Suppress close to 400 warnings generated by gcc on the win/X11/*.c code,
most due to -Wwrite-strings which makes string literals implicitly have
the 'const' attribute. (Since modifying a string literal results in
undefined behavior, that is an appropriate check to have enabled, but
it can be troublesome since string literals have type 'char *' and code
that uses them that way is correct provided it avoids modifying them.)
113 warning: initialization discards qualifiers from pointer target type
127 warning: assignment discards qualifiers from pointer target type
29 warning: passing argument discards qualifiers from pointer target type
109 warning: unused parameter
12 warning: comparison between signed and unsigned
The nhStr() hack casts to 'char *', explicitly removing 'const', for
situations where it isn't feasible to make code directly honor const.
The vast marjority of uses are for the second parameter to XtSetArg(),
which is a macro that actually performs an assignment with the second
argument rather than passing it in a function. It takes values like
'XtNtop', which doesn't need to be altered (although in many places I
changed that to nhStr(XtNtop) for uniformity with the surrounding code,
and 'XtNbottom', which does need to have the extra const stripping to
avoid a warning. Go figure.
The nhUse() hack actually uses its argument in a meaningless way if the
code is compiled with FORCE_ARG_USAGE defined. When GCC_WARN is defined,
FORCE_ARG_USAGE will be enabled if it hasn't been already. Example:
/*ARGUSED*/
int foo(arg)
int arg; /* not used */
{
+ nhUse(arg);
return 0;
}
The extra line will expand to ';' when FORCE_ARG_USAGE is not defined
or too
nhUse_dummy += (unsigned)arg;
when it is. I figured direct assignment might lead to a different
warning by some compilers in a situation like
nhUse(arg);
nhUse(otherarg);
where the first assignment would be clobbered by the second, and using
bitwise operations or safer '+= (arg != 0)' would most likely generate
more non-useful code. Some tweaking might turn out to be necessary.
modified files: include/hack.h, src/decl.c, sys/unix/Makefile.src
Groundwork for cleaning up the X11 sources, where gcc with the option
settings specified in the OSX hints file currently generates close to
400 warnings for win/X11/*.c.
lint.h is included by hack.h, and I've moved the debugpline stuff from
the latter to the former to hide it better. (By rights it belongs in
debug.h or something of the sort, but I didn't want to go that far.)
Makefile and project dependencies need to catch up.
nhStr() hides a cast to char *, and is intended to by used on string
literals where it isn't feasible to maintain the 'const' attribute.
(A pernicious problem with X11 code, where the include situation can
become very convoluted, and many, MANY string literals are hidden
behind macros to look like keyword-type tokens.)
nhUse() can be used to force a fake usage on something which triggers
an unused parameter warning. There are a 6 or 8 or 10 places in the
core code where that applies, but so far I have't touched any of them.
There's a tradeoff since it will result in some worthless code being
generated and executed, but is much simpler than tacking on compiler-
specific workarounds like '#pragma unused' or gcc's __attribute__ hack.
Handle !fixinv by forcing gold to have slot '$' all the time; that
particular type of object is 'fixed' regardless of user preference.
Also add a couple of checks for non-'$' gold when selecting from
inventory, just in case the issue of multiple gold stacks reappears.
Lit candles and burning potions of oil can be on the migrating objects
list and get deleted before arrival, so they need the same cleanup as
rotting corpses to prevent obfree from complaining that they're worn.
Thinko fix: sanity checking for owornmask was mis-treating
OBJ_MIGRATING as OBJ_MINVENT of migrating monsters rather than as
unattended objects and would have had problems similar to obfree's
inappropriate impossible check.
Sanity checking for objects worn in invalid slots (amulet worn in
a ring slot and so forth) is extended to items worn by monsters.
Also add a check for wielded coins since the loophole that let them
become wielded has been closed.
Migrating objects overload obj->owornmask with a destination code,
so rot_corpse needs to clear that before deleting corpses. (Buried
objects don't touch owornmask, so rot_organic, which does the actual
object deletion, shouldn't need any similar change.)
The corpses with owornmask 3 that have been observed recently were
slated to arrive on the up stairs, so presumeably fell down the down
stairs of the current level and rotted before the hero went down.
Put plainly, it was the [post-3.4.3] impossible() check which was in
error, not the active game data.
'w$' reported "you can't wield gold" but
'w*$', choosing from inventory and picking gold, let you wield gold.
The old code checked whether gold had been picked before checking
whether '?' or '*' had been picked to request selection from inventory.
This wasn't an issue with 3.4.3's !GOLDINV configuration (but probably
was for anyone who explicitly switched to GOLDINV) because getobj()'s
callers only inserted gold into inventory when they intended to accept
it as a valid choice.
Fix is just to swap two adjacent 'if' blocks in getobj() so that '*'
is processed before the test of whether '$' has been chosen. Most of
the diff is indentation and other minor reformatting.
* Add missing entry for include/tile.h to (top)/Files; also
add new entry for generated file util/tiletxt.c (Unix only);
* Add several missing entries for tile utility programs that
can be built by sys/unix/Makefile.utl to util/.gitignore;
* Update sys/unix/Makefile.utl to build 'tilemap' differently so
that it won't leave behind an unwanted subdirectory tree under OSX:
util/tilemap.dSYM/
util/tilemap.dSYM/Contents/
util/tilemap.dSYM/Contents/Info.plist
util/tilemap.dSYM/Contents/Resources/
util/tilemap.dSYM/Contents/Resources/DWARF/
util/tilemap.dSYM/Contents/Resources/DWARF/tilemap
It now generates util/tiletxt.c on the fly, to be compiled into
tiletxt.o, so that tilemap.c can be compiled in the ordinary
manner and tilemap.o can be kept around for dependency checking.
(Creating real source file win/share/tiletxt.c would be a little
bit cleaner, but it's effectively two lines long so seems silly
to be in the source distribution.) I looked to see whether I
could find a linker or compiler option to suppress that stuff but
failed. I'm sure something of the sort must exist but didn't
pursue it. Someday I might actually learn about how OSX works....
Fix the problem with erroneously detecting wielded or quivered
chained ball which legitimately has more than one mask bit set.
Add an additional check for valid wornmask of an item somehow worn
in an invalid slot (such as an amulet in one of the ring slots).
Only lightly tested; it really needs debugger assistance to force
various invalid situations but I don't know gdb well enough for that.
Fine-tune lycanthropy feedback by combining "you are a werecritter"
and "you are in beast form" into one message. Also, add some new
feedback when lycanthropy and intrinsic polymorph are blocked by
intrinsic unchanging.
When gold is stolen by a leprechaun or lost when being "overwhelmed
by an urge to take a bath" while dipping in a fountain, if you had
99 gold pieces or less, you'd lose all of it (in the bath case, only
if it was at least 10 to start with), but if you had 100 or more,
you would lose a random amount which could be as little as 1. And
in the bath case, if the random amount was less than 10, you would
lose nothing but be told that "you lost some of your money in the
fountain". After this change, it is still possible to lose less
when starting with more, but not as likely and not as extreme a case
as maybe losing only 1 when starting with thousands.
The fountain-dip bath case has code to handle mutiple denominations
of coins, possibly the only place in the program where that exists.
I've left that alone although it should probably be taken out....
Extend the processing done by the wizard mode 'sanity_check' option
to look for anomalies with obj->owornmask since there seem to have
been a few lately. I haven't actually triggered any so this code
isn't very well exercized yet.
sanity_check uses pline() rather than impossible() or debugpline()
to deliver messages so might not be very useful with keymasking.
A sizeable chunk of this diff is just cleaning up indentation so
that I could see what I was working with....
Fix filtering used by the 'D' command, and a few other activities that
allow both object class filtering and bless/curse state filtering, so
that when both class(es) and state(s) are specified, objects need to
match both rather than either. D?C will present the player with cursed
scrolls to drop rather than all scrolls plus all other cursed objects.
This also fixes another instance when gold could end up with its bknown
flag set.
Allow the 'I' command to show inventory of known blessed items via
pseudo object classes B, C, U, and X. That's instead of an showing
inventory of specific object class. The two can't be combined
because 'I' operates on single character input.
I had to modify tty_yn_function to prevent it from forcing a BUCX
character into lower case (simply using lower case would cause a
conflict with 'u' and 'x' for inventory of shopping bill), and did
that by checking whether any of the acceptable response characters
are upper case. Pretty straightforward and shouldn't impact any
other uses that don't specify upper case choices.
I did the same thing for X11. Other interfaces most likely need
to do something similar. If they don't, a response of 'B' or 'C'
(for menustyle:traditional or menustyle:combination) will simply
not work, without causing any problems, same as typing an invalid
choice, and 'U' or 'X' will give shop feedback instead of the
requested subset of inventory.
The Guidebook revisions are untested.
Replace most uses of isspace() with a simple test for ' ' after
processing the string buffer with mungspaces (which replaces tab
with space, converts instances of consecutive whitespace into a
single space, and removes leading and trailing spaces). The uses
where this wasn't done now cast their argument to (uchar) so that
platforms with signed chars will never pass negative values to it.
I didn't mess with the menu coloring code (except for casts to the
isspace() argument); it almost certainly could benefit from using
mungspaces. I did mess with the symset processing quite a bit,
and hope I haven't accidentally broken anything. Default symbols
and DECgraphics symbols still parse and display ok, so the rest of
dat/symbols should be ok too. I didn't test symbols in the user's
config file because I don't remember how that's supposed to work.
Avoid initializer for auto-class array since many pre-ANSI compilers
didn't support that. Also, when ESC is used to cancel inventory
selection, don't continue on searching for that selection in invent.
Still doesn't address the reported stack corruption.
* 'alt' points to a buffer which has already been processed by
lcase(), so remove the redundant call to that routine;
* common error exit accessed via 'goto' could potentially leave
a dangling window structure (only if 'data' is corrupted though;
it's the only way a failure at that late point could occur).
I didn't find Michael's crasher, but I did find a couple of minor
problems:
knight's saddled pony wasn't found in the file lookup either though
there is an entry for pony ("tame saddled <foo>" stripped off "tame "
but not "saddled "; "saddled <foo>" would only be found of there was
a wildcard entry key of "*<foo>", like "*horse" to catch both horse
and warhorse);
"wombat" matched the bat entry rather than reporting an unknown
entity.
Remove the code that converted statues shown as monsters into stautes
shown as big rocks when saving and then reversed the effect when
restoring. It was done to preserve save file compatability with 3.4.3
where statue-as-monster glyphs didn't exist, so is no longer useful.
* remove several unreachable panic() calls; alloc() already panics
rather than return a null pointer;
* replace a couple of calls to malloc() with alloc() so that
MONITOR_HEAP won't produce spurious alloc/free discrepancies;
* replace several instances of variable 'd' with 'd0' so that it
doesn't conflict with the name of a function;
* comment out opvar_new_region() since it isn't used anywhere.
mongone() would send away a vault guard, whether alive or already
dead and parked at <0,0>, losing the information needed to remove
the temporary vault corridor once whatever obstacle (ie, the hero)
that's keeping it in place has been cleared. Have mongone() call
grddead() like mondead() does, so that a guard will just be moved
to <0,0> with 0 HP and kept there rather than being purged during
the next dead monster removal.
Move polearm context reset to m_detach() [and, unfortunately,
duplicated in grddead()] rather than in having it in dmonsfree()
since execution of the latter is deferred. I'm not sure offhand
whether a fast hero could get a second move before end-of-turn
purging of dead monsters.
Give 20 experience points the first time the hero reads a passage
from a tribute novel. It's enough to go from level 1 to 2 or from
2 to 3. By the time a book store is found, that's too trivial for
most to care about, but it's potentially useful to a pacifist.
When minimal_xname() set up a dummy object containing as few details
as possible, it wasn't setting up the fruit id field, so xname()
couldn't figure out what type of fruit it had and issued a warning.
I haven't managed a test case that uses minimal_xname so testing of
the fix is less than comprehensive. [Pasi got it through dopay, but
that only resorts to minimal_xname if the formatted name is really
long and would otherwise cause the shopkeeper's prompt to overflow.
Long fruit name combined with long individual object name wasn't
long enough to trigger that. Maybe uncursed, greased, rustproof
the like, or possibly just a longer shopkeeper name than I had?]
The previous fix prevents the crash from 'the()' when NO_GLYPH was
used as an index into the defsyms array, but it resulted in giving
feedback of "you attack thin air" regardless of what was at the
target location, reverting to the situation that the buggy code was
attempting to address in the first place. Handle that differently
by removing the unseen monster glyph sooner. Also, the underwater
handling wasn't working as intended.
I blamed Derek's pudding farming patch for introducing the problem,
but all that did was replace the offending line(s) with different
indentation. The older post-3.4.3 patch which produced the problem
was mine. Sorry, Derek.
For the POSIX regexp interface, move local declaration to beginning
of block to avoid requiring C99. Also switch to alloc() from bare
malloc() so that MONITOR_HEAP won't log a free which doesn't match
up to any allocation. This results in a change in behavior: if
the allocation fails, nethack will panic rather than report an
option parsing error. In practice there will be no difference
because nethack is not going to run out of dynamic memory during
initial options processing.
Convert search in tty menus from pmatch to case-insensitive pmatchi;
convert search in X11 menus from substring strstri to wildcard pmatchi.
tty bug: if the menu is full screen, the search prompt and subsequent
user input clobbers the menu header.
Add pmatchi() to perform case-insensitive wildcard matching, and
pmatchz() which is also case-insensitive and ignores spaces, dashes,
and underscores like the type of matching done during wish parsing.
At the moment, neither is being used, although DEBUGFILES handling
uses pmatch and needs to be taught to distinguish between case-
sensitive and case-insensitive filenames so will eventually use
pmatchi when appropriate.
From a bug report two years ago, mithril armor should have color
HI_SILVER rather than HI_METAL. Tolkien describes mithril as "like
silver which never tarnishes". (The bug report had a more precise
quote, and a github URL for a patch which I've ignored.) The Amulet
of Yendor has material set to MITHRIL but I haven't changed its color.
All amulets are HI_METAL, so if any modification is needed for it,
change the material rather than the color.
Mithril-coats are displayed as cyan both before and after this patch,
so it's hard to tell whether any change actually happened.
The previous USE_OLDARGS worked with gcc on Intel, but was inherently
unsafe. This method is completely safe, just obnoxiously intrusive.
It you disliked debugpline*(), you're bound to hate this....
Remove the requirement for <stdarg.h> that was introduced to lev_comp.
USE_STDARG still works. USE_OLDARGS required hackery but has been
tested and actually works, although I wouldn't trust it on platforms
where 'long' and 'char *' aren't the same size. USE_VARARGS didn't
require any hackery--aside from the conversion to core's pline code--
but has not been tested: <varargs.h> supplied with OSX won't compile,
with an #error directive that basically says "switch to <stdarg.h>".
I changed several printf formats of %i and %li to %d and %ld because
I'm not sure how widespread the 'i' variant was back in days of yore.
[TODO: avoid use of snprintf since pre-ANSI systems won't have it.]
DEBUGFILES set to "wintty.c" reported a bad cursor positioning attempt
at the end of the RIP tombstone, and when set to "questpgr.c wintty.c"
reported a whole bunch right at the start of the game when enumerating
all the quest messages for the chosen role. Both were triggered by
this x==0 call to tty_curs() near the end of process_text_window().
if (i == cw->maxrow) {
if(cw->type == NHW_TEXT){
tty_curs(BASE_WINDOW, 0, (int)ttyDisplay->cury+1);
cl_eos();
}
...
The x value is always decremented in tty_curs, so passing in 0 yields
a bad value of -1. The bad call returns without doing anything, and
when DEBUG is disabled, it does so silently.
No fixes entry; it was caused by a post-3.4.3 fix for something else.
Same functionality as was recently implemented for tty. If a
character like ':' is an explicit menu selector and the player
types it, select that menu entry rather than treating it as a
search request. (Same for other menu commands like '>', but
offhand I can't think of anything besides container looting's
': - look inside' that uses any non-letter selectors other
than '$', which isn't used as a menu meta command.)
Add a couple of missing .gitignore entries for files put into
place when building. include/tile.h is genearated for tiles
and dat/NetHack.ad is copied for X11 resource definitions.
The 'zdir' function parameter has the same name as 'zdir' global
variable, triggering a 'shadowing' warning. I had to read the 'if'
statement multiple times to convince myself it was doing what was
intended. It was, but I think this rewrite is easier to understand
(at least for my feeble reptilian brain).
I don't know who Tim Wright is, but his 15 mintues of fame has
lasted for at least a decade so I cut his comment out. (The 28.5
year old GAN one a dozen lines lower was a tempting target for
removal, but I managed to stop myself; otherwise it never ends.)
gcc complained about assigning string literals to variables that aren't
const. String literals don't have const type but modifying them results
in undefined behavior, so the warning is justified. Except in this case
the strings were never being modified so if wasn't justified after all.
Switch to a static buffer anyway.
Looting a container generates a menu which contains ': - look inside'
but the recent change to make ':' be a menu command for selecting
items which match a search string made it impossible to pick that item.
(Well, I suppose you could enter a search string which matched it, but
that's a nuisance compared to just directly picking a choice.) This
makes menu selection for tty give precedence to menu choice characters
over mapped menu commands when some character happens to be both. I'm
not sure whether it ought to be expended to group accelerators too, so
didn't do that.
There's bound to be a better way to do this, but it works.
Bag of tricks that had been used at least once was being described
as "empty" regardless of charge count, because it always fails the
Has_contents() test. After half this patch fixed that, it started
being flagged as "empty" as soon as the last charge was used rather
than after attempting to use it again after that, since 'cknown' was
being set whenever it was used. Only set that flag when applying
the bag has been observed to fail.
My dog bit an acid blob and triggered a crash, caught by SYSCF panictrace
but yielding confusing information. The backtrace included a call from
'rustm()+N' that turned out to be passivemm(), which was deferencing a
null pointer since no weapon was involved.
Unix command line processing required that the initial 'd' of
"-DECgraphics" be lowercase so that it wouldn't conflict with -D for
wizard mode. This retains -D for wizard mode and now also recognizes
"-debug" (case insensitive, but full 5 letters necessary) for the same
thing, and allows "-DECgraphics" to be capitalized as it is throughout
the rest of the program (actual matching is case-insensitive, so "-dec"
and "-decgraphics" still work. It now requires that anything after
"DEC" match the rest of that string instead of accepting "-DECanthing"
as a synonym for "-DECgraphics". Likewise for "-IBMgraphics": when
more than 3 letters are supplied, the extra ones must be an initial
substring of "graphics" rather than arbitrary characters.
The raw_printf() warnings don't actually work as intended, but that
isn't a change from the old behavior so I've left them in for now.
DECgraphics, IBMgraphics, and MACgraphics used to be recognized when
at least 3 letters were supplied back when they were true boolean
options. When they got demoted to shortcuts for the symset option,
they started needing 10 (DEC and IBM) or all 11 (MAC), otherwise
triggering "bad syntax in NETHACKOPTIONS" (or config file). Revert
to having the first three letters be sufficient.
Reported by Alex, the probabilites shown in the comment about the effect
of the mysterious force have been wrong ever since they were added 20 or
so years ago. Lawfuls and neutrals are much more likely to go down just
one level rather than 2 or 3.
No, not a blindness cure. :-} Post-3.4.3 revisions to makesingular()
inadvertently made it impossible to successfully wish for "the Eyes
of the Overworld" because the string got changed into "the Eye of the
Overworld" which doesn't match anything. So don't singularize "eyes".
(After this fix, wishing for "the Eyes of the Aethiopica" no longer
yields the Eye of same, but I think that's the correct behavior.)
Restore the variant phrasing used when more than one stack of potions of
acid explode while being inflicted with water damage. First message:
"a potion explodes" or "some potions explode"; second and subsequent
messages: "another potion explodes" or "more potions explode". This
trivial feature stopped working when erosion handling was overhauled and
old water_damage was split into current water_damage_chain+water_damage.
Augment the message so that vague "potion" is only used when the object's
dknown flag isn't set (ie, object hasn't been seen yet). Use "<color>
potion" or "potion of acid" otherwise, depending upon whether such potions
have been fully discovered.
If getenv("DEBUGFILES") yields a value then it takes precedence over
sysconf.DEBUGFILES or sys.c's #define DEBUGFILES. (It probably should
only be controlled via environment since it is not a system-wide
attribute, but I haven't taken out the SYSCF handling for it.)
Bug report included a pointer to a fix; this patch is a superset.
Gold pieces dropped on an altar by the player got their bknown flag set,
which is incorrect since bless/curse doesn't apply to coins. If a
monster (in reported case, a slain temple priest) dropped gold there too
then the two stacks wouldn't merge. For the normal !GOLDOBJ config, the
problem goes away as soon as the gold gets picked up. I didn't test for
GOLDOBJ but think two inventory slots containing gold can result.
The superset part is to not break agnostic conduct by dropping gold
on an altar since no information is revealed when doing that.
[This was one of the very last patches checked into the old cvs repository,
where the somewhat out of date message above was accidentally omitted.]
* Replace variadic debugpline() with fixed argument debugpline0(str),
debugpline1(fmt,arg), and so on so that C99 support isn't required;
* showdebug() becomes a function rather than a macro and handles a
bit more;
* two debugpline() calls in light.c have been changed to impossible();
* DEBUGFILES macro (in sys.c) can substitute for SYSCF's DEBUGFILES
setting in !SYSCF configuration (I hope that's temporary).
If getenv("DEBUGFILES") yields a value then it takes precedence over
sysconf.DEBUGFILES or sys.c's #define DEBUGFILES. (It probably should
only be controlled via environment since it is not a system-wide
attribute, but I haven't taken out the SYSCF handling for it.)
Bug report included a pointer to a fix; this patch is a superset.
Gold pieces dropped on an altar by the player got their bknown flag set,
which is incorrect since bless/curse doesn't apply to coins. If a
monster (in reported case, a slain temple priest) dropped gold there too
then the two stacks wouldn't merge. For the normal !GOLDOBJ config, the
problem goes away as soon as the gold gets picked up. I didn't test for
GOLDOBJ but think two inventory slots containing gold can result.
The superset part is to not break agnostic conduct by dropping gold
on an altar since no information is revealed when doing that.
[This was one of the very last patches checked into the old cvs repository,
where the somewhat out of date message above was accidentally omitted.]
* Replace variadic debugpline() with fixed argument debugpline0(str),
debugpline1(fmt,arg), and so on so that C99 support isn't required;
* showdebug() becomes a function rather than a macro and handles a
bit more;
* two debugpline() calls in light.c have been changed to impossible();
* DEBUGFILES macro (in sys.c) can substitute for SYSCF's DEBUGFILES
setting in !SYSCF configuration (I hope that's temporary).
PatR