00:08:10 -!- svendre_ is now known as svendre
00:34:26 <Cheibriados> Allies can't be spawned on traps 13https://crawl.develz.org/mantis/view.php?id=11883 by CanOfWorms
00:41:38 <Cheibriados> 03gammafunk02 07* 0.24-a0-116-gd3e027d: Don't let vampirism drain temporary monsters (blorx1) 10(85 minutes ago, 5 files, 36+ 14-) 13https://github.com/crawl/crawl/commit/d3e027da1558
00:41:38 <Cheibriados> 03gammafunk02 07* 0.24-a0-117-g53b5160: Tidy up a monster class method 10(37 minutes ago, 1 file, 18+ 6-) 13https://github.com/crawl/crawl/commit/53b51606c3e9
01:07:05 <Kelbell> Unstable branch on crawl.kelbi.org updated to: 0.24-a0-117-g53b51606c3 (34)
01:28:22 <Gretell> Unstable branch on crawl.develz.org updated to: 0.24-a0-117-g53b5160 (34)
01:59:37 <Gretell> Windows builds of master branch on crawl.develz.org updated to: 0.24-a0-117-g53b5160
03:00:05 <Gretell> Monster database of master branch on crawl.develz.org updated to: 0.24-a0-117-g53b5160
03:20:59 <Cheibriados> D:1 monsters differ between Webtiles and Windows offline tiles on the same seed 13https://crawl.develz.org/mantis/view.php?id=11884 by NormalPerson7
03:30:50 <Kelbell> Fork (bcrawl) on crawl.kelbi.org updated to: 0.23-a0-1729-gdd55c3d1e3
03:42:14 <Cheibriados> Ozocubu's Refrigeration's damage is one third of what it used to be 13https://crawl.develz.org/mantis/view.php?id=11885 by NormalPerson7
03:44:58 <Pinkbeast> FWIW it works here with pyyaml 3.11 (which dates from 2014, so I suspect your problem is not "not new enough")
03:45:06 <Pinkbeast> ECHAN
10:02:43 <Lasty> !tell gammafunk critical update: did a second gauntlet, had a great time, didn't die this time. I ended up approaching it pretty differently than last time, but still felt challenged. Mino was still scary.
10:02:44 <Sequell> Lasty: OK, I'll let gammafunk know.
12:27:16 <Henzell> Unstable branch on crawl.akrasiac.org updated to: 0.24-a0-117-g53b5160 (34)
13:02:50 <advil> great, the seeding bug NP7 just found seems to be specific to cross-compiled builds
13:02:58 <advil> that'll make it fun to track down
13:08:14 <advil> also, the very same seed demonstrates that any randomization in the des cache is potentially bad
13:13:33 <advil> so I guess I have two options: 1. protect every single random call in des with is_validating, or disable randomness entirely during validation
13:14:04 <advil> *or 2.
13:25:11 <gammafunk> does validation determine what tags the vault will have?
13:25:41 <advil> yes, but I've already dealt with all the tags case
13:25:43 <advil> s
13:25:44 <gammafunk> nice
13:26:21 <advil> I had thought other stuff wouldn't matter, but I think any additive tag potentially does; the one that comes up here is onia_arrival_zoo_common
13:26:25 <gammafunk> yeah seems like preventing randomization during validation would be ok if there's nothing about that necessary for setting up vault selection
13:26:37 <gammafunk> !vault onia_arrival_zoo_common
13:26:38 <Sequell> 1/1. https://github.com/crawl/crawl/blob/master/crawl-ref/source/dat/des/arrival/small.des#L2436
13:27:24 <gammafunk> right, that looks like typicall placement time selection stuff
13:27:24 <advil> if it rolls the 1/10 thing seeds that try to place it can be completely different
13:27:26 <gammafunk> which we definitely want to allow that randomization to happen at placement time
13:27:31 <gammafunk> but don't need it before then
13:27:33 <advil> yeah
13:28:02 <gammafunk> oh that's the infamous "Crawl" vault
13:28:06 <advil> yep
13:28:15 <advil> this is the seed that broke manman's streak, incidentally
13:28:54 <gammafunk> well in terms of arrival vault terrain it's fairly typical
13:29:07 <advil> NP7 noticed it's different on windows cross-compiling than on servers (or linux/os x generally I think, and on windows when I've built it myself)
13:29:09 <gammafunk> that vault is bad for other reasons, but obviously we need to allow the kind of randomization it's doing there
13:30:04 <advil> yeah, what I mean is that if it rolls "crawl" during des validation, the behavior when placing it is completely different regardless of the reroll
13:30:20 <advil> well, maybe not
13:30:21 <advil> hm
13:30:35 <advil> I might have tested for this wrong
13:31:10 <advil> anyways, I still have no clue why the cross-compiled builds do something different
13:35:30 <advil> ok yeah, this des cache randomization worry was a false alarm, I was just making a stupid mistake in my testing
16:28:25 <gammafunk> advil: not sure if you've seen this, but this is an amazing bug: https://cdn.discordapp.com/attachments/325651923514556417/546805049070845993/unknown.png
16:46:53 <alexjurkiewicz> WHAT
16:53:02 <Rast2-> I'm OK with that
16:59:20 <advil> hah yep
16:59:42 <advil> it's actually been around for a little while
17:01:28 <advil> https://crawl.develz.org/mantis/view.php?id=11615
17:02:02 <advil> I was able to replicate it at one point, but then I took a break from crawl for a few months and haven't gotten back to it
17:02:26 <advil> yeah, at least I put some replication instructions into that issue
17:02:40 <gammafunk> gonna form part of the basis of my upcoming crawl-integrated twitch bot
17:03:12 <gammafunk> donate bits to send buff or debuffs to the player, debuffs include god wrath, xom effects, zot trap effects, forced ascii mode
17:03:34 <Pinkbeast> Thank you, SLEX
17:04:07 <gammafunk> they actually have something like said bot for super mario world and some other classic console roms
17:05:23 <alexjurkiewicz> advil: I think I understand what is causing some identical qw runs to diverge. Shaft traps can change the order of generation
17:05:27 <advil> ah, that'll do it
17:05:59 <advil> qw got more deterministic with these changes but there probably are still things that draw on the gameply rng that aren't stable
17:06:33 <alexjurkiewicz> I wonder if traps could use the some sort of consistent hashing so that the squares which are trapped can become deterministic
17:07:07 <alexjurkiewicz> ...I half said "like magic mapping" and half left the question open
17:07:41 <advil> it's doable, I'm not actually sure I would want that to be stable across instances of a seed though
17:09:35 -!- slphil1 is now known as slphil
17:09:55 <advil> well, having randtrap location be stable if the sequence of actions is identical seems reasonable (I just know I haven't achieved that because I've ignored trying to work on random calls in things that don't have to do with levelgen)
17:11:13 <alexjurkiewicz> yeah. It took me a while for "trap reform" to connect with "rng" while staring at these morgues
17:11:39 <alexjurkiewicz> the good news is I ran qw across seeds 1-1000 and the win rate was identical despite this. It's just the specific won games changed :D
17:11:53 <alexjurkiewicz> "score_mean": 362420,
17:11:55 <alexjurkiewicz> "score_median": 86151,
17:11:57 <alexjurkiewicz> vs
17:11:59 <alexjurkiewicz> "score_mean": 362251,
17:12:01 <alexjurkiewicz> "score_median": 87579,
17:12:22 <advil> if you run it with pregen, that should avoid all of this (and take not much more time in aggregate)
17:14:39 <advil> what was most impressive to me about the xp bug is how *good* the map display looks in that popup
17:15:04 <advil> compared to glyphs mode
17:21:39 <alexjurkiewicz> average run takes about 6mins, so pregen would slow down tests by a reasonable amount sadly
17:27:50 <advil> well, it's just frontloading levelgen that would mostly still happen
17:28:07 <advil> you could disable e.g. hells and tomb if you wanted, but those aren't much time
17:29:04 <advil> e.g. generating depths levels takes the same amount of time whether it's done up front or one by one
17:30:05 <advil> jfc trying to replicate this seeding bug is really annoying, it takes me like 20 minutes to do the cross-compiled build and I've gotten it wrong how to get a cross-compiled debug build at least three times so far
17:30:55 <advil> well, they have all replicated the bug, but with no new info
17:37:48 <gammafunk> advil: curious, why does it take 20 minutes? do you not do native cross-compilation from OS X?
17:37:52 <gammafunk> using mingw
17:38:08 <gammafunk> or are you building from a VM with msys2 or something like that?
17:57:36 <advil> I'm building in a docker container, I haven't been able to get it to work natively on os x
17:58:03 <advil> pretty sure it's not impossible but needs some makefile tweaking
17:58:34 <advil> I can't actually replicate this bug when building directly in msys2 in a windows vm, only by cross-compiling
18:03:19 <gammafunk> oh, that's interesting to know; I sort of assumed if you had mingw set up on os x that crosscompilation would work
18:06:09 <advil> I think it's because the build system needs to use pkg-config to build the tilegen util when cross-compiling, but os x for historical reasons never uses pkg-config
18:06:57 <advil> so I could either change it to allow pkg-config, or change it to build a non-cross-compiled version of contrib libpng for tilegen, but at that point it seemed easier just to do it in a container
18:07:50 <advil> in retrospect that was maybe wrong, but too late
18:08:18 <advil> anyways, 1 episode of dragon prince season 2 = 1 cross-compiled build
18:15:11 <gammafunk> Lasty: thanks for the feedback re: gauntlets, glad you found it fun
18:17:14 <Lasty> gammafunk: np. I think they're looking pretty solid. Interesting choices, some good layouts, a good level of threat / terrain that can be managed but usually not without cost.
18:22:50 <Lantell> Unstable branch on underhound.eu updated to: 0.24-a0-117-g53b51606c3 (34)
18:24:24 <Lasty> gammafunk: how're you feeling about them?
18:25:08 <gammafunk> definitely like it's a success; one direction I might go in is to not have the map fully revealed, but to have them so that there are sections
18:25:08 <gammafunk> at each section you get some kind of vault choice + exit
18:25:18 <gammafunk> so you can't fully see ahead
18:26:12 <gammafunk> if you don't reveal the map and don't have exits when you do progressively reveal sections, you have a problem
18:26:22 <gammafunk> in that it's impossible to know the worst case you'll face
18:26:48 <gammafunk> which is a problem in terms of spoilers and not giving weaker chars a chance to try the portal
18:26:59 <gammafunk> but if you have an exit at each section that's not such an issue
18:27:01 <Lasty> That's not necessarily a problem if it's easy enough to back out
18:27:06 <Lasty> yeah
18:27:24 <gammafunk> elliptic suggested the portal would be better if it were structured this way
18:27:37 <gammafunk> and he's probably right about about that
18:27:51 <Lasty> It would fight the current "check all vaults and loot and think carefully" piece that you get now
18:27:52 <gammafunk> but other than maybe trying that at some point (would take some significant work), pretty happy with it
18:27:54 <Lasty> which can be slow
18:27:58 <gammafunk> yes
18:28:48 <gammafunk> I think it's in a good state that's better than most other portals, so maybe some other portals could get some attention (and of course crawl has lots of other areas to work on)
18:28:58 <gammafunk> would love to see some species reform for some of our 28 species
18:29:06 <gammafunk> as well as some more god reforms (I'm looking at you, Fedhas)
18:29:18 <Lasty> yeah, that's fair
18:29:49 <Lasty> Clearly the answer to species reform is to merge all the simple species together, all the intermediate ones together, etc.
18:29:55 <Lasty> three is more than even
18:29:55 <gammafunk> good idea
18:30:35 <gammafunk> Weak Slime / Squishy Slime / Tough Slime / The Royal Jelly
18:30:40 <gammafunk> the 4 crawl species
18:31:30 <gammafunk> I do have a kind of grand vision for how I'd like branches to have some meaningful gameplay differentiation in terms of unique branch mechanics and kind of pair that will general shortening of the game
18:31:43 <gammafunk> smaller levels with a lot less trivial fights and popcorn monsters
18:31:57 <gammafunk> but that's a really big and general idea
18:32:40 <Lasty> that definitely seems like a good ambition
18:37:22 <Lasty> I still want to fix ranged combat and luring, but lately I've wanted to just refactor stuff. Like moving damage 100% into an object.
21:49:59 <advil> what even
21:50:02 <advil> volatile union luai_Cast u; u.l_d = (d) + 6755399441055744.0; (i) = u.l_l;
21:50:15 <advil> I'm pleased to say that that is *not* crawlcode
22:02:21 <gammafunk> what's up with (d)
22:09:46 <advil> it's in a macro def
22:10:05 <advil> #define lua_number2int(i,d) \
22:10:05 <advil> { volatile union luai_Cast u; u.l_d = (d) + 6755399441055744.0; (i) = u.l_l; }
22:10:34 <advil> but you didn't guess that *that* is what it's doing
22:11:10 <|amethyst> I also like that this is just a block, not a do-while(0)
22:11:11 <advil> this is because some lua hacker, at some point in the past, deemed casting to int too slow (and, presumably, safe)
22:11:45 <advil> the bug I am chasing down turns out to be because cross-compiled builds (and only those) use this particular macro, everything else is falling back on casting
22:14:11 <advil> |amethyst I think that's just because it's called from exactly one place where that is safe
22:14:33 <advil> this is in luaconf.h so it's intended to be replacable in case you come up with something better(??)
22:15:20 <gammafunk> using a block would be unsafe sometimes?
22:16:08 <gammafunk> I'm guessing the whole i vs u business is some kind of lua stack thing
22:16:41 <advil> https://stackoverflow.com/questions/1067226/c-multi-line-macro-do-while0-vs-scope-block
22:17:59 <gammafunk> gah, right
22:20:34 <advil> if I'm right (doing the final check now) I think I might just remove this stuff from being even possible in the contribs
22:20:51 <advil> the problem (95% sure) is that the different cast methods they provide have different rounding behaviors
22:22:13 <advil> which of course can have major butterfly-style effects on levelgen from stuff like layout_rooms, which passes all sorts of fractional numbers back to c++ code
22:23:11 <advil> I wonder if this all evens out or if this has led to some qualitative differences in how these layouts play out in cross-compiled windows builds vs other platforms
22:24:24 <gammafunk> heh, sort of thing that we'd never really know unless someone was playing on windows and online and *really* paid attention, I guess
22:26:15 <advil> yeah, I have no idea if it would even be noticeable
22:26:47 <advil> the two layout_rooms that result are structurally similar, I think the "room" parts are the same in fact, but the paths are quite different
22:27:08 <advil> both look like "normal" crawl levels to me though
22:28:16 <gammafunk> !time
22:28:18 <Sequell> Time: Feb 18, 2019, 03:28:17 AM, UTC. The 2019 0.23 tournament ends in 6 days, 16 hours, 31 minutes and 42 seconds.
22:28:52 <alexjurkiewicz> wow so crawl will need to fork its own lua and not use the system one to fix this?
22:29:07 <advil> uh good point I guess people might use lua from pkg-config?
22:29:24 <advil> it might not be an issue except for some windows builds
22:29:43 <gammafunk> hrm
22:29:45 <gammafunk> looks like cwz being down causes the tournament scripts to not continue processing
22:29:51 <advil> tbh I haven't figured out why it's triggered on cross-compiling but not anything else
22:30:04 <advil> (we already do fork our own lua though)
22:30:16 <gammafunk> that needs to be changed to just have it continue processing the sources that do work
22:30:45 <gammafunk> might it happens that cwz has been down for an hour so no updates
22:30:55 <gammafunk> I guess I'll go to a one-off without cwz enabled
22:32:07 <advil> yep, it's exactly this rounding behavior
22:35:08 <advil> can anyone who knows about gcc provide any guesses as to why the checks at the following link seems to be true specifically on the cross-compiled builds right now? https://github.com/crawl/crawl-lua/blob/master/src/luaconf.h#L557
22:35:15 <advil> as compared to building directly in msys2
22:36:19 <advil> or anything else for that matter, I have no clue when stuff like __i386__ is or isn't set
22:39:25 <|amethyst> Do the native msys2 builds use 64-bit?
22:39:42 <advil> I thought both do but I'm also not 100% sure how to tell
22:39:51 <advil> I'm more sure that the native ones do
22:40:36 <|amethyst> what arch is used for the CROSSHOST?
22:40:42 <advil> CROSSHOST=i686-w64-mingw32
22:41:28 <|amethyst> that's 32-bit (i686)
22:42:20 <|amethyst> and also I guess i686 is old enough it includes pre-Pentium 4 archs that don't have SSE2
22:43:00 <advil> oh yeah the native build is using x86_64-w64-mingw32
22:43:10 <|amethyst> which would mean that one of those __i386__ things is defined (32-bit Intel architecture), and __SSE2__ isn't defined
22:43:18 <|amethyst> which would trigger that #if
22:43:28 <advil> yeah, I see, I had just missed that those were in fact different
22:45:41 <gammafunk> we specifically do 32-bit builds for our official once since 32-bit windows tablets and such are still in wide use
22:45:51 <gammafunk> I switched to x86_64-w64-mingw32 only for one release and people complained
22:46:22 <advil> hm ok doesn't seem like there's an easy way to do this for non-contribs builds (for contribs builds it's easy, and that covers all windows builds)
22:46:34 <|amethyst> could do -msse2 maybe
22:46:38 <advil> ah
22:46:42 <gammafunk> we used to do both 32-bit and 64-bit but it's just such a hassle since it's twice the number of builds and people don't care about any performance difference
22:46:58 <|amethyst> assuming none of these tables use something older than Pentium M or Pentium 4
22:47:05 <|amethyst> s/bles/blets/
22:49:42 <|amethyst> you might also need -mfpmath=sse
22:50:23 <|amethyst> probably not to get that #if to trigger, but not doing so might result in different behaviour between x86 and x86_64?
22:50:38 <advil> ugh
22:52:26 <|amethyst> AIUI the 387, which is what gcc uses on non-Darwin x86-32 targets, uses 80 bits for intermediate results when working with doubles, while sse2 is 64-bit all the way through
22:53:36 <advil> it seems like if I just change the contrib to remove that #ifdef, that leaves in the worst case whoever is still doing 32bit builds on linux
22:53:56 <advil> we also probably want to prevent whatever tf the msvc branch there is doing, too
22:54:17 <|amethyst> debian etc do still do 32-bit linux Crawl builds
22:54:19 <|amethyst> hm
22:55:13 <gammafunk> we actually do them ourselves!
22:55:31 <gammafunk> I do both 32-bit and 64-bit debs with every release for the cdo repo
22:55:46 <gammafunk> by we I guess I mean me, but yeah we do make those
22:56:18 <gammafunk> cdo also does regular trunk debs still I think, don't know if it does 32-bit or what though
22:57:33 <gammafunk> ah, perhaps we don't
22:57:38 <gammafunk> last ones I see are 0.15
22:57:49 <gammafunk> but for those I do see both archs
23:02:11 <advil> ok, reading the docs I think I'd be comfortable adding `-mfpmath=sse -msse2`
23:02:20 <advil> sse2 is from 2000
23:02:29 <gammafunk> looks like http://crawl.newtzie.com/ stopped doing those android tiles builds back in june
23:02:44 <gammafunk> maybe something broke because we didn't update it
23:02:46 <advil> I think they broke and no one fixed them
23:02:53 <advil> with the sdl2 update iirc
23:03:03 <gammafunk> would make sense
23:03:13 <advil> he came in channel and said something, but there wasn't much we could do ourselves
23:04:16 <advil> in fact, not using x87 could cause some seeding issues in and of itself, to the extend we use floating point math in levelgen
23:04:25 <advil> *extent
23:04:33 <advil> er
23:04:39 <advil> *using x87
23:05:46 <alexjurkiewicz> while you're talking makefile issues, it would be nice if AUTO_OPT to set -march=native worked or was removed
23:06:05 <alexjurkiewicz> (immediately below the check for that variable, the -march=native value is overridden with a CFOPTIMIZE := -O2 -parallel