Time  Nick       Message
00:03 MTDiscord  <jordan4ibanez> Introduces a lot more bugs than it solves, clamp to the max collideable nodes iirc is 3
00:04 MTDiscord  <jordan4ibanez> Also you can still access oob iirc with get_node
00:07 Guest54    jordan4ibanez what introduces more bugs than it solves, cleaning up the datatypes?
02:44 MTDiscord  <jordan4ibanez> Dang, I don't remember. I think it was something about null
06:59 nrz        Guest54, ok pathfinding with server side calculation, that leads to a suggestion show before where we send nodes on the path (and should resync every x seconds in case of drift)
10:34 Krock      will merge #13380 and #13337 in 15 minutes
10:34 ShadowBot  https://github.com/minetest/minetest/issues/13380 -- Limit formspec fields to 640K by SmallJoker
10:34 ShadowBot  https://github.com/minetest/minetest/issues/13337 -- Lua API: Catch serialization error for chat messages by SmallJoker
10:49 Krock      merging
12:51 Guest54    Krock I may be a bit late for this, but would https://github.com/minetest/minetest/pull/13337/files limit putting a large item in an inventory or so? i am asking because of nested shulkers.
12:54 Krock      Guest54: how come I have to answer this question twice?
12:54 Krock      they are not related whatsoever
12:54 Guest54    i must have forgotten
12:54 Guest54    sorry!
12:54 Guest54    thanks
12:55 Krock      also I think you meant the other PR
12:55 Guest54    i just woke up, so could be
12:56 Guest54    regarding the filtering of bad characters that luatic proposed for minetest_game btw, i recently talked with a coworker about something like this. my conclusion was: you need to filter dangerous ANSI escape sequence at the output, not the input.
12:56 Guest54    like, filter anything that messes with lua at the input, but filter anything that messes with your terminal at the output
12:57 Guest54    context was: the coworker found a terminal control character injection vulnerability in a somewhat-popular file transfer app and asked if filtering the input codepoints was enough.
12:59 Guest54    and in general, you need to act on the serialization of the output as bytes, given the encoding – not codepoints given the input encoding. otherwise an attacker might get around your filters.
13:00 Guest54    i read somewhere that people used to try XSS with “¼script¾alert(¢XSS¢)¼/script¾” because in 7-bit US-ASCII that became a “<script>alert("XSS")</script>” or so
13:01 Guest54    luk3yx if you have a list of all the things that mess with lua, please share
16:04 sfan5      has anyone tested #13374 yet
16:04 ShadowBot  https://github.com/minetest/minetest/issues/13374 -- [NO SQUASH] Particle spawner: Fix missing params by SmallJoker
16:04 sfan5      in particular protocol compat
16:20 Krock      well... I did.
16:21 Krock      of course, it would require another protocol bump if you want to have it compatible with 5.7.0-dev since last week
16:22 Guest54    how many protocol bumps can you afford anyways?
16:22 Krock      2¹6 - 42
16:22 Krock      * 2¹⁶ - 42
16:23 Guest54    65494
16:23 Guest54    given that some people run random dev versions, why not do it then?
16:23 Krock      can we please first focus on the actual issue first?
17:40 sfan5      we won't do another bump
17:40 sfan5      asking again because I don't remember: do we have a changelog written yet?
17:42 Desour     the thing on the wiki was written by Krock
17:42 Desour     changelog thing
17:42 Krock      it's outdated by a week but not much happened since then
17:43 sfan5      i see, great
18:00 Guest54    i think particles leak memory
18:04 Guest54    if i understand it correctly, in Particle::~Particle() in src/client/particles.cppit seems to delete only 104 bytes, but the allocated memory is 112 bytes
18:04 Guest54    has anyone else found something like this before?
18:09 Krock      I think the best strategy is to first find "new" keywords that effectively allocate stuff
18:09 Krock      or base classes that do not have a virtual destructor
18:10 Guest54    well you can use semgrep to do static analysis and fsanitize=address for dynamic one
18:10 Guest54    so far either i am getting a lot of false positives or people believe that new / new[] and delete / delete[] can be mixed and matched arbitrarily
18:11 Guest54    and i am pretty sure it's false positives ;)
18:11 Guest54    the particle thing is not though
18:11 Guest54    unless the missing bytes get lost somewhere else
18:32 Guest54    okay, it does not happen every time. i compiled minetest with ”cmake -DCMAKE_CXX_FLAGS='-fsanitize=address -g' .” and “make”
18:32 Guest54    then in devtest started spamming particles
18:33 Guest54    Krock here is the result: https://mister-muffin.de/p/jbgF.txt
18:37 Pexin      Krock: would it be appropriate to put a short desc in changelog about improved bouncy? (jump to increase height, sneak to decrease)  ..as I'm not sure the best way to inform players
18:38 Desour     if you find an issue, submit a fix, or describe the codepath that triggers the issue. otherwise this is just about how you're trying out a tool, and hence offtopic.
18:40 Guest54    Pexin isn't it the job of games to inform players about trampolines? you only need to inform game/mod devs
18:40 Krock      Guest54: from how it looks, this is actually is about freeing ClientTexture
18:41 Guest54    Krock so is this an issue or not? i.e. does every particle leak 8 bytes eventually?
18:43 Guest54    Desour i am not sure if “minetest crashes if you compile it with asan” is an “issue” worth of a fix, which is why i am asking people hopefully more knowledgeable than me.
18:43 Krock      this looks a bit suspicious https://github.com/minetest/minetest/blob/master/src/client/particles.cpp#L551
18:44 Krock      oh nvm
18:44 Krock      it's actually a constructor call
18:45 Guest54    the same thing *does* eventually happen with mcl2 when particles are disabled btw, but i am not sure how
18:45 Krock      I can't tell what's going on without digging deeper. not quite interested in that. feel free to open an issue
18:45 Guest54    well i take that as confirmation of “it's not nothing, at least at first glance”
18:45 Krock      Pexin: will update
18:46 Guest54    thanks Krock
18:48 Pexin      ty
18:55 Krock      okay. the changelog is up-to-date again
18:56 Desour     why is the particle stuff using decltype so much?!
18:57 Krock      it's also using enable_if and other high level c++ magic
18:57 Guest54    btw, the zstd corruption recently fixed in zstd 1.5.5 was introduced with zstd 1.5.0 – if there is a version macro, maybe someone can just blacklist them in cmake dependency declarations?
18:57 Krock      feel free to propose a PR
18:58 Desour     it's kinda weird that that big particle PR was merged in less than a year. it's so big and spaghettily
18:59 Krock      it took a year, though.
19:00 Desour     no full year
19:00 Desour     (I'm not proposing this as a requirement x))
19:12 Desour     looked a bit into the particle stuff. ~Particle deletes the tex in a ClientTexRef, which points to the tex member (no a ptr) of a ClientTexture. see also https://github.com/minetest/minetest/blob/9c9309cdbb053598aaf08506928a4824e78b4622/src/client/particles.cpp#L778
19:14 Krock      it's passed as reference
19:25 rubenwardy That particle PR is a good example of polishing a turd
19:58 sfan5      would you make fundamental changes to particles?
19:58 sfan5      not talking about rendering
20:04 Desour     I'm already doing a little cleanup now
21:30 Desour     #13394
21:30 ShadowBot  https://github.com/minetest/minetest/issues/13394 -- [NOSQUASH] Particle cleanup by Desour