:
../win/Qt4/qt4inv.cpp:41:8: error: ‘obj_to_glyph’ was not declared in this scope
glyph=obj_to_glyph(nhobj);
^
Allow sys/share/random.c to be included in the build
always, even if USE_ISAAC64 is defined, by making most
of its contents conditional in that case.
That avoids Makefile tinkering when going back and
forth between USE_ISAAC64 and not during testing.
GCC doesn't set __STDC_VERSION__ for older versions. It is verified to
be set by default on 5.3.0.
Older versions set __INT64_MAX__ though (oldest version tested is
4.5.4).
If compilation with older GCC versions is required, set -std=c99. Even
3.4.6 is supported with this setting.
It's a gameplay-affecting action (it updates the character's
memory of the map), and there's no way to spam it without using
resources, so rn2 is safe.
