Other half of the Debian reproducible-builds patch. For Unix
Makefile.top, explicitly set locale to generic 'C' when running
'dlb' during install, so that wildcard expansion will yield a
predictable ordering regardless of collation order specified by
the local enviroenment. Otherwise contents of the 'nhdat'
container might be different--when viewed as a single data file
itself--during subsequent rebuild even when no changes to source
or data have been made and each module inside remains the same.
This assumes that locale doesn't matter during generation of any
of the data files (whether destined for dlb or not). I don't
think the utility programs attempt any sorting on the fly or
other locale-dependent output but wouldn't swear to that....
Take the 4-5 line Debian patch and turn it into six dozen lines of
new code. The submitted patch introduces use of several C library
routines that aren't presently in use, so would need testing by all
functional or nearly-functional ports to verify that it wouldn't
break anything. It also switched the formatted build date+time
from localtime to UTC. This makes the code conditional so it can
be ignored by anybody and avoid the risk of breakage. And a lot of
the increase in size is comments attempting to explain what the new
conditional is for: when REPRODUCIBLE_BUILD is defined, makedefs
will use getenv("SOURCE_DATE_EPOCH") (whose value is an integer
representing seconds since 1-Jan-1970) instead of current date+time
when generating date.h. The purpose is to be able to rebuild at a
later date and produce an identical program, which doesn't happen
when compile time gets incorporated into the binary.
I've added some sanity checking to try to make sure the getenv()
value obtained isn't bogus. And the version string put into date.h
will be slightly different, allowing someone who sees date.h or 'v'
output to tell whether SOURCE_DATE_EPOCH was involved: showing
"<port> NetHack <version> last revision <date>" instead of the
usual "... last build <date>".
To test, checkout a new branch for building, make any local edits
to unixconf.h and config.h, including enabling REPRODUCIBLE_BUILD,
git add+commit them, then use
SOURCE_DATE_EPOCH=`git log -1 --pretty=%ct` make install
Other ports will need a bit more work to set up the environment,
but can still use git to track file dates and supply the latest.
Building with alternate configurations could be accomplished by
using tags instead of 'log -1' or by using distinct build branches
where nothing is commited/merged/rebased after completed build.
Unresolved issue: BUILD_DATE, VERSION_ID, and COPYRIGHT_BANNER_C
contain formatted date+time but omit timezone. SOURCE_DATE_EPOCH
is assumed to be UTC but the formatted values don't say so, so it
might appear to be incorrect when compared with local time. We
definitely don't want to start mucking about with timezones within
nethack, so I think we just live with this. It's not an issue for
default configruation where REPRODUCIBLE_BUILD is left disabled.
The Makefile race condition report included a link to a log file
of the build attempt, and it contained this:
makedefs.c: In function 'do_grep_control':
makedefs.c:611:26: warning: suggest parentheses around operand of '!'
or change '|' to '||' or '!' to '~' [-Wparentheses]
#define ST_LD(old, opp) (!!(old) | (!!(opp) << 1))
^
makedefs.c:722:37: note: in expansion of macro 'ST_LD'
grep_stack[++grep_sp] = ST_LD(grep_writing, !isif);
^
They're using a more recent version of gcc than I am, because my
CFLAGS includes -Wparentheses (via -Wall) and I don't get that.
It's a little confusing, but I think it's whining that we might
have meant !!((old) | (!!(opp) << 1))
rather than (!!(old)) | (!!(opp) << 1).
The latter is what we get (and what we intended--no bug here).
I changed it to something that more directly reflects the intent
since it's not bit twiddling within some crucial innermost loop.
Update sys/unix/Makefile.src to force it to build monst.o and
objects.o before attempting to build makedefs, so that building
the latter doesn't use the rules for those two object files in
Makefile.utl. Prior to this, if 'make' was building multiple
targets in parallel monst.o and/or objects.o might be clobbered
when a process using util/Makefile tries to build them while its
parent is also building them via src/Makefile.
This includes a bit of reformatting which wasn't present in
yesterday's email attachment.
Make 'zeromonst' global instead of local to makemon.c. Its address
isn't used as a special value like &zeroobj, but it is useful to
have available for initializing various pseudo-monsters.
modified:
include/decl.h
src/decl.c, makemon.c, mkobj.c, mplayer.c, teleport.c
Some worthwhile stuff from abandoned 'git stash':
is_plural() macro wasn't comprehensive;
a couple of return values where writing with a magic marker was
causing time to elapse even though nothing happened;
comment formatting for saddle being left in shop by dying steed.
Rescuing an old revision from bit rot: If one of fog clouds or
vampire bats has been genocided and you try to polymorph a vampire
disguised as the other, it won't change form because the shape it's
currently in is the only candidate shape left for vampshifting.
This makes shapechangers who fail to take on a new shape when
polymorphed try again, specifying original form on the second try.
It's unlikely to affect chameleons, but disguised vampires will
sometimes become undisguised instead of seeming to be immune from
polymorph.
The meta keystroke commands which use an uppercase letter were all
missing from dat/hh:
M-A annotate level
M-C show conduct
M-N name something (synonym for M-n, which is a synonyn for 'C'all)
M-O display dungeon overview
M-R ride/unride steed
M-T tip a container
(All meta keystroke command shortcuts are missing from dat/help.)
Noticed when testing tty menucolor changes recently, using the 'O'
command to interactively add a new entry would accept one and then
quit, unlike remove which accepts multiple at a time and then goes
back to the menu where you could choose 'remove' all over again.
Adding is still done one entry at a time, but instead of finishing,
it goes back to the menu where you can choose to add another.
After some permutation of commands which displayed items, the 'd'
command presented a prompt with the list of letters scrambled (in
loot order or pack order rather than invlet order), so explicitly
sort when getobj operates. Done for ggetobj too.
For menustyle:Traditional, ',' followed by 'm' presented a pickup
list in pile order even when sortloot was 'l' or 'f'. That was an
unintentional change during the 'revamp'.
There was a report during beta testing that menu lines which were
displayed in color showed the whole line in color, unless/until an
item was selected or unselected, in which case the '-', '+', or '#'
was rendered in monochrome. The suggestion then was to redraw the
selection character in color, but I went the other way. Menu
entries will render the selector letter and selection indicator in
monochrome all the time, and only the text of the menu entry will
honor menucolors.
Reported directly to devteam for recent git code, the "sortloot
revamp" patch could trigger an object lost panic after calling
query_objlist() when menustyle was full or partial and player
picked up a subset of available items. Modified head-of-list was
not being propagated to its source in pickup(). Use an extra layer
of indirection.
Reported directly to devteam (for 3.4.3 but still present in 3.6.0):
an unseen landmine explosion which caused scatter() to break a
boulder or statue would give feedback as if the hero could see the
boulder or statue being destroyed.
Also, a couple of landmine explosion messages didn't take deafness
into account.
This one had been intended for longer than several years, but I
hadn't gotten around to it. When consorting with succubi and
incubi, very high Cha+Int no longer guarantees that a positive
outcome will occur.
Chance of positive outcome is still quite high and most of the
negative outcomes are pretty easy to repair, so this isn't likely
to make a significant impact. However, the possibility of losing
spell power will matter for some players....
This was sitting around for several years. When positive luck
lessens a negative effect from sitting on a throne, reduce luck so
that repeated occurances will eventually get the intended result.
Most of the humanoid species have Lords and several have Kings,
but none of them have Ladies or Queens. When a female grows up
to reach that level of monster, she changes into male. This fix
gives an alternate message acknowledging that change rather than
prevent taking on the stronger form. A better fix would be to
add ogre ladies and dwarf queens as separate monsters, but doing
so will break 3.6.0 save file compatibility.
(I started out with an alternate fix, adding mons[].fname for the
dozen or so creatures which warrant an alternate name for females.
But that requires statues, figurines, corpses, tins, and maybe
even eggs to track gender [some statues already do, and corpses
and statues with attached mtraits also implicitly do] and to not
stack with equivalent ones of the opposite gender. Plus glyphs
to track them, and new tiles. It was becoming too complicated
for such a relatively unimportant feature. Separate monsters is
the way to go, deferred until save file format changes again.)
Blindness due to face covered by pie was ignored for several cases
of magically curing blindness--cleaning the face seems better than
adjusting timeout to account for u.ucreamed for those cases. A few
instances of taking stun or confusion damage overrode existing stun
or confusion rather than increasing it. Plus a copy/paste mistake
for dual stun+confusion when casting an expired spell.
There was also a suggestion that vomiting when already nauseated
should decrement the timer instead of increasing it. But there is a
negative effect for as long as it's in effect, so I left that as is.
Fix some typos in the sort-by-invlet code and a logic error in the
lately added subclass sorting for sort-by-pack. Regular inventory
display only works correctly for the latter if invlet is the tie-
breaker within object classes. When helmet/gloves/boots/&c and
ammo/launcher/missile/&c sub-categories already break ties for armor
and weapon classes, inventory ended up out of alphabetical order.
When removing items from a container via menu, list gold as '$'
instead of 'a' when it is the first item. Requested during beta
testing last year....
When gold isn't first ('sortpack' false, or custom 'inv_order[]'),
it uses the next letter in sequence instead of '$', otherwise it
would be the only item out of sequence.
Change the sortloot option to use qsort() instead of naive insertion
sort. After sorting, it reorders the linked list into the sorted
order, so might have some subtle change(s) in behavior since that
wasn't done before.
pickup.c includes some formatting cleanup.
modified:
include/extern.h, hack.h, obj.h
src/do.c, do_wear.c, end.c, invent.c, pickup.c
Accept "male" or "female" when specifying monster type for ^G.
Groundwork for testing and hopefully eventually fixing "female
gnome" grows up into "gnome lord" and becomes male.
Polyself with gender change into a creature with fixed gender
would deliver a message containing "a <creature>" regardless of
whether "an" was warranted.
(Into any creature which supports both genders it yielded
"a male <creature>" or "a female <creature>" so "an" was never
needed. And when no gender change was involved, it used an()
so got "a <creature>" or "an <creature>" as applicable.)
There have been two or three reports on getting feedback about
amulets rusting. Object formatting doesn't display erosion for
them, so being told about damage then not seeing that damage
feels like a bug. Even if damage was displayed, it has no effect
on them so would still feel somewhat strange. It does display
erosion for wands and rings, which is strange too.
This limits erosion damage--and its feedback--to items which are
actually impacted by erosion: armor, weapons and weapon-tools;
also heavy iron balls and iron chains since they've traditionally
shown rust even though it has little effect.
A side-effect of this change is that flammable items (other than
armor and weapons) which don't burn up immediately will no longer
become burnt, then very burnt, thorougly burnt, and finally be
destroyed. Since the player couldn't see or possibly repair the
erosion state, it seemed incomplete. It could be reinstated by
making other flammable items be subject to erosion and displayed
as such by xname() & co.
Wishing now avoids applying erosion and erosion-proofing to items
that aren't affected by it, regardless of material. It also now
allows wishing for "rusty rustproof <iron-object>" which used to
suppress "rusty" in that combination and triggered a couple of
old bug reports.
Heavy iron balls and iron chains can have rust repaired and can
be made rustproof by wielding, then reading enchant weapon while
confused, as if they were weapons.
Pt 1 was about the wrong message delivered when a high priest
rejects being given a name by the player, and was fixed weeks ago.
Pt 2 is about zaps on the Elemental Plane of Air which reach the
edge of the map not having their temporary display effect removed
after "the <zap> vanishes in the aether". There was a 'goto' in
use which bypassed the tmp_at(DISP_END) call. I guess Dijkstra
earns an "I told you so" here.
"Looting many containers via menu cannot be stopped". When the
player uses #loot command at a location with multiple containers,
a menu of which ones to loot is presented and player can pick any
or all of them. But if you terminate the looting of a particular
container with ESC, it goes on to the next selected one rather than
stopping the loot action because that's what the 'q' choice does.
The simplest fix would be to allow choosing only one container
from the "loot which?" menu, but this retains the ability to loot
multiple containers on a pile in one turn. It makes looting
stoppable by extending the ":iobrsq or ?" prompt, adding 'n' for
"next container" and changing 'q' from "done with this container"
to "done looting" (with ESC still a synonym for 'q'). When just
one container is being looted, or when on the last of N containers,
'n' is not shown but is still accepted (and treated as 'q').
Also, use_container() was using a menu for ":iobrsq" if player had
menustyle set to Full when it was intended to be for Partial (name
confusion...). This switches Partial to use menu for loot action,
and leaves Full with that since that's how 3.6.0 has been behaving.
Traditional and Combination use the prompt string and single char
response.
The bug report was actually about letting monsters use fire horns
without checking whether they could actually use wind instruments.
The previous fix probably handled most cases by excluding animals
and mindless creatures, but this is a more specific fix for MUSE
of fire and frost horns--they must pass the same test as the hero
and it's not limited to stopping being turned into slime.
There was no check for being capable of using items when an attempt
to cure being turned into green slime picked scroll, wand, or horn
of fire.
Also, implement a 'TODO' in the same section of code. Monsters
can enter fire traps to cure themselves from slime. I made that
be for monsters smart enough to use items too, even though there's
no actual item involved.
Umpteenth revision of the X11 extended command menu. Add a new
resource to NetHack.ad to control its initial size.
I still hope there's a better way to do this, but this is my last
shot at it.
Let monsters who have a weapon attack for non-physical damage dish
out physical damage instead of doing the drain life or drain
strength they usually do if they happen to be wielding cockatrice
corpses or a couple of particular aritfacts that do more harm
than just level drain. (Other artifacts are candidates, but I
don't think it's worth checking for them since the monsters
involved have such a small chance of acquiring and wielding them.)
Also switch to physical if monster's ability has been cancelled.
Only barrow wight, Nazgul, and erinys are affected. Yeenoghu and
the Master Assassin have a weapon attack for physical damage and
another one for non-physical damage (not necessarily delivered in
that order). They haven't been changed--only the physical damage
attack has a chance to apply their weapon's special damage.
While looking at #H4265 ("Bug - Monsters opening doors" about
feedback naming the unseen monster who opened a door), I didn't
find the the problem. But I did notice a couple of suspicious
constructs. Fix an assignment that gave a boolean variable a
value of 16, and add parentheses around 'a & b' in (a & b && c).
The latter isn't incorrect, it just looks strange.
Report states that using OSX Xcode IDE results in use of 'clang
Modules', whatever those are, and role.c's 'filter' struct ends up
conflicting with a function declared by <curses.h> (or possibly
<ncurses.h> since one includes the other). src/role.c does not
include <curses.h>, so this smacks of the problems caused by using
precompiled headers on pre-OSX Mac.
Instead of trying to import nethack into Xcode, I temporarily
inserted '#include <curses.h>' at the end of unixconf.h. gcc did
complain about 'filter' in role.c (but not in invent.c, despite
-Wshadow), and then complained about termcap.c using TRUE when it
wasn't defined (after in had been #undef'd, where there's a comment
stating that it won't be used in the rest of that file), and also
complained about static function winch() in wintty.c conflicting
with external winch() in curses.
This renames 'filter' and 'winch()' to things that won't conflict.
Also, our winch() is a signal handler but had the wrong signature
for one. And the troublesome use of TRUE was in code that was
supposed to be dealing with int rather than boolean.
Lawful angels deliver taunt messages from a pool of messages which
might mention the lawful god; demons and non-lawful angels draw from
another pool which doesn't mention any gods. Since it is odd for a
'renegade' angel to claim to be operating for its god, choose taunts
from the other pool of messages for renegade lawful angels.
Not related: some formatting fixups in include/mextra.h.
One entry among many in #H4216: make shuriken be a pre-discovered
item for monk role. The word "shuriken" comes from Japanese and
martial-arts monk is primarily Chinese, but shuriken/throwing-star
is a martial-arts type of weapon and monks get a multi-shot bonus
for it (even though they can't advance its skill beyond basic...).
Two different reports complaining that having the Wizard steal the
hero's quest artifact is a bad thing. This doesn't change that,
but it does make all quest artifacts become equal targets so that
wishing for other roles' artifacts doesn't offer such a safe way to
have whichever special attributes they provide.
Quest artifacts are actually higher priority targets for theft than
the Amulet. I suspect that probably wasn't originally intended,
but I left things that way. Taking quest artifacts leaves the hero
more vulnerable to future thefts, and once they're gone the Amulet
has priority over the invocation tools.
When using a stethoscope or wand of probing on a long worm, report
the number of segments it has in the feedback given.
Some of the extra bhitpos and/or notonhead assigments may not be
necessary. They were added when I was trying to figure out the
question of why probing of a tail segment revealed a long worm's
inventory even though the code explicitly prevents that. (Answer:
it didn't; I had misinterpreted bz 12 to think that that was what
was being reported. You need to use wand of probing--or "insigtful"
Magicbane hit--on the head in order to see its inventory or be told
"not carrying anything".)
I initially misunderstood this bug report about a nymph who was
polymorphed into a long worm while carrying a cursed figurine.
It wasn't about a long worm having inventory or about probing of
the worm's tail revealing that it had inventory, it was about the
message given when the cursed figurine activated itself. If that
happened while the head was out of view but at least one tail
segment was visible, the message about the new monster emerging
from the long worm's backpack implied that that pack was carried
by the tail segment.
Only give the emerge-from-backpack message when the worm's head
is visible. Likewise if a carried egg hatches.
Requested during beta testing last year, include a menu entry of
"- - your bare hands" (or "your gloved hands") for wielding,
"- - empty quiver" for readying quiver,
"- - your fingertip" for engraving, or
"- - your fingers" for applying grease
if the user responds with '?' or '*' at the
"What do you want to {wield|ready|write with|grease}? [- abc or ?*]"
getobj prompt. (First dash is inventory selector 'letter', second
dash is menu separator between the letter and its choice description.)
I'm not sure whether I'll do the same for Guidebook.tex.
Guidebook.txt wasn't up to date. I've got some substantive (but minor)
changes coming so am not checking in an updated one yet.
Even out the summoning distribution by adding more lawful candidates.
There used to be only 4; now there are 10. Chaotics have 14, so are
still more likely to get "neutral or own alignment" and stop, but the
difference is now pretty small once you factor in the 18 neutral ones.
In theory nasty() could summon 200 critters at a time, although the
chance seems fairly remote. But it was biased towards having lawfuls
summon more critters than others since there are fewer lawfuls in the
nasties[] list. This puts a cap of 8 successful makemon() calls,
enough to completely surround the hero. More than 8 monsters can be
generated, if any of the makemon() calls produces a group. (I think
fire giants are the only thing in nasties[] that ever come in groups.)
It's still biased toward lawful summoners trying more times hoping to
produce a lawful creature and generating chaotic ones in the process.
The bug report also thought there was some problem between chaotic
and unaligned or with the Wizard, but unaligned is treated as if it
were chaotic (due to use of sgn() in the two or three places where
alignment type is manipulated) so that isn't an actual problem.
I was thinking about iflags rather than context and didn't realize
that the change to maxinum number of passages would breal save files.
Put the tribute context back to 3.6.0 size.
Anyone who grabbed from public git yesterday is potentially in for
some temporary trouble. That's the risk they take for trying to stay
on the bleeding edge.
Death Quotes have reached the current limit of 30 passages per 'book'.
Instead of increasing that, change the selection code to be able to
operate on a subset (dropped from 30 down to 20) at a time while
keeping the excess available for later selection.
Chatting with Death (more than 20 times since he also delivers non-
tribute messages) should cycle through 20 of his 30 passages without
repeating. After that, another subset of 20 out of the 30 will be
chosen, independent of the first set so might contain all, some, or
none of the 10 that left out before.
Sometimes you can see a hidden monster without bringing it out of
hiding (wand of probing, blessed potion of monster detection) but
look_at wasn't mentioning the fact that the monster was hidden and
probing described mimics accurately but lumped all hiders together
as "concealed". Describe all hidden monsters more consistently.
Reported directly to devteam, the 'R' command would let you take off
a suit from under a cloak or a shirt from under a suit and/or a cloak,
and it didn't require any extra turns. 'T' doesn't allow either of
those. ('A' lets you take off a suit from under a cloak, adding in
extra turns to implicitly take the cloak off and put it back on,
but doesn't allow the same for shirt.)
'R' also let you attempt to take off embedded dragon scales if you
were wearing 2 or more accessories (or just 1 with paranoid_confirm
set for takeoff/remove), which triggered impossible "select_off:
<scales object> (embedded in skin)???".
When the extended command menu auto-scrolls as the player types in
characters, scroll so that all matching entries come into view rather
than just the first one. For example, it someone types 'w', instead
of just highlighting and showing "wipe", it will highlight "wipe"
(because that has become the default response when <return> is used)
but also show "wiz-this", "wiz-that", and "wmode". It actually shows
one extra entry beyond the last matching one--so you can see that
there aren't any more ambiguous choices--except for 'w' where "wmode"
is the very last extended command.
Previously, on subsequent popups of the extended command menu, the
scrollbar's slider was left drawn in the position it was in during
the previous time even though menu content wasn't scrolled. Now it's
forced back to the top (non-scrolled) position when that menu is
popped up.
When the extended command menu is big enough to need a scrollbar,
leave more elbow room when forcing its height to fit on screen.
The last entry was frequently obscured by OSX's "docking tray"
desktop decoration and the resize hotspot (bottom right corner of
the menu popup) could be hard to access.
I'm not particularly happy with this code. There really has to be
a better way to accomplish what's needed.
Newer compiler found new things to complain about.
NHinternal/shared/sys/share/flexhack.h has been updated.
Generated sys/share/ sources have to be committed separately so
they happen after this update fills in new banner comment for
their source.
Another OS upgrade (OSX 10.6.8 -> 10.8.5) with different toolset,
another change in compiler behavior. Earlier 'gcc -Wwrite-strings'
didn't complain about passing string literals as 'char *' paremeters
if there was no prototype in scope. This one found one or two of
those in options.c and several in makedefs.c (fix coming soon in a
separate commit...). This adds some missing prototypes and reorders
the existing ones to match their order within the file. There were
also several functions which were declared static in their advance
declarations but not in the definitions, which can be confusing when
reading the source.
Color was only being tracked for locations that had the pile of
objects flag set. And hilite_pile made a monster on a pile take
on the color of the top object of the pile.
This restores the tracking of color for the whole map, and makes
highlighted piles be drawn in inverse like highligted pets. The
drawing routine doesn't know the difference (but could tell, if
necessary, by testing whether the glyph is an object or a monster).
Also, variables 'inbuf', 'inptr', and 'incount' were global; limit
their scope to winmap.c.
Force a screen redraw if the tiled_map option is toggled via the 'O'
command. The X11 interface switches map modes even without this, but
conceptually it's something that must be done when the option setting
is changed.
The three line change I made previously to implement highlighting for
prompts that ask for single-character input was easy and worked well
for a tiles map, but it didn't look very good for a text map. This
handles both text map and tile map and also adds a configurable
'highlight_prompt' X resource to let the user enable or disable the
feature. The resource template file (win/X11/NetHack.ad, copied to
$HACKDIR during install) now has it enabled by default.
The highlighting--more specifically, the "lowlighting" when no prompt
is active--still looks bad if the map window has a vertical scrollbar
on left edge. I don't have any inspiration about how to fix that up.
If the user hasn't explicitly loaded application defaults (which I
haven't been doing), the X11 interface behaves differently if invoked
via the shell script than if the executable is run directly, because
the script sets up a path so that X can find $HACKDIR/NetHack.ad.
This hides the difference by reading in that file during initialization
and feeding its contents to XtAppInitialize as fallback resources.
The scroll bar on the message window doesn't work for me, just like
the one on the extend commands menu. Moving the pointer to it does
change the cursor, but neither trying to drag the slider nor clicking
above it will make it do anything. However, at some point I managed
to accidentally scroll the message window, and new messages never
restored it to the unscrolled state. New messages were hidden until
enough even newer ones had been delivered to push the hidden ones
into view. So this adds key translations to scroll the message
window via the arrow keys. Clicking on the scroll bar doesn't change
focus to the right place, so I have to click the text display area of
the message window. That triggers a beep (I suspect it's interpreted
as an attempt to move the hero beyond the map.) But then the messages
can be scrolled via the cursor keys. Getting focus back to the map
seems tricky since doing it via pointer is interpreted as a travel
command. Not quite optimal....
Anyway, being able to scroll the message window let me figure out how
to unscroll it when new messages are delivered.
For the 'slow' config (prompting for single-char input done on a
fixed line at the top of the map rather than via a popup window),
invert the background and foreground when creating it so that it
looks like part of the map, then invert again when a prompt is
active in order to highlight that prompt.
weight() didn't know how to calculate a glob's weight. When one glob
absorbs another, that isn't used, but when the hero eats part of a
glob, it is, and the result was incorrect.
For make install or update, dlb.o was being recompiled unnecessarily
when building util/dlb because the dependency in Makefile.utl listed
hack.h instead of config.h for it. Then attempting to run nethack
under debugger control could cause debugger complaints about dlb.o
being newer than nethack.
Updated for unix and vms.
Another one from 4.5 years ago: the Tsurugi of Muramasa ought to
be able to split puddings in keeping with its special attack effect
of slicing things in two. The suggestion was more extreme: always
split instead of kill. This generalizes the less extreme version;
METAL weapons (scalpel and tsurugi) can now split puddings like IRON
weapons do even though their user faces no risk of having the weapon
become rusty. Sam's quest artifact receives no special treatment.
Bashing puddings with wielded iron-tipped projectiles was splitting
them. This prevents that (and also for metal-tipped ya).
Every role has a specific spell that they having an easier time
casting. Samurai's special spell is clairvoyance but samurai is
restricted in divination spells. Requested about 4.5 years ago:
allow samurai to achieve basic skill in divination.
Similar for barbarian, special spell is haste self but escape spells
are restricted. All the other roles can already get at least basic
in their special spell's category.
There were several choices:
1) leave things as they are;
2) give those two roles different special spells;
3) allow them to reach basic in the spell category of the existing
special spell;
4) #2 for one of those roles, #3 for the other.
I went with #3. To compensate, reduce attack spell skill limit from
skilled to basic for both. (#4 might be better, since the reason to
want divination enhanced is most likely identify and magic mapping,
not interest in clairvoyance.)
Free askname's widgets after use and free getlin's and yn_function's
persistent widgets at end of game.
When loading an entire text file into one long string in memory,
use strcpy on a pointer to the end of the string instead of having
strcat repeatedly churn through the entire string as it grows for
each line. [Since that's only used for small help files (biggest
is dat/history), this optimization is probably not noticeable.]
Also, a handful of new comments and quite a bit of reformatting.
I upgraded from OSX 10.5.8 via 10.6.3 to 10.6.8, plus Xcode to whatever
version was on the 10.6 dvd, and ended up with a more recent version of
gcc that is configured to use 64 bit longs and 64 bit pointers (by
default; presumably that can be changed if necessary). It triggered
several warnings about converting int to pointer of different size or
vice versa even when explicit casts were in use, and a couple of other
things.
Most shop messages use shkname() to give the shopkeeper's accurate
name (or hallucinatory substitute) even if he or she can't be seen.
stolen_value() was using mon_nam(), which calls shkname() if the
monster is a shopkeeper who can be seen, but produces "it" when not
seen. Change it to use shkname() like the rest of the shop routines.
Also, replace Monnam() (quite a few instances) with new Shknam() to
do the same duty when the name is at the start of a sentence.
There was also a very obscure bug where if you could see two
shopkeepers at the same time, you could probe the map one spot at
a time with repeated use of the 'p' command to locate monsters in
general and other shopkeepers in particular. Very tedious and not
very useful, but now fixed.
The big memory allocation for tiles that was unfreed according to
heaputil was actually freed by X according to a comment in the code.
But free it explicitly for #if MONITOR_HEAP so that the alloc/free
tracking stays accurate.
Also, the cached extended commands menu was not being freed, so take
care of that. I wasn't sure where to handle it; I ended up making it
happen when the map window is torn down.
I just tried to build with MONITOR_HEAP defined in unixconf.h plus
WANT_WIN_X11=1 for make. tile2x11 wouldn't link. It doesn't use
alloc() and free(), but it does link with drawing.o, and drawing.o
has calls to free(), so MONITOR_HEAP makes it need nhfree() from
alloc.o.
All the miscellaneous tile utilities seem to link with drawing.o, so
they all need alloc.o too, and that drags in util/panic.o as well.
I've only changed it for Unix where I can actually test the change
but other platforms probably should do this too. Or we need to redo
drawing.c so that the small subset of stuff utilities want is
separate from the code the core uses for that same stuff.
Attempting to read a cursed spellbook fails with a nasty effect. But
a non-cursed book can become cursed while being read (malignant aura
after Wizard has been killed). Assuming no interruption for other
reasons, the read would finish, the spell be learned, and then the
nasty effect would be given. This changes things so that if the book
being read becomes cursed and the hero notices (book's bknown flag is
set), the read-in-progress will be interrupted. Resuming will take
the attempting-to-read-a-cursed-book path. Unfortunately, if the
hero doesn't notice, the old behavior still applies. Maybe the new
behavior should happen even if bknown isn't set (but then player
won't be told why the interruption occurred).
X11 had been ignoring add_menu(..., MENU_SELECTED) to specify a
pre-selected menu entry. This adds support for that.
Attempt to implement pre-selected entry for PICK_ONE menu sanely by
returning the pre-selected entry instead of toggling it off if the
user chooses it explicitly. Inner workings of menus are convoluted
so I'm not sure it's 100% correct, although testing hasn't found any
problems. (tty currently returns 0 for "nothing picked" when
explicitly picking a pre-selected entry in a PICK_ONE menu, and the
core jumps through hoops to handle it. That can't be cleaned up until
all interfaces which support pre-selected entries achieve sanity.)
Make "random" be chosen for <return> or <enter> during role selection
and highlight it to reflect that. (Role selection for X11 uses its
own code instead of nethack menus, so pre-selection isn't applicable.)
After the recent shopkeeper fix, I wanted to find out what happens if
you turn to stone in the spot inside the shop door. It didn't go too
well--a change of mine from three weeks ago caused a crash due to
passing a null pointer to strcmp(). Death from being turned to stone
or from starvation when there was no while-helpless reason (probably
not possible for starving) triggered it. This fixes that.
As far as the test goes, the shopkeeper takes your inventory and moves
it all the way into the shop, and a statue of the petrified hero is
left without contents in the spot in front of the door. That shk was
awfully quick....
Post-3.6.0 bug, so no fixes entry.
The impossible about partly eaten glob having more nutrition than
an untouched one after another glob had been absorbed into it was
not a 3.6.0 bug, it was an interim situation when I converted glob
nutrition to be based on glob weight rather than on the creature
corpse weight. So take that fixes36.1 entry out.
Add one about black puddings that was included in the glob patch.
Globs on the floor used different criteria (anything goes) than globs
in inventory (mostly requiring same ownership when in shops and same
curse/bless state--other stuff generally isn't applicable) when
deciding whether two globs should merge. That was okay as long as
the globs on the floor were from being left behind when a pudding or
ooze was killed, but not if the player had picked some up, dipped
them in holy or unholy water, and dropped them again. This changes
things so that globs on the floor use the same criteria as globs in
inventory when deciding whether to coallesce.
Also, my earlier fix was modifying globs in the mergeable() test (to
make bknown and rknown match) rather than during actual merge, which
would be a problem if the merger didn't take place for some reason.
The report that a tame Archon got two "<pet>'s long sword is not
affected" messages thought there was some duplication error when a
flaming sphere exploded, which was incorrect. Since an Archon has
two weapon attacks, getting that message twice just meant that both
attacks hit. However, the player has only 1/6 chance to suffering
passive fire damage to weapon, where monster-on-monster or monster-
on-polyd-hero was inflicting that for every successful hit, so
there was a bug here after all. Give monsters the same 1/6 chance.
Also, add even more verbosity to that message--now that it won't be
delivered so often--to mention what didn't affect the item.
While investigating this, I noticed that hitting a steam vortex
with a flammable weapon was doing fire damage to that weapon. Fire
damage in the steam vortex definition makes some sense in that it
allows fire resistance to give protection, but dishing out actual
fire damage makes no sense and is now prevented for passive counter-
effects.
The code to handle monsters fleeing up the upstairs on level 1 was
expecting those stairs to be normal upstairs but they are actually
sstairs like the ones down into the mines. So monsters fled to the
Plane of Earth instead of escaping the dungeon as intended.
When destroy_item() or destroy_mitem() burned up a glob of green slime,
they had the message index and damage amount reversed. This could give
a nonsense message ("the glob of green slime freezes and shatters") or
go out of array bounds and wreak havoc. Even if the message index had
been correct, fatal damage would have produced an incorrect cause of
death since it would have used a potion or scroll string.
Now globs will boil and explode like potions, and damage will be
proportional to the size (weight) of the glob, which seems to be the
original intent.
PatR