03:33:32 Experimental (bcrawl) branch on underhound.eu updated to: 0.23-a0-5261-gd9800d219b 06:25:26 New branch created: spell-ui (3 commits) 13https://github.com/crawl/crawl/tree/spell-ui 06:25:28 03CrawlOdds02 07[spell-ui] * 0.35-a0-1314-g4d222eda59: Fix some errors and crashes in skill target estimation 10(5 hours ago, 3 files, 63+ 37-) 13https://github.com/crawl/crawl/commit/4d222eda5965 06:25:28 03CrawlOdds02 07[spell-ui] * 0.35-a0-1315-g450a86508d: Show the effect of training skills on spell failure 10(7 days ago, 2 files, 135+ 55-) 13https://github.com/crawl/crawl/commit/450a86508dbc 06:25:28 03CrawlOdds02 07[spell-ui] * 0.35-a0-1316-gc930725692: Show the skills needed to get a spell castable 10(18 hours ago, 7 files, 150+ 40-) 13https://github.com/crawl/crawl/commit/c93072569282 06:28:02 <11O​dds> If anyone wants to glance at the broad sweep of that it would be welcome - as well as the decisions around exactly what to show, there's a bit of fiddliness around various ways skills can be modified 07:56:04 MoodEB (L10 DsEn) ASSERT(!in_bounds(you.pos()) || !cell_is_solid(you.pos()) || you.wizmode_teleported_into_rock) in 'main.cc' at line 4429 failed. (D:8) 08:06:32 <11O​dds> !crashlog moodeb 08:06:35 <04C​erebot> 1. MoodEB, XL10 DsEn, T:10106 (milestone): https://crawl.akrasiac.org/rawdata/MoodEB/crash-MoodEB-20261021-145603.txt 08:07:03 <11O​dds> Heh, v0.14 09:47:49 <12g​e0ff> This version gives the player an avg spell school(s) target and how many XLs it'll take to get to a specific fail rate, w/o taking into account Spellcasting, Armour, and aptitudes, right? 09:47:58 <12g​e0ff> The XL estimate would be incorrect - too high - for Gns and Djs, as they train Spellcasting together with magic schools. 09:48:19 <12g​e0ff> Species with good Spellcasting apts like DE, Mu, and even Oni could get a spell online faster by investing more into Spellcasting instead of bringing the relevant magic schools to a suggested level. Which means this feature would give them estimates that are pretty far away from an optimal path. 09:48:38 <12g​e0ff> Also, what exactly the player is supposed to do with this information? Suppose I'm playing a Gr - Spellcasting -1, Forgecraft -1, Fire 0, and Earth 2 - and want to get the Mortar online. The game tells me that I need 16 in Forgecraft, Fire, and Earth, and it'll take 3 XLs. Now what? 09:48:51 <12g​e0ff> A new and naive player would set 16 into these three schools and would get the Mortar much later than they could. 09:49:00 <12g​e0ff> Another player would open an online calculator, put these aptitudes and target into a spreadsheet, or get a trusty arithmometer, and calculate the optimal targets. 09:49:15 <12g​e0ff> The remaining 90% of the players won't even use this feature, because it's too impractical on its own, compared to more straightforward UI solutions like setting a min delay skill target for a weapon. 09:49:27 <12g​e0ff> DCSS is a tactical game where the player has to do a lot of tactical and strategic decisions. It's not about memorizing formulas, crunching numbers, or solving equations. 09:49:35 <12g​e0ff> My dream is to have something more precise and immediately useful, which at the very minimum means taking the apts and Spellcasting into account: 09:49:44 <12g​e0ff> "Getting to 10% fail rate with your current stats and equipment requires at least Xs Spellcasting, X1 School_1, X2 School_2, and X3 School_3. At 100% training you would reach that in about Y XLs. Press (s) to set these targets." 09:57:30 I don't see that helps unless skills can be trained other than at single or double rate - just turning them all on (as (s) on a weapon does) will reach the Spellcasting target first so won't actually be spending XP on lowering fail rate as efficiently as possible. 09:57:32 <12g​e0ff> @gammafunk , by the way, the old dev wiki https://crawl.develz.org/wiki/doku.php?id=start is showing 404 since last week. Kinda worried, 'cause DokuWiki uses php, and there's a php-apocalypse going on right now (two (?) severe vulnerabilities in all PHP versions since 2004) 10:03:27 <12g​e0ff> Lowering the fail rate as efficiently as possible would require some Armour/Shields training too. I'm worried more about giving the player something less incorrect and more useful 10:05:16 <09g​ammafunk> it was 404 for a reason, which is that CDO has been having some problematic web traffic. It will hopefully return when Napkin sorts that out. But right now, crawl.develz.org httpd is actually offline because of said PHP vuln. Once Napkin gets it patched, the website will return. Not sure on the ETA of doku wiki after this 10:05:46 <09g​ammafunk> But yes, Napkin was fully aware of the 404 and PHP issues and is dealing with both 10:07:14 <11O​dds> Now you know roughly how far away mortar is from being online and whether it's reasonable to shoot for it now, is the idea. I agree that the advice will often be far from an optimal path, and that the version you dream of is more useful. The concern is that it is more likely that people then follow it precisely, when it's probably a bad idea for a number of reasons and also likely to change while you follow it because it's a fairly 10:07:14 long range estimate. I'm personally pretty torn on this. 10:08:49 <12g​e0ff> https://klipy.com/gifs/lisa-disaster 10:08:53 <11O​dds> (See also the conversation on this last week after my initial mock version which more less your dream) 10:31:48 <12g​e0ff> Yeah, I've seen that and thought about this problem for a while. Comparing to a similar case from the past, there probably were players who opposed the addition of the min delay skill targets back in 0.21 (d461b0ec43), since it made all their knowledge of equipment-specific numbers less relevant. But I don't remember any arguments that this made the game feel more hands-on or took away agency from the player. 10:32:56 <12g​e0ff> > also likely to change while you follow it because it's a fairly long range estimate My intuition is that it mostly affects only a subset of estimates, when it's say 2+ XLs away. And even for such cases, taking more things into account and therefore being less wrong is better 10:33:06 <12g​e0ff> > a bad idea for a number of reasons OK, let's go through them 10:33:18 <11O​dds> %git d461b0e 10:33:19 <04C​erebot> advil * 0.21-a0-304-gd461b0ec43: Add a command to set skill targets from an item description (9 years ago, 5 files, 176+ 43-) https://github.com/crawl/crawl/commit/d461b0ec437a 10:35:08 <11O​dds> Other then “things will change”, the other big one is that all the skills you’d train for this spell do other things (ac, sh, mp, spellpower, other spells) such that it’s very likely this is the wrong plan 10:36:48 <12g​e0ff> (also, if I really need to convince DO instead of preaching to the converted, just say that 🙂 ) 10:38:21 <11O​dds> You’re preaching to the undecided 🙂 As a player I’d like the version you’d like. As a dev I’m super unsure. 10:38:55 <11O​dds> Eh I don’t actually want the skill target button you want even as a player 10:39:04 <11O​dds> I think the plan will be too wrong for that 10:39:43 <11O​dds> Though ofc I could simply not press it 10:40:00 <12g​e0ff> the player takes a pen and pencil and writes the number down, then opens the m screen and puts the number in 10:40:20 <11O​dds> Right but I’m almost never actually following this plan for the reasons above 10:40:23 <12g​e0ff> 0 hand holding, 0 agency taken away 10:41:25 <11O​dds> (I guess the one time I might follow it is when it’s “just train translocations from scratch to get gells on my mage) 10:41:27 <12g​e0ff> just to check, when we talk about the plan, we don't assume that the player turns all others skills off? 10:41:48 <12g​e0ff> like, i'd enable the plan, but still train other things in parallel 10:42:16 <12g​e0ff> or enable the plan, like set the target for 3-4 skills, but train them one at a time, particleface-way 10:42:51 <11O​dds> Yeah I was thinking we were talking about just setting the targets 10:42:56 <11O​dds> Ofc neither currently exists 10:52:43 <12g​e0ff> > all the skills you’d train for this spell do other things I'm trying to understand this part but failing. Sure, more Spellcasting means more MPs, more Fire gives more spellpower to my current fire spells. Like, I want the mortar online, but I still need to survive until it gets castable. My current spells might get some benefit from training Mortar-related skills. Otherwise I'd need to train, say, Conj or something else, to 10:52:43 survive. This feels disconnected from having a more precise estimate vs. a less precise one, from having one avg skill target vs. having aptitude-aware ones. 11:10:26 <11O​dds> The idea is that a precise estimate looks more like a recommendation and is more likely to be followed like one, when it's unlikely to be a good one 11:12:43 <04d​racoomega> A new and naive player is making skilling decisions far far worse than anything this model would actually suggest. Have you seen how many different superfluous things most of them train at once? ^^; (Which is to say that I don't feel this advice is misleading to them in a practical way, because it's probably already better than what they would do without it, and it's also focused on a specific part of the problem domain.) 11:13:31 <04d​racoomega> I don't feel this is a fair comparison, since 'reaching mindelay' has a trivially calculable target in exactly one skill. It's easy for anyone to have calculated it themselves and there's no room for conflicting or subjective advice on the fastest way to reach it. There's exactly one way to reach it and no other skill will contribute even slightly. Whereas 'fastest' way to reach 10% fail rate on a spell is a lot more nuanced, 11:13:32 changes based on other stat and gear choices along the way (which is extremely true when looking at the path towards a very distant spell - and you know how excited new players get about the level 9 spells they find in early D...), and involves many skills with secondary effects of variable value to the player. (So whether or not one feels the 'optimal path' calculator for spell failure is good for the game, I don't think one can fairly argue 11:13:32 that it's the same situation as mindelay targets.) 11:17:07 <11O​dds> (Opening my current game and checking... yeah, 3 focused and 4 non-focused skills #optimaltraining) 11:22:46 <12g​e0ff> > Have you seen how many different superfluous things most of them train at once? in several early games, I've trained shields to 27 while wearing a buckler, so I know about that > reaching mindelay' has a trivially calculable target it doesn't require any calculations, it's a fixed weapon-specific number > There's exactly one way to reach it [min-delay target] similarly, there exists a way to get to 10% fail rate by spending a 11:22:47 minimal amount of XP. Are you sure that more than one way exists (if we are minimizing the xp spent)? 11:24:09 <04d​racoomega> Are you... are you really quibbling over the exact degree to which knowing a number is trivial?? 11:25:14 <12g​e0ff> i don't understand your question 11:27:05 I agree there's only one way (if we don't account for INT / STR / armour etc changing en route); my concern is more that there's nothing magic about 10%, I want to know what skill will most improve failure rate _right now_ 11:28:17 <11O​dds> ^ That is not my intent with this feature; it can be worked out (albeit in a few more button presses) with the other less controversial commit on this branch 11:28:20 A lot _can_ change en route, most obviously tripping over a ring of wizardry - then I want to have spent my XP as efficiently as possible up to that point, not dumped extra into Spellcasting. 11:29:24 <12g​e0ff> tripping over a relevant manual would be also interesting, and would also require redoing skill targets 11:29:36 <11O​dds> Yeah I nevertheless think there's value in knowing how long the route is 11:29:43 <04d​racoomega> Anyway, it may be true that only one way exists to reach 10% in our current stats/gear while absolutely minimizing XP spent, but I think some of the debate is about how closely 'Reaching this while spending the least XP possible' is to how one should actually train towards anything here. 11:33:01 <12g​e0ff> Sure, the game could show the player skill targets, but the player has to choose how to train them, in which order, and what other skills to train. I'm not suggesting to lock their skills until they reach them 11:35:30 <12g​e0ff> Also, I'm not suggesting to include Armour/Shields, which were in the Odd's initial idea, in the mix. But ignoring Spellcasting and especially the aptitudes feels very wrong 11:35:48 <04d​racoomega> Oh, this version ignores aptitudes? 11:36:00 <04d​racoomega> I hadn't looked at specifics yet, I will be honest 11:36:17 <12g​e0ff> (and to reiterate, for Djs and Gns, spellcasting is a part of their magic) 11:36:47 <12g​e0ff> (and you can't turn it off and ignore its effect) 11:36:53 <11O​dds> It currently gives the level you'd need to get to and the total XP you'd need to get there. It doesn't show the latter for gnolls, but yeah it's wrong for djinni atm (as are skill target calculations actually) 11:37:43 <11O​dds> "Training evenly" felt probably less right on average - if you have tons of summoning and no translocations, you don't want to be told to train those evenly to get Forest online 11:38:28 fr: let me select a spell to single or double train which automagically trains the appropriate skill 11:39:22 <11O​dds> And also "you need an average of 16 in these schools" is a pretty good way of conveying how much training is needed 11:40:07 <11O​dds> Actually I guess both versions have that 11:40:09 <12g​e0ff> I wonder how close to the optimal™ training the current auto-training mode (90% sure that it takes apts/spellcasting into account) 11:40:56 <04d​racoomega> I mean, absent any ongoing debates here, I think this would be clearly more valuable to have over the nothing that we currently do, to be clear. 11:41:19 <04d​racoomega> I would be staggered if it turned out auto-training was even slightly close to optimal at anything. 11:41:35 (I hope you'll forgive my interjections but this is very interesting) 11:41:55 <04d​racoomega> Every time I am on a new install and haven't remembered yet to turn it off, the ratios it gives me always seem wild 11:42:43 <11O​dds> I feel like maybe it's not awful for melee brutes, partly because they don't use many skills and they are all pretty useful most of the game 11:43:07 <04d​racoomega> I feel this may be by virtue of it having less room to fuck up rather than it being any bit more sensible about its priorities 11:43:24 <11O​dds> For mages.... it would be probably better if it didn't exist, it's misleading to suggest you can play the game with it on 11:44:40 <12g​e0ff> it would be an interesting challenge - trying to win a mage w/o opening the m menu at all 11:44:52 <04d​racoomega> Quite a while ago, I actually suggested removing it. Like, I've never seen a 'training wheels' system in any game where the universal first suggestion given to a new player is to turn it off. 11:45:10 <04d​racoomega> Which is a heck of a red flag 11:45:12 <12g​e0ff> (for a non-gn/dj, obviously) 11:45:34 <11O​dds> Yeah. I suppose there was disagreement on this? 11:45:40 <11O​dds> On removing it, that is 11:46:02 I don't suggest that! I think autotraining is not so bad for a melee brute and there's huge scope to fuck it up manually by, say, never training Fighting 11:46:03 <04d​racoomega> There was some pushback at the time, yeah. (I do earnestly wonder if just training all your starting skills evenly is any worse than what autotraining does right now in practice. Honestly unsure.) 11:46:56 <11O​dds> We could make sure that the starting skills for each class aren't totally awful as we did it 11:47:05 <04d​racoomega> Given that certainly nobody has examined it in detail in a very long time, I feel like the degree to which it results in sensible choices is largely by accident 11:47:38 <04d​racoomega> Like, I think the code that does this is all effectively left over from victory dancing days, which had very different design intents 11:47:42 <11O​dds> Do you actually use autotraining? 11:47:53 <11O​dds> (And does quoting like that come across correctly IRC-side?) 11:48:36 No. But if that was to me, no. I'm not a new player. But I don't think it's automatically bad for a new player on a melee character # which is a faint endorsement, I'll grant you 11:48:38 <12g​e0ff> you need to ping irc peoples by name 11:49:52 <12g​e0ff> there was some work on autotraining around 0.19-0.20 11:50:00 <04d​racoomega> Oh? 11:50:01 <12g​e0ff> %git 103566974 11:50:02 <04C​erebot> PleasingFungus * 0.19-a0-1531-g1035669744: Make autotraining a bit smarter about weapons (10 years ago, 4 files, 41+ 19-) https://github.com/crawl/crawl/commit/10356697443c 11:50:15 <04d​racoomega> Oh, huh 11:50:16 <12g​e0ff> (but nothing more recent, it looks like) 11:51:52 Train starting skills evenly, turn that off when one reaches 10 and "you must enable at least one skill for training"? 11:53:25 <04d​racoomega> Like, obviously it's a good idea to guide beginners in vaguely sensible directions, but it's not clear to me that autotraining even does that. It's, in fact, somewhat opaque that the game is even doing that. (And I myself, on new installs, have manually set which skills I want to train and sometimes been surprised later that one of them was only training at ~10% because I forgot to disable that part) 11:53:28 <12g​e0ff> actually, the mega-commit that added talismans, 188002318 (0.30), had some auto-training for them too 11:54:48 <04d​racoomega> I wonder be honestly curious to take a closer look at what decisions autotraining actually reaches about things in real gameplay (but that does, of course, mean playing moderately long into actual games using autotraining on a few types of characters, so not the quickest to gather info on.) 11:56:42 <12g​e0ff> > // Skill training for when the player casts or miscasts a spell. > ... > // Fill all disciplines into a vector, then shuffle the vector, and exercise skills in that random order <--- optimal training, confirmed 11:59:46 Couldn't you make the morgues reflect what proportion of XP was autotrained and let other people do the work 12:02:16 <11O​dds> Hmmm yeah probably lots of the "training events" are captured such that if you could find autotrained morgues you'd get a lot of info from them 12:16:47 <11O​dds> On skill targets, I'll probably go ahead with the current version as I think it's useful and from conversations so far I doubt we're going to reach a consensus view on a better option. 12:17:42 <04d​racoomega> No arguments from me 12:18:04 <04d​racoomega> Things can always be incrementally improved in future, as always 12:19:23 <11O​dds> @dracoomega the code is a bit fiddly, and if you had a moment to glance at it and check you can't spot a better overall approach that would be appreciated 12:19:59 <12g​e0ff> (unless i missed something, exercise.cc doesn't care about apts and trains Spellcasting more like incidentally than deliberately) 12:20:54 <04d​racoomega> I am quite a bit more sleepy than I would like to be when analyzing something like this, but I can take a little look in a few minutes anyway 12:31:18 <11O​dds> Thanks! No rush if you want to leave it for a more wakeful moment 13:38:56 <04d​racoomega> So, having read over the code for a bit and then fired up the branch, I first want to say how immediately useful it feels like the in-game numbers already are. Like, yes, it's not an optimal training path, but it gives a much better ballpark figure for 'How far away is this?' than I'd ever intuitively have and generally just feels good. Took me a moment to realise you were essentially doing a binary search of spell level 13:38:57 targets to decide which was best. Kind of cute. I did wonder if there was a less brute-force way to calculate the desired target, since we already know the raw spell fail to hit, and we know the contribution of int and spellcasting and spell difficulty (and it would be easy to refactor _apply_spellcasting_success_boosts() to something that can return the wizardry-adjusted spell fail target. I think then it might be straightforward to derive how 13:38:57 much skill contribution is needed to cover the difference between the target raw fail and the current raw fail? But it's also very possible that doing it just slightly brute-forcey is fine, since that's more guaranteed to remain in sync with spell failure code (if any part of it changes in future) and the binary search does reasonably minimize the number of calculations it tries. The amount of new include_temp all over the spell is mildly 13:38:58 concerning, since it feels ripe for a future oversight. Is this mostly factoring out Heroism and Ash wrath and such (or am I overlooking something else obvious)? (Is it even correct for the calculator to do so, I wonder? It does talk about your 'present state' and some of those 'temp' effects last quite a while) It does feel like a real shame that the XL estimate doesn't work for gnolls and comes close to directly lying for djinni, though. Would be 13:38:58 nice if it were possible to improve on that in particular. 13:46:59 <11O​dds> Thanks! Glad it seems useful 🙂 Yeah, I did consider "inverting" some of the spell logic to work out the target... but as you say thought in the end that this was more likely to remain in sync and in fact this isn't ever going to be expensive (we only call it for an explicit UI action, and the search is pretty quick anyway) The include_temp stuff - I think it's better if e.g. heroism is ignored... this is a long term guide which 13:46:59 probably doesn't want to change with that (in the same way that XP estimates for skill targets don't). I'll do something about djinni and gnolls before merging; perhaps keep the same formulation of average spell school, and for the XP calculation tell Djinni how long it will take to get there with magic-only training, and gnolls how long it will take. The target level is somewhat misleading for both of those, because at least spellcasting will also 13:46:59 contribute in both cases, but I don't especially want to have them be told about more skills than all the other characters. 13:49:25 <04d​racoomega> Yeah, I think it's fine to ignore spellcasting, but like... it's still directly derivable it will take to arrive at a given skill level for each of those species - it's just not the same as any other 13:50:18 <04d​racoomega> (Ash wrath feels long enough term that I could see the argument that it should be included in this, but I guess I don't have strong feelings either way. It's certainly a wildly different scale of 'temp' than Heroism is!) 13:52:43 <11O​dds> Yeah, I just went by what you.skill considers temp, which does indeed bundle timescales together 13:53:50 <11O​dds> (Dilettante being temp is the other longer one there) 13:54:22 <11O​dds> I overall didn't find a configuration of temporariness that seemed completely obvious and natural, but this one seemed as good as any 13:55:21 <04d​racoomega> Yeah, it feels like a sort of compromise with reality 13:55:39 <04d​racoomega> Not perfect, but the other versions wouldn't be either 13:56:04 <04d​racoomega> (Like, talisman forms are really less temporary than god wrath in a lot of ways since you can leave them immediately. And yet...) 13:56:06 <11O​dds> Also if you are under Ash wrath being confused is thematic 😛 15:40:16 03CrawlOdds02 07* 0.35-a0-1314-g9d4aed78f7: Fix some errors and crashes in skill target estimation 10(14 hours ago, 3 files, 63+ 37-) 13https://github.com/crawl/crawl/commit/9d4aed78f77b 15:40:16 03CrawlOdds02 07* 0.35-a0-1315-gfdbd7b20ba: Show the effect of training skills on spell failure 10(8 days ago, 2 files, 135+ 55-) 13https://github.com/crawl/crawl/commit/fdbd7b20ba3a 15:40:16 03CrawlOdds02 07* 0.35-a0-1316-g32f959a6b0: Show the skills needed to get a spell castable 10(27 hours ago, 7 files, 175+ 40-) 13https://github.com/crawl/crawl/commit/32f959a6b094 15:43:45 Unstable branch on underhound.eu updated to: 0.35-a0-1313-g2a75ff33b5 (34) 16:10:59 <08n​icolae> fucken finally 16:11:17 New branch created: nicolae_0826_shoals_decor (3 commits) 13https://github.com/crawl/crawl/tree/nicolae_0826_shoals_decor 16:11:18 03nicolae02 07[nicolae_0826_shoals_decor] * 0.35-a0-819-g0436fd900f: Add decor vaults for Shoals, featuring a large number of subvaults 10(3 months ago, 1 file, 1049+ 0-) 13https://github.com/crawl/crawl/commit/0436fd900fc6 16:11:18 03nicolae02 07[nicolae_0826_shoals_decor] * 0.35-a0-820-ga081fec39d: Make some changes to the archipelago functions 10(8 weeks ago, 1 file, 14+ 3-) 13https://github.com/crawl/crawl/commit/a081fec39d46 16:11:18 03nicolae02 07[nicolae_0826_shoals_decor] * 0.35-a0-821-g1d1ad1ec18: Add more decor vaults to shoals 10(2 minutes ago, 2 files, 1836+ 1060-) 13https://github.com/crawl/crawl/commit/1d1ad1ec1810 16:14:49 <08n​icolae> not sure what exactly changed in the archipelago file in that last one that made it get entirely redone, but 16:26:55 04Build failed for 08nicolae_0826_shoals_decor @ 1d1ad1ec 06https://github.com/crawl/crawl/actions/runs/39691666758 18:22:39 <08n​icolae> it's occuring to me that i don't remember how to actually make that into a pull request 18:34:09 fwiw when I opened one of the commit URLs the branch they're on is a link that shows a diff against master and offers to make a PR 18:34:48 (specifically, https://github.com/crawl/crawl/compare/nicolae_0826_shoals_decor) 18:37:53 from the actual branch you get there by clicking "3 commits ahead of". if you're on your local checkout, I suggest installing `gh` and it's `gh pr create` 18:38:40 otherwise you push it up and use the web ui as I showed; often, shortly after new pushes there'll be a banner at the top of the web UI that offers to make a PR 18:39:02 ("new pushes to " IIRC) 22:37:05 Unstable branch on crawl.develz.org updated to: 0.35-a0-1316-g32f959a6b0 (34) 23:02:47 Windows builds of master branch on crawl.develz.org updated to: 0.35-a0-1316-g32f959a6b0 23:41:06 Unstable branch on cbro.berotato.org updated to: 0.35-a0-1316-g32f959a6b0 (34) 23:58:33 Monster database of master branch on crawl.develz.org updated to: 0.35-a0-1316-g32f959a6b0