The devteam feedback was to place casts in the code
in question.
This puts explicit casts on some code that was being
compiled into 'int64' then stuffed into smaller types with
VC2005.
I'm pretty sure that some pre-standard compilers don't know how to
apply an initializer to a variable of type union. Unfortunately, I don't
have access to one to check. Fortunately, there's no need to explicitly
initialize `zeroany' since the default value is what we want--the first
field will be set to zero or null as appropriate (null in this case).
Strictly speaking, this isn't adequate; what if long is wider than a
pointer rather than narrower? Using `= {DUMMY}' didn't handle that case
either; the ordering of the union's fields controls which bits get stored.
As a practical matter, it should make no difference. As long as the code
reading a union accesses the same field as the code writing that union set
up in it, anything in extraneous bits should be irrelevant--except perhaps
when a debugger tries to format things. The whole issue has always been
implicitly based on the assumption that null pointers have all bits zero
in the first place; that's typical but not guaranteed.
botl.c conversions. All the ports seem to be using genl_status_update(),
rather than a window port specific version, so botl.c was the only place
this had to be adjusted.
Also a uudecode cast for the result of strlen, since it isn't using
config.h
Remove some more code that forced pointers into a long int, and
vice versa where information could be lost (P64 platforms such as
WIN64 have a 64 bit pointer size, but a 32 bit long size.)
This 3rd part deals with region functions switching
some arguments from type genericptr_t to 'anything'.
Like the previous 2 parts, this needs to increment
EDITLEVEL in patchlevel.h.
Remove some more code that forced pointers into a long int, and
vice versa where information could be lost (P64 platforms such as
WIN64 have a 64 bit pointer size, but a 32 bit long size.)
This 2nd part deals with timeout functions switching
some arguments from type genericptr_t to 'anything'.
Like part 1, this needs to increment EDITLEVEL in patchlevel.h.
[the problem in the earlier rev was tracked to cleanup_burn(),
where arg was holding a (genericptr_t) timer id, and
passed directly to del_light_source() as is.]
P64 (Win64) has a 64 bit pointer size, but a 32 bit long size.
Remove some code that forced pointers into a long int, and
vice versa where information could be lost.
This part deals with light source functions and their
arguments mostly, and switches some arguments
from type genericptr_t to 'anything'.
P64 (Win64) has a 64 bit pointer size, but a 32 bit long size.
Remove some code that forced pointers into a long int, and
vice versa where information could be lost.
This part deals with light source functions and their
arguments mostly, and switches some arguments
from type genericptr_t to 'anything'.
Hide pointer formatting in alloc.c by eliminating the need for callers
to know how big a buffer is required. I generally prefer the caller to
pass in its own buffer for this sort of thing, but in this case the usage
is almost entirely for debugging so using static buffers results in less
clutter in the rest of the code.
> Michael Allison wrote:
> There are unresolved functions in the trunk if you build without WIZARD
> defined and attempt to link:
Also to a lesser extent, in the 3.4.4 branch.
<Someone> wrote on Tuesday, July 27, 2004 at 06:46:15
> In the region.c function rest_regions allocates storage for the possible
> enter_msg and leave_msg strings. But the function free_region does not relese
> this storage.
Also ensures that some code that is currently ifdef'd out
makes copies of the strings into memory from alloc()
to ensure that no problems with free() result if the function
gets passed a literal string.
<Someone> wrote on Tuesday, July 27, 2004 at 06:46:15
> In the region.c function rest_regions allocates storage for the possible
> enter_msg and leave_msg strings. But the function free_region does not relese
> this storage.
Also ensures that some code that is currently ifdef'd out
makes copies of the strings into memory from alloc()
to ensure that no problems with free() result if the function
gets passed a literal string.
> NetHack feedback form submitted by
> <email deleted> on Friday, June 30, 2006 at 17:31:12
> ---------------------------------------------------------------------------
>
> mailversion:1.35
>
> nhversion:3.4.3
>
> nhfrom:Our 3.4.3 source release, unmodified
> comments:
> telnet nethack.alt.org with the terminal set to 21 rows.
> Choose to pick a char, not accept pot luck, and game segfaults.
> (same happens from linux console)
I was able to reproduce something similar in win32 by setting
the console to 21 rows. As he stated, don't let the game pick you
character for you to reproduce the problem. As soon as I chose
Archeologist the problem occurred:
Where:
In hack.c, weight_cap()
if (Levitation || Is_airlevel(&u.uz) /* <email deleted> */
#ifdef STEED
|| (u.usteed && strongmonst(u.usteed->data))
#endif
)
Variables:
carrcap 200
u.usteed 0x00000000
&u.uz 0x005e54aa
youmonst.data 0x00000000
Examination of the preprocessor output of that section
of code reveals that
"Levitation" becomes:
(u.uprops[47].intrinsic || u.uprops[47].extrinsic ||
((youmonst.data)->mlet == 5))
so it is the is_floater(youmonst.data) causing the crash.
Call stack:
weight_cap() line 2300 + 24 bytes
inv_weight() line 2342 + 5 bytes
calc_capacity(int 0) line 2354 + 5 bytes
near_capacity() line 2365 + 7 bytes
bot() line 607 + 5 bytes
docorner(int 47, int 19) line 2378
erase_menu_or_text(int 5, WinDesc * 0x00a22550, char 0) line 994 + 25 bytes
tty_dismiss_nhwindow(int 5) line 1664 + 15 bytes
tty_select_menu(int 5, int 1, mi * * 0x0006fc40) line 2248 + 9 bytes
tty_player_selection() line 442 + 16 bytes
pcmain(int 3, char * * 0x00a20eb0) line 457
main(int 3, char * * 0x00a20eb0) line 91 + 13 bytes
This adds a check for a valid youmonst.data in
bot().
The latest Micrsoft compilers complain when a function is
assigned to a function pointer, and the function's argument
list does not match the prototype precisely.
It was evem complaining about the difference between this:
int x()
{
[...]
}
and a prototype of
int x(void);
when assigning that function's address to a function pointer.
This quiets those warnings, without suppressing the mismatch
check altogether for more serious mismatches.
From a bug report, attempting to attack
a hidden monster who is revealed by ongoing monster detection (blessed
potion or skilled spell) gave "wait! there's a <monster> hiding there"
response and prevented the attack. Make it behave the same as when the
hidden monster is revealed by telepathy; the monster comes out of hiding
and the hero's attack proceeds.
The recent fix for "breaking glass wand in tool shop" looked suspect,
adding a call to costly_alteration after an existing call to stolen_value.
Either one or the other ought to suffice. (For items on the floor,
costly_alteration() calls stolen_value(); for items in inventory, or just
released from inventory and not placed on floor yet, costly_alteration()
adds a usage fee to the shop bill but doesn't annoy the shopkeeper into
adding surcharges to prices or summoning the kops if already hostile.)
In 3.4.3, stolen_value() wasn't smart enough to charge for an out-of-
shk's-field item (like a wand in a tool shop) taken from a shop container,
and that's the problem the user was reporting. But the post-3.4.3 code was
changed to handle that by checking billable() instead of saleable(); this
bug should have been gone. Unfortunately, billable() treats items already
on the bill as not interesting--from the perspective of adding things to
the bill--so the change accidentally resulted in stolen_value() no longer
handling objects which are marked unpaid, triggering the same symptom for
a different reason. (Other events besides the breakage of thrown objects
suffered from the bug's new incarnation since various places deliberately
call stolen_value() for unpaid objects.) This updates stolen_value() and
stolen_container() to account for the behavior of billable(). And a few
calls to subfrombill() go away since stolen_value() now takes care of that.
Bug in #tip handling for horn of plenty. Emptying one while levitating
would trigger an "obj not free" panic by flooreffects() due to following
hitfloor() with redundant/inappropriate dropy().
<Someone> reported that he applied an unID'd bag and it became
discovered as a bag of tricks even though a spellbook appeared on the floor
next to him rather than having a monster show up (the monster was a mimic).
Suppress the bag discovery unless you can see or sense a monster appear.
(This doesn't really achieve much for most players, who'll recognize the
bag because they know that only one type of container doesn't prompt to
take things out and/or put things in, but I think it does make sense.)
While mucking with bag of tricks I decided that to be consistent with
the behavior of other containers, the #tip command should release all the
monsters in the bag instead of just one.
And after doing that, I realized that horn of plenty ought to behave
much the same, so #tip will operate on it now. However, it won't be listed
as a likely candidate in the "which item?" prompt unless/until it has been
discovered. (Attempting to empty any other type of horn yields "nothing
happens", same as for a horn of plenty with no charges left.) Emptying a
horn of plenty in a shop can be extremely verbose, but I don't think that
qualifies as a bug and don't currently have any plans to alter it.
From a bug report: [ slashem-Bugs-1206099 ] Torches are not extinguished with rust traps).
A rust trap that hits the torso candles causes all lit objects being carried
to be extinguished, but one which hit the weapon arm didn't have same affect
on a wielded light. This fix causes wielded candles or lamps (not Sunsword)
to go out if affected by any rust, such as hitting a rust monster with one,
rather than use his patch that just handled the trap case. It also excludes
wielded lights from the existing torso splash since having them be hit by
both instances is too obviously buggy.
I think brass lanterns ought to be exempt from being extinguished by
water (at least splashing which is less drastic than total submersion) since
there are references to them operating by batteries rather than fire, but I
didn't want to track all the places which would be affected.
<email deleted> wrote:
> - when in a hardware store, I put a glass wand out of a sack (the glass wand
> will cost you 266 zorkmids) and threw it in the shop => shattered into a
> thousand pieces BUT if I try to pay, I do not owe the shopkeeper anything !!!
> If I break a potion with a /oS, I have to pay !
Some recent code shuffling introduced an unintended change in behavior
(not observable to the player; just unnecessary deletion and re-creation of
light source with identical radius when polymorphing from one light emitting
form to another). The fixup for light range 1 would need to be repeated for
`old_light' when outside the `if (old_light != new_light)' block; move it
back inside where that isn't required. Also, youmonst.data is valid all the
time so a couple of `Upolyd' tests in the surrounding code can be dropped.
- remove an unreferenced variable
- continue with recent code trend towards having DEADMONSTER()
check in its own if/continue statement in a few more places
Several polymorph tweaks, most dealing with specifying form under
polymorph control or for wizard #polyself:
1) allow "were<critter>" and "human were<critter>" for your type of
<critter> when you're inflicted with lycanthropy; now you'll toggle
shape rather than be told "you cannot polymorph into that".
2) allow your own role; now you'll become a new man (or whatever race)
rather than get "you can't".
3) allow "human" to force a new man (or whatever) regardless of race.
No change for human characters, but elves, dwarves, and such can now
use either their own race or "human". (They never become humans.)
4) for wizard #polyself only, override the 20% chance of becoming a new
man instead of taking on the selected form. (This implicitly prevents
the annoying "your new form isn't healthy enough to survive" death
since your experience level won't drop below 1.)
5) remove a redundant drowning check in polyself(); it's already handled
in polymon() and polyman(for newman()) via spoteffects().
This also gets rid of an old use of 0 as not-a-valid-monster (not
responisble for any bugs though since giant ants aren't lycanthropes).
The user (<email deleted>) who recently suggested a
dump command for containers also wanted atmospheric sounds on levels which
have altars. Right now we'd have to find unattended altars the hard way
(by scanning the entire level) but we could add a counter (or set of
counters, one per alignment) like for fountains and sinks if we really
wanted to do that. [Now that I think about it, the #overview patch may
have already done something of the sort.] But what noises would an altar
be expected to produce? This only adds sounds for temples, where the
attending priest can be the source of the noise.
I'm not real thrilled with the initial set of sounds, particularly
the hallucinating one, but the implementation works. The "carcass" one is
a little clumsy; it's intended to add a hint for new players who haven't
figured out what the #offer command does.
Saving the game while punished, not carrying the attached ball,
and while swallowed by a purple worm resulted in losing the
ball and chain.
Since the required information was not being written to the
save file at all, I couldn't come up with a clean way to do this
for the branch, and preserve save file format. I could think
of lots of kludgy ways to do it (insert ball and chain into
the hero's inventory prior to saving, and remove it on restore, etc.)
Saving the game while punished, not carrying the attached ball,
and while swallowed by a purple worm resulted in losing the
ball and chain.
Since the required information was not being written to the
save file at all, I couldn't come up with a clean way to do this
for the branch, and preserve save file format. I could think
of lots of kludgy ways to do it (insert ball and chain into
the hero's inventory prior to saving, and remove it on restore, etc.)
When testing the spoteffects/drown hack I noticed that draining myself
with Stormbringer (toss up, get hit on head) while in iron golem form gave
messages about it drawing or draining life. (I'm sure that this has come
up before....) I've altered the artifact hit message rather than making
golems become resistant, although the opposite approach seems at least as
valid. The drain life spell uses different wording and isn't affected.
Attempt to fix a buglist item: if hero poly'd into iron golem form
enters a pool of water and drowning triggers reversion to human/whatever
form due to water damage, he will fall into that pool again, crawl or
teleport out, then redundantly crawl or teleport out for the initial entry.
[ spoteffects -> drown -> losehp -> rehumanize -> polyman -> spoteffects
-> drown ]
I don't have a lot of confidence in this fix. It does handle the
reported problem, and hasn't broken a couple of earlier tricky cases
(ice melting to water, land mine turning into a pit). But drown() and
lava_effects() seem to leave a trail of special case handling wherever
they appear so they--or spoteffects, or both--ought to be redone somehow.
I can't find the report about this; I must have deleted it after
reading, or else recently reread something so old that I'm not going back
far enough now. When you perform the invocation ritual to create the
stairs down to Moloch's Sanctum, any trap at your location gets deleted.
But if you were in a trapped state at the time then you got left in that
state. Descending stairs doesn't check for traps so you wouldn't notice
unless you tried to move around on the same level first. Then you'd get
"you're stuck in a pit/beartrap/web" even though it wasn't there anymore.
While testing the hiding vs traps patch, I became a mimic and hid.
It gets stuck to you.
Huh? Nothing was visible; nothing became visible (aside from the ']' at my
position changing back to 'm'). Display the invisible monster glyph when
an unseen monster bumps into you while you're hiding or mimicking gold.
That's usually handled by the hit and miss routines, but they aren't used
when you're just brought out of hiding instead of attacked.
From a bug report, the game gave feedback
about a monster becoming stuck in a web but there seemed to be no monster
around because it immediately began hiding under an object at the web's
location. Prevent monsters--or poly'd hero--from hiding when trapped in
anything other than a pit or spiked pit. Also, prevent them from hiding if
they're holding you or you're poly'd and holding them. I'm not sure whether
either of those cases ever actually happened but big mimics are capable of
both hiding and grabbing on.
