gcc has recognized various "magic comments" for white-listing
occurrences of implicit fallthrough in switch statements for
a long time:
The range and shape of "falls through" comments accepted are
contingent upon the level of the warning. (The default level is =3.)
-Wimplicit-fallthrough=0 disables the warning altogether.
-Wimplicit-fallthrough=1 treats any kind of comment as a "falls through" comment.
-Wimplicit-fallthrough=2 essentially accepts any comment that contains something
that matches (case insensitively) "falls?[ \t-]*thr(ough|u)" regular expression.
-Wimplicit-fallthrough=3 case sensitively matches a wide range of regular
expressions, listed in the GCC manual. E.g., all of these are accepted:
/* Falls through. */
/* fall-thru */
/* Else falls through. */
/* FALLTHRU */
/* ... falls through ... */
etc.
-Wimplicit-fallthrough=4 also, case sensitively matches a range of regular
expressions but is much more strict than level =3.
-Wimplicit-fallthrough=5 doesn't recognize any comments.
Plenty of other compilers did not recognize the gcc comment convention,
and up until now the compiler warning for detecting unintended
fallthrough had to be suppressed on other compilers. That's because the code
in NetHack has been relying on the gcc approach, and only the gcc approach.
The C23 standard introduces an attribute [[fallthrough]] for the
functionality, when implicit fallthrough warnings have been enabled.
Several popular compilers already support that, or a very similar attribute
style approach, today, even ahead of their C23 support:
C compiler whitelist approach
--------------------------- -------------------------------------
C23 conforming compilers [[fallthrough]]
clang versions supporting
standards prior to
C23 __attribute__((__fallthrough__))
Microsoft Visual Studio
since VS 2022 17.4.
The warning C5262 controls
whether the implict
fallthrough is detected and
warned about with
/std:clatest. [[fallthrough]]
This adds support to NetHack for the attribute approach by inserting a
macro FALLTHROUGH to the existing cases that require white-listing, so
other compilers can analyze things too.
The definition of the FALLTHROUGH macro is controlled in include/tradstdc.h.
The gcc comment approach has also been left in place at this time.
GitHub issue #1315 points out that it is possible for
a downstream function to change an object's nobj field
to point to a completely different chain.
The cited example by @vultur-cadens was:
for (obj = gi.invent; obj; obj = obj->nobj)
if (obj->oclass != COIN_CLASS && !obj->cursed && !rn2(5)) {
curse(obj);
++buc_changed;
}
curse() drops the weapon with drop_uswapwep(),
which calls dropx(),
which calls dropy(),
which calls dropz(),
which calls place_object().
place_object alters the nobj pointer, to point to the floor chain:
otmp->nobj = fobj;
fobj = otmp;
The result was that the next loop iteration was then using floor
objects from the floor chain.
This alters several for-loops to use a more consistent approach,
particularly when the obj is being handed off to a function,
where a downstream function might, or might not, alter the nobj
field.
References:
https://github.com/NetHack/NetHack/issues/1315https://www.reddit.com/r/nethack/comments/1gkc9ub/even_if_you_drop_an_item_before_drinking_from_the/
Changes to setuhpmax() a couple of days ago to deal with sanity_check
for "current hero health as monster better than maximum" ended up
triggering sanity_check about "current hero health better than maximum"
when gaining experience level(s) while polymorphed.
Issue reported by ars3niy: non-fireproof water walking boots are
supposed to be destroyed if worn on lava, but a post-3.6 change
made that only happen if the hero died and left bones.
The boots remained intact if hero was fire resistant or survived
6d6 damage. Staying intact should only happen if they're fireproof.
This seems to work but each time lava_effects() gets modified it
becomes more fragile. Having deleted objects stick around doesn't
help with this problem, which is to keep an item which is being
stolen--and whose loss causes the hero to drop into lava--from
being burned up before being transferred to the thief's inventory.
Fixes#1291
The g? structs had a mix of variables that were written to
the savefile, and those that were not.
For better clarity and to distinguish those that end up in
the savefile, relocate some g? variables that get written
directly to the savefile into different structs.
This updates EDITLEVEL, although technically it probably
didn't need to, since savefile contents are not changing.
Details:
gb.bases -> svb.bases
gb.bbubbles -> svb.bbubbles
gb.branches -> svb.branches
gc.context -> svc.context
gd.disco -> svd.disco
gd.dndest -> svd.dndest
gd.doors -> svd.doors
gd.doors_alloc -> svd.doors_alloc
gd.dungeon_topology -> svd.dungeon_topology
gd.dungeons -> svd.dungeons
ge.exclusion_zones -> sve.exclusion_zones
gh.hackpid -> svh.hackpid
gi.inv_pos -> svi.inv_pos
gk.killer -> svk.killer
gl.lastseentyp -> svl.lastseentyp
gl.level -> svl.level
gl.level_info -> svl.level_info
gm.mapseenchn -> svm.mapseenchn
gm.moves -> svm.moves
gm.mvitals -> svm.mvitals
gn.n_dgns -> svn.n_dgns
gn.n_regions -> svn.n_regions
gn.nroom -> svn.nroom
go.oracle_cnt -> svo.oracle_cnt
gp.pl_character -> svp.pl_character
gp.pl_fruit -> svp.pl_fruit
gp.plname -> svp.plname
gp.program_state -> svp.program_state
gq.quest_status -> svq.quest_status
gr.rooms -> svr.rooms
gs.sp_levchn -> svs.sp_levchn
gs.spl_book -> svs.spl_book
gt.timer_id -> svt.timer_id
gt.tune -> svt.tune
gu.updest -> svu.updest
gx.xmax -> svx.xmax
gx.xmin -> svx.xmin
gy.ymax -> svy.ymax
gy.ymin -> svy.ymin
Related note:
There are some pointer variables that are heads of chains that were not
moved from 'g?' to 'sv?', because they are not actually written to the
savefile directly, but the objects/monst/trap/lightsource/timer in the
chains they point to are. That can be changed, if desired.
Examples: gi.invent, gm.migrating_objs, gb.billobjs, gm.migrating_mons,
gf.ftrap, gl.light_base, gt.timer_base
Replace several upstart(y_monnam(mon)) with new YMonnam(mon) to
produce "Your little dog" and such.
Also change one or two Monnam(mon) to YMonnam(mon) and one pline(...)
to pline_mon(mon, ...).
Add a theme room with multiple visible teleportation traps
which will always teleport to specific locations in the same level.
Teleport trap change from xNetHack by copperwater <aosdict@gmail.com>.
Containers can't become fireproofed so the line of code in
fire_damage() which tested for that led to confusion.
Also, add missing handling for statues as containers.
Landmine blew up, scatter exploded a potion of oil, which melted
the ice on which the landmine was, resulting in the landmine trap
being deleted. The code then tried to access it to make a pit.
Issue reported by Umbire: if hero dies by drowning on the Plane
of Water, cause of death was reported as "drowning in a limitless
water".
Reported for livelogging but applied to tombstone and logfile too.
Omit the article "a" in this situation.
For 3.6.7, it would have started as "drowning in a water" and been
updated on the fly to be "drowning in deep water". 3.7 changed
terrain type WATER to be "wall of water", where "a" is expected,
and also added "limitless water" for Plane of Water, but it was
neglecting to include a similar fixup for the latter. The "deep
water" fixup is still present but doesn't get triggered anymore.
Fixes#1248
In 3.6.2 parts of the wakeup code were merged together, and this
caused pets consider any noise made by the hero - such as hitting
iron bars or digging - as whistling for them to come to the hero.
Change it to only consider actual whistling and ringing a bell.
Reported directly to devteam: if a magic trap gave its uncurse
effect, scroll of remove curse could become discovered.
Turns out that it would happen if hero was wielding a stack of
unholy water potions. It didn't matter whether they were known
as water or known to be cursed or whether hero was carrying any
scrolls of remove curse.
The recent "trap.c reformatting" commit included a non-formatting
change switching from gb.bhitpos.x,.y to local x,y in the rolling
boudler trap routine. The part of that routine used when a rolling
boulder hits another boulder and transfers its remaining momentum
to that other one got switched to wrong x,y and the first boulder
basically kept going, possibly hitting itself at each new position.
Note: Original change is from xNetHack by copperwater <aosdict@gmail.com>,
but this commit comes from HACKEM-MUCHE by Erik Lunna, with
some minor code formatting.
From xNetHack commit a0a6103bea:
'The original goal: nerf item destruction using a method I initially
proposed for SpliceHack, in which the number of items subject to
damage from any single source is limited by the amount of damage the
effect caused. The intent was to be more fair all around and prevent
aggravating situations where, for instance, a chest shock trap zaps
you for 4 damage and immediately ten of your rings and wands blow up.
Problem 1: no easy way to limit the items destroyed without biasing
heavily towards the start of the invent chain. The old code was able
to get away without bias by just indiscriminately destroying
everything eligible with a 1/3 chance. Here, I had to introduce
reservoir sampling in a somewhat more complex form than I've applied
it elsewhere, since there are a pool of potential items.
Problem 2: destroy_item no longer worked remotely like destroy_mitem,
which still destroyed 1/3 of items indiscriminately. Commence the
process of squishing them into one function that handles both the
player and monsters. (Which required making a lot of adjustments to
destroy_one_item, now named maybe_destroy_item, on nits such as
messaging and when to negate damage. An annoying consequence of the
merge is that in the player case, their HP is deducted and they can
be killed directly, but for monsters they need to add up the
destruction damage and return it.)
Unifying destroy_item and destroy_mitem has some advantages: in
addition to the obvious code duplication removal, it ensures monsters
now take the same damage as players for destruction (previously they
took a piddly 1 damage per destroyed item). Now when you hit
something with Mjollnir and their coveted wand of death breaks apart
and explodes, you at least get the satisfaction of knowing they took
the standard amount of damage from it. Monsters also now get
symmetry with players in having extrinsic elemental resistance
protect them from item destruction, and damage negation from item
destruction if they were appropriately resistant.
Problem 3: a lot of callers didn't preserve the "amount of incoming
damage" that this refactor relies on. E.g. if the defender resisted
that element, the local dmg variable would be set to 0. So I had to
do some wrangling with callers to save that original damage
value. The rule of thumb is: all *incoming* damage counts. So that
includes the player's spellcasting bonus if applicable, but not
things like half damage, negation due to resistance, or extra damage
due to being vulnerable to cold/fire.
Then I figured, while I'm here let's get rid of all those silly cases
where destroy_items is called multiple times for various different
object classes, and cut the object class parameter out of it. This
has a few minor effects:
- Places where different object classes previously rolled
independently for destruction to happen at all now roll
once. (Which, by my calculation, generally means less incidences of
destruction - a fire attack now won't have three separate chances
to hit your scrolls, potions, and spellbooks. On the flip side, a
lucky roll will no longer save an entire object class in your
inventory.)
- Callers can no longer specify different probabilities for
destroying different object classes. The only place this was really
used was to call destroy_item with a slightly lower probability on
SPBOOK_CLASS. With the nerf in this commit, less of them ought to
be destroyed anyway.
- A very edge case of where explosion-vs-monster damage was totted up
differently for golems, which could result in differences of a hit
point here or there.
- All object classes being processed in one go means that less items
are destroyed than would be if they were still processed
independently. This is not really visible compared to the old
baseline of just destroying 33% of everything, but would be a
marked difference versus a copy of the game that still called
destroy_items separately for different object classes. To
compensate, I adjusted my planned damage-to-destruction-limit
scaling factor down from 8 to 5.
Not done: merging in ignite_items(), though that would probably be
really easy now.'
Notes from porting from xNetHack:
- It might be necessary to reexamine at all the conditional checks for
calling destroy_items. Because item destruction is much more
restrained and uses the actual damage from an effect, we might now
need to check 'if (!rn2(3))' and similar in all the places item
destruction occurs.
Monstly changing '<type>* <var>' to '<type> *<var>'. Wrap or shorten
a few wide lines.
One of the rolling boulder routines accessed gb.bhitpos.x,y many times;
change that to use local x,y instead.
If a trap is blocking levitation or flight and you use wand of opening
or spell of knock to get out, you'd get things like
|You start to float in the air!
|You are released from the bear trap.
Give the release message before the actual untrapping.
lava_effects() item destrunction had the logic for handling Book of
the Dead wrong. (However, that didn't matter since the obj_resists()
check earlier would prevent it from being burned up. Fix it anyway.)
Three months ago to prevent an "object lost" panic situation when
stealing an item that let hero survive water (several candidates)
would result in drowning, remove_worn_item() was changed to flag the
item being removed as in_use and emergency_disrobe() was changed to
avoid dropping in_use items while drowning. That seemed to work ok.
But for lava instead of water, in_use is a flag to destroy the item
(set in advance, before issuing messages that can give the player a
chance to trigger a hangup save). So instead of keeping the item
around for theft to finish, it was deallocating it. steal() would
format the freed object and then access some of its fields, leading
to havoc.
This adds a hack to allow one item already flagged as in_use to be
treated differently by lava_effects() from the ones it flags for
destruction. This also seems to work ok, but we may need to start
putting freed items on a deferred deallocation list similar to how
dead monsters are kept around for the rest of the current move.
The fix/hack has revealed two more bugs that this doesn't address.
An item being stolen is removed without any message, then if that
removal doesn't kill the hero a theft message is given. The message
sequencing is wrong. Flying hero who loses amulet of flying just
gets affected by lava; player is only told why after life saving.
The other issue is that life-saving from lava can teleport the hero
to where the thief can no longer be seen, yielding "It steals <item>"
even though "It" was visible when the theft started.
Some minor cleanup of artifacts from the splitting up of trap effects
into the various trapeffect_foo functions: consolidate redundant
variables in trapeffect_pit (tt vs ttype), and simplify the definition
of 'inescapable' traps a couple functions (since the functions are now
specialized to a particular trap, it's unnecessary to check ttype
there in determining whether it's an inescapable Sokoban pit or hole).
Issue reported by Umbire: suggestion to always destroy adjacent webs
via 'F'<dir> if wielding Sting or Fire Brand.
Sting already did that; this adds Fire Brand.
This also augments the #untrap command when wielding either of those,
or any other blade. And rephrases successful untrap message
"You remove {the or your} {bear trap or webbing} from Fido." to
"You extract Fido from {the or your} {bear trap or web}." since the
trap remains intact.
Forcefight and #untrap against webs ought to be reconciled to remove
[some of] their differences and/or share code. But not by me...
Closes#1201
Do isok check apriori to accessing the levl array to filter
out dx/dy that have grown too large.
This fixes:
trap.c:3455:19: runtime error: index 80 out of bounds for type 'rm [80][21]'
Adds a new boolean option, accessiblemsg. If on, some game messages
are prefixed with direction or location information, for example:
(west): The newt bites!
(northwest): You find a hidden door.
I added the info to the most common messages, but several are
still missing it.
Two variations:
IndexOk(idx, array) validate that idx is a valid index into the array
IndexOkT(idx, array) validate that idx is a valid index into the
array, excluding the final Terminator element
