4.0.2 Bugs

[Nick]

Crash bug in 4.0.2 windows version.

Digging through rubble. Light goes out the moment the digging is done.
Edit to add: Light source was a torch. Auto-pickup is on.

Assertion failed!
Program: ...
File: obj-pile.c
Line: 134
Expression: pile_contains(*pile,obj)

Last messages:

Your light is growing faint.
You dig in the rubble. <17x>
You have removed the rubble.
Your light has gone out!

I can neither reproduce this, or see why the code here should be failing. Any clues, or a savefile if it happens again, would be welcome.

[Nick]

My variant just crashed while processing objects. Found out the following loop is faulty:

	while (obj) {
		struct object *next = obj->next; [B][I]<-- crashes here[/I][/B]

		/* do something */

		/* Delete the object */
		object_delete(&obj);
		obj = next;
	}

When leaving the loop, "next" being local to the loop is freed, "obj" points to nowhere...

As redlumf points out upthread, there is nothing wrong with this loop as such. My first guess for what would have caused a crash is that something else corrupted the object structure, so that obj->next became an illegal access.

[AnonymousHero]

As redlumf points out upthread, there is nothing wrong with this loop as such. My first guess for what would have caused a crash is that something else corrupted the object structure, so that obj->next became an illegal access.

Since the API is already such that it can, I think object_delete should actually set the pointed-to-pointer to NULL[1]. In my experience it's a much more reliable way of causing an immediate bad access to fail rather than relying on something like ASAN/UBSAN/&c. Obviously, I realize that either is undefined behavior, but in practice NULL access is pretty reliable at causing an immediate failure as long as the compiler doesn't optimize it away[2].

[1] AFAICT from current master, it doesn't.

[2] There are actually some interesting cases of this in the Linux kernel which have caused security issues, but I don't think it's particularly pertinent to Angband, though it can obviously still happen and obscure other errors when using optimizations. The compiler is quite unlikely to be able to "peer through" object_delete

[Timo Pietilä]

I can neither reproduce this, or see why the code here should be failing. Any clues, or a savefile if it happens again, would be welcome.

I'm guessing that light actually goes out while digging, and it doesn't disturb the digging repeat, so immediately after digging is successful lights go out.

[nevermind, I just read the referred post and it talks about crash, not odd message sequence]

[PowerWyrm]

As redlumf points out upthread, there is nothing wrong with this loop as such. My first guess for what would have caused a crash is that something else corrupted the object structure, so that obj->next became an illegal access.

Yeah my bad. Didn't reproduce the crash either after having fixed two occurences of object_delete called on inventory objects without doing a gear_excise first.

[PowerWyrm]

Just now I got a message with no subject, that said:

makes a soft 'pop'.

I'm used to it saying "Something makes a soft 'pop'," when it's about an unseen monster. I don't know much about programming- is this just a case of the message not having an assigned value because the monster is offscreen?

DLvl 39; it could have been any number of things teleporting/phasedooring as I approached a big, dark room. Seems minor, but thought I'd report it since I don't see anyone mentioning it yet.

The typo is in monster_spell.txt. There are other typos I'm currently reviewing, I'll note them all in a post.

[PowerWyrm]

The typo is in monster_spell.txt. There are other typos I'm currently reviewing, I'll note them all in a post.

And here we go:

- ARROW_4: "thwang" (no idea if it's correct, it's spelled "twang" everywhere else)
- BA_POIS: it was "creates a cloud of poison" in 3.5
- BA_WATE: it was "creates a whirlpool of water" in 3.5 (and "Something gurgles")
- CAUSE_2: missing punctuation (--> "points at you and curses horribly!")
- CAUSE_4: idem (2x "DIE!".)
- HASTE: "{name} mutters." (not when blind --> "Something mutters.")
- TPORT: missing "Something" ("Something makes a soft 'pop'.")
- S_KIN: "You hear something appear nearby!" (S_KIN is not limited to 1 --> "some things")

[Nick]

ARROW_4: "thwang" (no idea if it's correct, it's spelled "twang" everywhere else)

I think a thwang is louder than a twang

CAUSE_4: idem (2x "DIE!".)

This looks OK to me - 'DIE!' is what the monster is saying, and it's in quotes because it's being referred to by the message.

[Nick]

OK, I believe all outstanding bugs for 4.0.2 are either fixed (latest build will be up soon), or filed here. Note that I've confirmed the "too many Dracoliches" bug.

[Ingwe Ingweron]

OK, I believe all outstanding bugs for 4.0.2 are either fixed (latest build will be up soon), or filed here. Note that I've confirmed the "too many Dracoliches" bug.

Information leak bug. If @ quaffs an unidentified potion you can get the message something like, "Choose a direction..." Ah, that's !Dragon's Breath, but I shouldn't know that until after it's identified. And since I didn't choose a direction, it is just a "Light Brown Potion". Not sure of what the solution should be. I don't remember the problem existing in 3.5.1.

[Derakon]

Information leak bug. If @ quaffs an unidentified potion you can get the message something like, "Choose a direction..." Ah, that's !Dragon's Breath, but I shouldn't know that until after it's identified. And since I didn't choose a direction, it is just a "Light Brown Potion". Not sure of what the solution should be. I don't remember the problem existing in 3.5.1.

Pretty sure it existed in 3.5. Ditto with Identify scrolls/staves, which prompt for the item to ID.

[Nomad]

Information leak bug. If @ quaffs an unidentified potion you can get the message something like, "Choose a direction..." Ah, that's !Dragon's Breath, but I shouldn't know that until after it's identified. And since I didn't choose a direction, it is just a "Light Brown Potion". Not sure of what the solution should be. I don't remember the problem existing in 3.5.1.

I'd think the solution should be that the potion IDs even if you don't choose a direction. (Seems to me not choosing would be the equivalent of spitting the potion back into the bottle, and you'd probably notice if you'd taken a mouthful of Dragon's Breath...)

[Ingwe Ingweron]

Pretty sure it existed in 3.5. Ditto with Identify scrolls/staves, which prompt for the item to ID.

Except it didn't identify when the direction wasn't chosen. Nomad's solution sounds like a reasonable justification. The change needed would be for the potion to identify when a direction is not chosen, much like the identify scroll/staff/rod when no item is chosen.

[PowerWyrm]

The typo is in monster_spell.txt. There are other typos I'm currently reviewing, I'll note them all in a post.

Forgot:
- a missing period for "breathes gravity"
- "Something mumbles loudly." for BA_DARK (3.5)
- a missing period for "Something commands you to go away"

[Timo Pietilä]

OK, I believe all outstanding bugs for 4.0.2 are either fixed (latest build will be up soon), or filed here. Note that I've confirmed the "too many Dracoliches" bug.

What about nexus resist preventing you from using teleport level? That was in one of the nightlies. Is that fixed? Can't find it from http://trac.rephial.org/