Bugs in 4.2.0

[Nick]

Encountered a couple of bugs in a 4.2.0 game, played on a Windows 7 machine.

The first bug was encountered when targeting and then teleporting away Gothmog, and then casting the spell Single Combat on an enemy using a direction key (i.e. not targeting). This resulted in the error attached as angband_error1.jpg. After the game crashed and reloading the save I was in the Single Combat room with Gothmog, not the enemy I actually cast the spell on. Fortunately I was able to beat him, but it was definitely not a fight I was expecting.

Steps to reproduce
1. Target an enemy using * or '
2. Teleport away using the rod of teleport in inventory
3. Stand next to a different enemy and cast Single Combat, using an arrow key instead of * or '

Some time after this when I tried to book cast recall out of the dungeon, I get the error message attached in angband_error2.png (side note, why does the website upload the first as .jpg and the second as .png? They're both .pngs). I can't remember for sure but I think this was the first time I recalled since encountering the first bug, so the two might be related.

Steps to reproduce
1. Cast Recall from book

I've attached a savefile for convenience, both bugs are reproducible in it. I'd really appreciate it if the savefile could be fixed, hopefully the bugs aren't too hard to track down.

Thanks for the detailed report. I think I have an idea what caused it - Single Combat requires a targeted monster, and so it used the one you had previously targeted, which caused bad things to happen because he was out of LoS etc. I'll look at the savefile.

[DavidMedley]

Tasty Coins!

I used Vampire Strike on creeping adamantite coins. Yummy!

I also was surprised that it was resistant to acid. Seems like the ideal weapon to use against metal.

[Derakon]

Adamantite, and mithril, are both corrosion-proof. That holds for armor made of the metal just as much as it does for monsters made of it.

[backwardsEric]

witch/blackguard/slice of meat with original tiles

With 4.2.0 using the original tiles on Mac OS X, witches and blackguards are rendered as 'p's and slices of meat are rendered as ','s.

[backwardsEric]

memory leak selling item with brand, curse, or slay

Running 4.2.0 on a Mac and collecting a report of memory leaks from it with the Leaks instrument in XCode reported leaks of 32 or 240 bytes where the memory was allocated from this partial call stack (I've omitted the functions before process_command()):

mem_zalloc()
object_copy()
object_copy_amt()
store_sell()
store_menu_handle()
menu_select()
use_store()
event_signal()
move_player()
do_cmd_walk()
process_command()

There was also a leak of 32 bytes where the memory was allocated from this partial call stack:

mem_zalloc()
object_copy()
object_copy_amt()
do_cmd_stash()
process_command()

From the code, space may be allocated for slays/brands/curses in the temporary object used in either store_sell() or do_cmd_stash() but that space is not released before the temporary object falls out of scope. The attached file is a patch from git format-patch for store.c and ui-store.c to plug the leak.

object_info_ego() in obj-info.c and wiz_reroll_item() in wiz-debug.c also look like they could leak the memory for slays/brands/curses, but I didn't exercise those functions when profiling Angband with the Leaks instrument.

[Nick]

I've attached a savefile for convenience, both bugs are reproducible in it. I'd really appreciate it if the savefile could be fixed, hopefully the bugs aren't too hard to track down.

OK, fixed savefile attached. The saved town was the problem, so when you go back there you will find the layout different; the shop and your home should be unaffected. Let me know if you have any further problems.

[DavidMedley]

Adamantite, and mithril, are both corrosion-proof. That holds for armor made of the metal just as much as it does for monsters made of it.

Makes sense

[DavidMedley]

Are You Stoned?

When I'm hallucinating, if I do things like hit space, escape, browse a book, etc., no time passes but my hallucinations flicker. Shouldn't they only change when time passes? Pretty easy to determine whether something is actually there or not this way.

[scrarth]

OK, fixed savefile attached. The saved town was the problem, so when you go back there you will find the layout different; the shop and your home should be unaffected. Let me know if you have any further problems.

No luck on my end unfortunately, I still get the chunk_copy() level bounds failed error message.

[DavidMedley]

Adamantite, and mithril, are both corrosion-proof. That holds for armor made of the metal just as much as it does for monsters made of it.

Mithril Golem does not resist acid, however.

[Derakon]

Mithril Golem does not resist acid, however.

Good news! We're in the 4.2.0 bugs thread, so hopefully Nick will see this.

(They may have been made vulnerable so that mages would have some way to fight them...)

[Nick]

No luck on my end unfortunately, I still get the chunk_copy() level bounds failed error message.

Sorry, I attached the wrong one

This one should be OK - I recalled back to town from your save.

[scrarth]

Sorry, I attached the wrong one

This one should be OK - I recalled back to town from your save.

Yep, it's working fine now. Thanks a bunch for that!

[backwardsEric]

Other memory leaks

Other memory leaks reported by XCode's Leaks instrument for Angband 4.2.0:

1) Some "known" versions of objects allocated from the following partial call stacks below (omitting the calls before process_command()) were leaked.

mem_alloc()
mem_zalloc()
object_new()
object_split()
gear_object_for_use()
ranged_helper()
do_cmd_fire()
process_cmd()

mem_alloc()
mem_zalloc()
object_new()
object_see()
square_know_pile()
square_note_spot()
update_one()
update_view()
move_player()
do_cmd_walk()
process_command()

mem_alloc()
mem_zalloc()
object_new()
object_sense()
square_sense_pile()
effect_handler_SENSE_OBJECTS()
effect_do()
use_aux()
do_cmd_read_scroll()
process_command()

The attached file, object-absorb-known-object-leak-patch.txt, patches object_absorb(), but that doesn't eliminate all of those leaks.

2) An object allocated from this partial call stack,

mem_alloc()
mem_zalloc()
object_new()
make_object()
place_object()
alloc_object()
alloc_objects()
classic_gen()
cave_generate()
prepare_next_level()

, was leaked.

3) The string allocated from this partial call stack,

mem_alloc()
string_make()
cmd_set_arg_string()
cmd_get_string()
do_cmd_inscribe()
process_command()

was leaked when inscribing an object.

4) During character creation, there's a leak if the player edits the character background information. The partial call stack for where the allocations occur is

mem_realloc()
textblock_calculate_lines()
edit_text()
textui_do_birth()
play_game()

5) During character creation, the background or quest information can be leaked, especially if the player backtracks to select a different combination of race and class. The partial call stacks for where the allocations occurred were:

string_append()
player_generate()
do_cmd_choose_race()
process_command()

string_append()
player_generate()
do_cmd_choose_class()
process_command()

string_append()
player_generate()
do_cmd_birth_reset()
process_command()

string_append()
do_cmd_roll_stats()
process_command()

string_make()
do_cmd_choose_history()
process_command()

string_append()
player_generate()
do_cmd_birth_init()
process_command()

mem_zalloc()
player_quests_reset()
player_init()
do_cmd_birth_reset()
process_command()

string_make()
player_quests_reset()
player_init()
do_cmd_birth_reset()
process_command()

The attached file, player-birth-memory-leaks-patch.txt, has patches to plug those leaks and the leak in (4).

[backwardsEric]

File not closed in remove_old_dump() if error

In ui-prefs.c's remove_old_dump(), if there is an error opening the new file, the current file won't be closed. The attached patch fixes that.