00:04:55 %git 91e3373 00:04:55 07|amethyst02 * 0.21-a0-321-g91e3373: Avoid a gcc warning. 10(27 hours ago, 1 file, 2+ 2-) 13https://github.com/crawl/crawl/commit/91e33734f4fd 00:06:05 all fails too early to be crawl fault 00:06:21 right, travis can't find its outbound again >.< 00:09:27 Unstable branch on crawl.jorgrun.rocks updated to: 0.21-a0-328-g6a5e2f3 (34) 00:13:28 Unstable branch on crawl.akrasiac.org updated to: 0.21-a0-328-g6a5e2f3 (34) 00:20:38 "XL: 2 Next: 1020%" 00:20:49 gnoll bug with 10x skill points thing? 00:26:31 uh, maybe? not trivially reproduceable for me though 00:26:50 !locate rchandra 00:26:50 rchandra was last seen on CJR (rchandra, L5 GnAM of No God). 00:27:16 it looks normal now though? 00:27:55 when did you get that? 00:28:48 killing sigmund, maybe he's just worth more than I thought 00:28:59 !log rchandra 00:29:00 1394. rchandra, XL27 MuAE, T:124279: https://crawl.jorgrun.rocks/morgue/rchandra/morgue-rchandra-20170930-191523.txt 00:29:35 !log 00:29:36 1394. rchandra, XL27 MuAE, T:124279: https://crawl.jorgrun.rocks/morgue/rchandra/morgue-rchandra-20170930-191523.txt 00:30:06 sigmund is often a couple levels worth when you are just a baby 00:30:09 Unstable branch on CRAWL.XTAHUA.COM updated to: 0.21-a0-328-g6a5e2f3 (34) 00:30:14 yeah, I guess that just happens when killing sigmund at low XL 00:30:32 at XL 1 I just got XL: 1 Next: 2080% 00:30:36 I often see 340% or so on the first --more-- prompt 00:32:03 nothing to do with gnolls aside from being more likely to get a --more-- from skills rising 00:32:03 yeah just never remembered seeing so much 01:03:53 just logged in and got the "you have quaffed a potion of experience screen" after updated (I had 0 XP to distribute) 01:14:01 !locate nikheizen 01:14:03 nikheizen was last seen on CAO (nikheizen, L10 DsEn of Kikubaaqudgha). 01:14:15 on that char? 01:14:26 oh, hm 01:15:04 I guess it will happen on transfered games during the transfer sometimes because skill training works better now 01:15:25 it's harmless anyway 01:17:21 Yeah, on that character. 01:20:25 Unstable branch on crawl.develz.org updated to: 0.21-a0-328-g6a5e2f3 (34) 01:28:25 03elliptic02 07* 0.21-a0-329-g12f17e1: Fix transferred non-gnolls being told that they quaffed a potion of experience. 10(3 minutes ago, 1 file, 1+ 1-) 13https://github.com/crawl/crawl/commit/12f17e1c3b07 01:29:08 &versions 01:29:21 CAO: 0.21-a0-328-g6a5e2f3, CBRO: 0.21-a0-321-g91e3373, CDO: none, CJR: 0.21-a0-328-g6a5e2f3, CPO: 0.21-a0-328-g6a5e2f3, CUE: 0.21-a0-324-gfc3775b, CWZ: 0.21-a0-277-geaec956, CXC: 0.21-a0-328-g6a5e2f3, LLD: 0.21-a0-142-gdd972a7a06 01:32:55 Unstable branch on crawl.akrasiac.org updated to: 0.21-a0-329-g12f17e1 (34) 01:33:48 Unstable branch on CRAWL.XTAHUA.COM updated to: 0.21-a0-329-g12f17e1 (34) 01:58:39 Windows builds of master branch on crawl.develz.org updated to: 0.21-a0-329-g12f17e1 02:10:01 Unstable branch on crawl.jorgrun.rocks updated to: 0.21-a0-329-g12f17e1 (34) 02:11:55 The build passed. (master - 12f17e1 #8780 : elliptic): https://travis-ci.org/crawl/crawl/builds/281830686 02:46:06 -!- rchandra is now known as rc_zzz 02:54:58 Monster database of master branch on crawl.develz.org updated to: 0.21-a0-329-g12f17e1 03:23:07 Unstable branch on crawl.beRotato.org updated to: 0.21-a0-329-g12f17e1 (34) 03:39:16 03gammafunk02 07* 0.21-a0-330-g461fa0c: Re-allow Trog worship for Gnolls 10(15 minutes ago, 3 files, 20+ 9-) 13https://github.com/crawl/crawl/commit/461fa0c58c53 04:04:45 -!- amalloy is now known as amalloy_ 04:09:53 Unstable branch on crawl.jorgrun.rocks updated to: 0.21-a0-330-g461fa0c (34) 04:29:27 -!- amalloy_ is now known as amalloy 06:35:52 -!- amalloy is now known as amalloy_ 09:13:22 looking at this morgue of a gn from before elliptic's gn-specific fixes (this is right before the even training commit) makes me think that skill point allocation on training is probably really broken: http://crawl.xtahua.com/crawl/morgue/Fingolfin/morgue-Fingolfin-20170930-212838.txt 09:14:04 there's a spread of 3.5 to 5.1 for skills that would start out 0 09:14:15 ok, maybe "really" is an exaggeration 09:25:49 by "really" I mean "in an incredibly subtle way unless you are training tons of skills at once with equal starting points and aptittudes" 09:46:38 advil: I think e65fa8f probably smooths that out somewhat even for non-gnolls 09:47:08 it was never really broken, just more randomized than it needs to be 09:47:14 oh nice, that also explains why I back off my "really", because I was testing with that one 09:48:25 I also suspect that before your commits, skill cost increases in the middle of training everything could introduce other oddities, but that just doesn't come up in the same way for normal characters 09:48:26 could maybe be improved more, but it was already quite a small thing in practice (where you aren't training that many skills) 09:48:34 yeah 09:48:54 I did a bunch of testing and even with 4 identical skills being trained on a Hu it's just not that noticeable 09:49:27 you can see it in the actual skill points in wizmode, but it doesn't translate to much except one skill occasionally being off by 0.1 for a bit 09:49:42 so nothing like that morgue I linked 09:59:35 there are still some really weird things happening in _train_skills that could be improved if desired 09:59:55 I'm honestly surprised everything works as well as it does 10:00:00 for non-gnolls 10:08:05 03advil02 07* 0.21-a0-331-gfb491b4: Clean up some fulldebug spam for gnolls 10(3 minutes ago, 1 file, 5+ 1-) 13https://github.com/crawl/crawl/commit/fb491b441f31 10:08:15 yeah, it's a total mess 10:09:12 at some point I'd like to at least brainstorm eliminating skill cost levels entirely (I think this has come up before a few months ago) 10:09:53 and rescaling skill points somehow 10:26:24 what's wrong with skill cost levels? all that they do is make skill points cost more as the game progresses, which seems good to me given that xp increases and we don't want a single monster later on to train all your skills from 0 to 8 10:26:35 So it's the 2nd of october in australia now 10:26:39 happy 20th birthday dungeon crawl 10:27:52 (basing the date off this post https://groups.google.com/d/msg/rec.games.roguelike.misc/AzYHcmpPSqw/2l-5yhzS_uIJ) 10:29:08 Strange behaviour of Tomb hatches 13https://crawl.develz.org/mantis/view.php?id=11244 by stickyfingers 10:33:36 elliptic: I think their goal is fine, but imo they're really confusing, both for us in terms of implementation details (at least, for me) and for the player 10:34:21 I actually think that what you had to do to get Gn to work cleanly kind of illustrates their code-cost :) 10:37:00 it's possible there's a better implementation that still uses skill cost levels, too, maybe what you did with Gn could be generalized & made more efficient? 10:39:35 yeah, I guess I just think that the concept of skill cost level is fine and the issue is that the current implementation does a lot of unnecessary things 10:40:05 and also everything has a bad name 10:40:22 I could imagine many alternatives that in principle could accomplish that goal, here's two: scale skill points differently per xl, have skill cost levels (or a skill cost function) per skill 10:41:07 either of those seems way more transparent (though I have no idea if they'd work) 10:41:13 I don't know what you mean exactly by "scale skill points differently per xl" but that is basically what skill cost level is already 10:41:32 having it be a per-skill thing would have a very different effect though 10:42:19 as I said before, the purpose is that raising a skill from 0 to 10 late in the game requires more xp than raising it earlier in the game 10:42:36 so it sort of needs to depend on something other than that skill's current level 10:42:52 re per xl, that may be how it basically works, but that's *really* far from how it's implemented 10:42:57 not really? 10:43:35 <|amethyst> it's based on total XP rather than actual XL so that it's not affected by XL apts 10:43:37 skill cost level is very very close to what your XL would be if you had a specific fixed xp apt 10:44:08 it's only different from that as far as !exp goes, because that trains skills by a different amount than the xp it gives you 10:44:48 I think you're making the case that you understand how it works, not that it's not confusing :-P 10:44:48 my understanding is that ancient skill code actually used your XL 10:45:01 and we changed that because it was weird to depend on your XL apt 10:45:34 it's not like it is visible to the player anyway 10:46:26 <|amethyst> You could replace skill cost levels and have exactly the same effect (modulo XL apts) by instead: (1) dividing all incoming XP by an equivalent amount, based on your level and (2) reducing the XP requirements for acheiving higher XLs accordingly 10:46:35 my case there is that it is one of many factors that increases the opacity of how what you do relates to how skills increase 10:47:09 orc warrior (08o) | Spd: 10 | HD: 4 | HP: 23-33 | AC/EV: 9/10 | Dam: 20 | 10weapons, 10items, 10doors, fighter | Res: 06magic(20) | XP: 133 | Sz: Medium | Int: human. 10:47:09 <|amethyst> %?? orc warrior 10:47:48 if it makes things easier to think about, there is some weird function F determined by the code such that raising your skills by N skill points over the course of the game requires F(N) xp 10:48:14 yes, and that function is quite weird -- breakpoint-y, and also never overtly present in the code 10:48:40 sure, we could make it smoother if we wanted but I'm not sure there is that much point 10:49:09 <|amethyst> elliptic: keep in mind advil just implemented something to predict how many XLs of training you'll need to reach a certain skill level 10:49:22 <|amethyst> elliptic: that would definitely be simplified if skill cost level were easier to invert :) 10:49:28 yes, that's exactly where I'm coming from 10:50:01 !source skill_level_to_diffs 10:50:01 1/1. https://github.com/crawl/crawl/blob/master/crawl-ref/source/skills.cc#L1298 10:50:27 well, I didn't implement that all, I just generalized + fixed existing code 10:50:48 but figuring out how that worked was not trivial 10:51:29 also unfortunate that this is an entirely separate code path to training (at least that precedes me) 10:52:48 hm, I probably need to double check that this does the right thing for gnolls, should be ok I hope 10:53:33 would it actually be that much simpler to think about if skill_cost_level was some explicit smooth function? I guess I have the opinion that piecewise-constant and piecewise-linear functions aren't such a bad way to describe things if we don't actually care too much about smoothness for gameplay reasons 10:55:42 The build has errored. (master - fb491b4 #8782 : advil): https://travis-ci.org/crawl/crawl/builds/281919653 10:58:38 well, if it allowed something other than an iterative implementation of training / inverting training, I think it would be nice 10:58:54 but I don't have a particular approach in mind yet 11:00:28 in any case, I guess we are at least all agreed that _train_skills needs some work :) 11:00:54 I think probably the simplest approach would be to pick a function G such that G(you.total_experience) = total skill points available to train this game 11:01:25 and then for training, just see how much you.total_experience increased by and distribute those skill points between current skills 11:01:33 that sounds great 11:06:19 current G is basically just the piecewise linear function given by integrating 1/skill_cost_level, so IMO skill_cost_level isn't a terrible description of it :P but I'm not opposed to smoothing it out a bit if someone wants to do that 11:09:13 Unstable branch on crawl.jorgrun.rocks updated to: 0.21-a0-331-gfb491b4 (34) 11:11:16 hm, skill cost level currently looks sort of like the cumulative distribution function of a gaussian 11:11:21 I guess that sort of makes sense 11:11:22 probably for training implementation we'd also want to replace you.exp_available with you.skill_points_available 11:11:23 https://www.dropbox.com/s/cp4336dwx1z6igv/Screenshot%202017-10-01%2011.07.53.png?dl=0 11:11:47 and just increment that using G inside gain_exp() when increasing you.total_experience 12:48:20 in fact skill cost level is a slightly off version of the cdf for the 27,0.5 binomial distribution, perhaps someone took that and hand-tweaked it so it maxes out a little bit later 12:52:01 scaled to 270 of course 13:05:06 Unstable branch on crawl.akrasiac.org updated to: 0.21-a0-331-gfb491b4 (34) 13:07:46 that might just be an accident, though, given https://crawl.develz.org/wiki/doku.php?id=dcss:brainstorm:skills:experience_system#new_skill_cost 13:09:43 crawlcode numerology 13:37:55 Lightli points out that the ! is now broken on !apt, because it only shows up for gnoll apts now 13:38:03 no idea how to fix that, myself 13:38:25 !apt gn 13:38:25 Gn: Fighting: 8!, Short: 8!, Long: 8!, Axes: 8!, Maces: 8!, Polearms: 8!, Staves: 8!, Slings: 8!, Bows: 8!, Xbows: 8!, Throw: 8!, Armour: 8!, Dodge: 8!, Stealth: 8!, Shields: 8!, UC: 8!, Splcast: 8!, Conj: 8!, Hexes: 8!, Charms: 8!, Summ: 8!, Nec: 8!, Tloc: 8!, Tmut: 8!, Fire: 8!, Ice: 8!, Air: 8!, Earth: 8!, Poison: 8!, Inv: 8!, Evo: 8!, Exp: 0, HP: 0, MP: 0 13:38:27 !apt mf 13:38:27 Mf: Fighting: 1, Short: 2, Long: 1, Axes: -2*, Maces: -2, Polearms: 4, Staves: -2, Slings: -2, Bows: -2, Xbows: -2, Throw: 0, Armour: -3*, Dodge: 3, Stealth: 2, Shields: 0, UC: 1, Splcast: -1, Conj: -2, Hexes: 0, Charms: 1, Summ: 0, Nec: -2, Tloc: -2, Tmut: 3, Fire: -3*, Ice: 1, Air: -2, Earth: -2, Poison: 1, Inv: 1, Evo: 0, Exp: 0, HP: 0, MP: 0 13:40:14 !source apt 13:40:14 1/1. https://github.com/crawl/crawl/blob/master/crawl-ref/source/delay.h#L30 13:40:25 !cmd apt 13:40:25 Built-in: !apt => https://github.com/crawl/sequell/blob/master/commands/apt.pl 13:40:34 advil: commit to that script in sequell 13:40:50 *change that script in the sequell repo 13:41:02 not sure if shalott will automatically pull that change 13:41:37 !apt summonings 13:41:38 Summ: Gn: 8!, Ba: 2, Te: 2, DE: 1, HE: N/A, VS: 0, Op: 0, Ko: 0, Fe: 0, Vp: 0, Mf: 0, Fo: 0, Ds: 0, Dr: 0, Hu: 0, Na: 0, HO: 0, Og: -1, Dg: -1, Gr: -1, Gh: -1, DD: -1, Ce: -1, Sp: -2, Ha: -2, Mu: -2, Mi: -3*, Tr: -3* 13:42:23 ah 13:42:47 I'll let someone who knows perl do that, I'm sure it's just a quick change to is_best_apt but I'll probably get the syntax wrong 14:03:55 +You feel you are being watched by something. @player_only@ 14:15:27 -!- Whistling_Bread is now known as Cerpin 14:18:59 -!- mol is now known as Guest9758 14:19:38 -!- Guest9758 is now known as molll 14:20:17 advil: the piecewise linear function that we actually want to approximate is the one passing through these points (rounded slightly): https://bpaste.net/show/52037a8c407c 14:23:37 this is a function from total xp to skill points? 14:24:22 yeah, from you.total_experience (which is usually 10 * you.experience) to the number of skill points you're allowed to train (not counting ones you start with) 14:26:54 e.g. using &k to go to L10 as HuFi while only training unarmed brings you.total_experience to 39080 (as per &=) and puts 4675 skill points in unarmed 14:27:53 and (39080, 4675) is very close to the line between (25727, 3647) and (50817, 5577) 14:29:59 I'm not really sure smoothing this out is worth it if it messes too much with early-game balance, but at least using this table instead of the calc_skill_cost table would probably be an improvement 14:31:03 yeah, I agree smoothing may not be worth it 14:31:30 (this table is just integrating 1/calc_skill_cost using the breakpoints from skill_cost_needed, which use the ugly formulas in exp_needed) 14:41:02 -!- amalloy_ is now known as amalloy 15:49:12 -!- molly is now known as Guest36784 15:49:23 -!- Guest36784 is now known as molll 18:15:41 Unstable branch on underhound.eu updated to: 0.21-a0-331-gfb491b4 (34) 19:19:15 "I can finally have a pain weapon with Trog!" in response to Gnolls being able to worship Trog again 20:26:18 a video on nethack https://youtu.be/SjuTyJlgLJ8 20:33:27 such interaction 22:07:18 -!- mikee__ is now known as mikee_ 23:28:05 another video on nethack https://youtu.be/D9oMSPzChgk 23:37:09 I appreciate that they used the DOS version just so they could put that obnoxious BEEEEEP in the video 23:40:24 where's the list of stuff that currently needs to be implemented / fixed in codebase? 23:44:18 nvm found it 23:44:43 there are a number of such lists. three i can think of: mantis bugs; github issues/PRs; implementable 23:44:49 ??implementables 23:44:49 implementables[1/1]: https://crawl.develz.org/mantis/search.php?project_id=1&category=Implementables&status_id%5B%5D=10&status_id%5B%5D=20&status_id%5B%5D=30&status_id%5B%5D=40&sticky_issues=on&sortby=last_updated&dir=DESC&highlight_changed=12&hide_status_id=-2 23:44:54 nice url 23:45:00 don't blame me 23:45:21 I was looking for the one that talked about the desolation of salt clouds 23:45:34 yeah PF did make a bug for that one 23:45:35 let me see 23:46:02 ...oh 23:46:10 !bug 0011244 23:46:10 https://crawl.develz.org/mantis/view.php?id=0011244 23:46:12 that's a good bug 23:46:37 if it's actually happening that way 23:49:23 hrm 23:49:34 can't seem to find that bug about desolation clouds 23:49:59 desolation clouds are mentioned at https://crawl.develz.org/wiki/doku.php?id=dcss:planning:0.20_plan 23:50:56 yeah, but was there a bug with some technical discussion? 23:51:03 I seem to recall one but maybe there wasn't 23:51:06 I certainly can't find one 23:51:12 also, the tomb stair thing seems legit. i just went to W:1 and took a non-default stair first. now that stair and the mani stair both take me to the center of W:2 23:51:44 non-default stair? 23:52:06 do you mean you took one of the gauntlet down-hatches? 23:52:21 the early one should take you to center of W:2 23:52:27 the later one should take you to W:2 side chambers 23:52:45 the early one is simply an "escape" stair for the gauntlet 23:52:50 the later one takes you to the rest of W:2 23:52:55 so it should go to the side chambers 23:53:03 whats the ! command here to search for fog_machine 23:53:21 don't think it will work for lua, Naruni 23:53:21 eww 23:53:22 dat/dlua 23:53:26 git git grep from dat/dlua 23:53:31 from that dir I mean 23:53:58 i've barely played since the tomb escape hatch change, so i guess i don't understand how the hatches are supposed to work 23:54:10 it's pretty similar to old stone stairs 23:54:20 so you enter gauntlet normally from W:2 23:54:33 in one of the W:2 side chambers connected to the W:2 center 23:54:38 where by "gauntlet" you mean...? 23:54:46 is a hatch taking you W:1 'gauntlet' 23:54:47 the other half of W:1? 23:54:53 which is the corridor on W:1, yes 23:55:00 that's not connected to the rest of W:1 23:55:31 taking that W:2 hatch to 'gauntlet' gets you to start of gauntlet on W:1, and a few squares from that point 23:55:39 is a hatch taking you back to the initial W:2 area you left 23:55:45 i.e. it takes you back to center of W:2 23:55:56 so, the bug report sounds like it is describing intended behaviour then? 23:56:07 oh, wasn't he shafting though? 23:56:11 no 23:56:14 oh, oops 23:56:18 I didn't read it carefully 23:56:31 yeah quite possibly just intended behaviour and he didn't realize where the hatch would lead 23:56:48 must have teleported into gauntlet from W:1 or somethhing 23:56:50 it surprised me too, to have two hatches go to the same place 23:56:52 LRD, he said 23:56:57 ah yeah right LRD 23:57:46 I mean, that 'escape' hatch in gauntlet could go to an area close to the hatch you took up but I think it's better that way 23:57:58 the whole idea is to have stairs you can't cycle quite so easilly 23:58:01 so going back to center is nice 23:58:08 normally it's irrelevant if you cleared W:2 23:58:17 but if you rushed through W:2 and went to gauntlet 23:58:34 and need to escape gauntlet through a hatch for some reason 23:58:34 well you're right back in the center of W:2 23:59:34 The only bug....is that they didn't die.... 23:59:49 Unstable branch on CRAWL.XTAHUA.COM updated to: 0.21-a0-331-gfb491b4 (34)