Home liberachat/#haskell: Logs Calendar

Logs: liberachat/#haskell

←Prev  Next→ 1,800,848 events total
2025-12-28 05:39:17 <iqubic> Essentially, I'm trying to model simple combinatorial games like "nim" where I can write a function like "nextMoves :: a -> [a]", but then I realized that different games might have the type for the state variable.
2025-12-28 05:39:28 <ski> yea, sounds like record of operations is more sensible, then
2025-12-28 05:40:43 × rekahsoft quits (~rekahsoft@70.51.99.245) (Remote host closed the connection)
2025-12-28 05:40:46 <ski> btw, sometimes it makes sense to make a function from a record-of-operations, to another (possibly existential) record-of-operations. this being one way to do something similar to the parameterized modules in the module system of the MLs (SML,OCaml) (cf. Backpack)
2025-12-28 05:41:11 <iqubic> Yeah, I get what you mean.
2025-12-28 05:41:16 rekahsoft joins (~rekahsoft@70.51.99.245)
2025-12-28 05:41:29 <haskellbridge> <slack1256> I liked backpack...
2025-12-28 05:42:16 merijn joins (~merijn@host-cl.cgnat-g.v4.dfn.nl)
2025-12-28 05:42:45 <ski> you'd open the returned record in a scope, bringing the skolem into scope, and then use the operations directly. this is a way to do Abstract Data Types
2025-12-28 05:43:03 <iqubic> In the basic game of "nim" you have a bunch of distinct piles, and each pile has a bunch of objects. Players take turns removing any amount of objects from a single pile. The winner is the player to make the last valid move. In other words, if you can't make a move, you lose.
2025-12-28 05:43:34 <ski> (unlike doing OO via existentials, where you'd open the record, apply an operation to the current state, and then rewrap new state with the operations again, every time)
2025-12-28 05:44:52 <iqubic> I can model this as with a state of [Int]. But if I want a similar game, but with the constraint of "after each move, the piles must all have a coprime number of objects", that's also a state of type "[Int]" but a different "nextMoves :: a -> [a]" function.
2025-12-28 05:45:16 <iqubic> BTW, Conway et al. call that version "prim", I think.
2025-12-28 05:45:41 <haskellbridge> <slack1256> These combinatorial games are not related in anyway right? They just happen to implement a nextMoves function, the lack of laws pushes you to record of functions
2025-12-28 05:46:55 × merijn quits (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 240 seconds)
2025-12-28 05:46:56 <iqubic> Yeah, correct.
2025-12-28 05:47:12 <iqubic> It's mainly just the "nextMoves" function that relates them.
2025-12-28 05:47:24 <iqubic> And I have no laws, because I have just the one function.
2025-12-28 05:47:42 <haskellbridge> <slack1256> Right, which is also a non obvious function
2025-12-28 05:48:54 <ski> presumably iterating `nextMoves' is meant to lead to a tree where no path is infinite ?
2025-12-28 05:49:23 <iqubic> Correct. I'm only interested in finite games here.
2025-12-28 05:49:33 <ski> some kind of well-founded order thing
2025-12-28 05:51:06 <haskellbridge> <slack1256> Multiple implementations for that in no way related to the days structure to give an instance
2025-12-28 05:51:12 <ski> all the outputs of `nextMoves' are presumably strictly less than the input
2025-12-28 05:51:17 <iqubic> Basically, I've been reading the book "Winning Ways For Your Mathematical Plays" by Berlekamp, Guy, and Conway, and I want to implement some of these things in Haskell.
2025-12-28 05:51:17 <haskellbridge> <slack1256> Data*
2025-12-28 05:51:33 <iqubic> ski: For some definition of "strictly less", yeah.
2025-12-28 05:52:02 <ski> yea, some arbitrary partial order
2025-12-28 05:52:30 <ski> (that is well-founded)
2025-12-28 05:52:47 <iqubic> But my main question was "Do I model this as a typeclass with one function or a record of one function"?
2025-12-28 05:53:12 <ski> yep. the latter seems more reasonable
2025-12-28 05:53:45 <ski> (you could put the order relation as a second operation in the record .. but perhaps you have no use for it, at all, at run-time)
2025-12-28 05:54:08 <iqubic> I agree. Because different games with completely different rules might want to use similar state representations.
2025-12-28 05:54:30 <iqubic> ski: Currently my planned API doesn't need a way to tell which states are smaller.
2025-12-28 05:55:01 <ski> in any case, even without an explicit order relation (computationally irrelevant or not), you should still be able to express a well-foundedness law for `nextMoves'
2025-12-28 05:55:20 <ski> (iow, you don't need more than one operation, to be able to have a law)
2025-12-28 05:55:43 <iqubic> What do you mean by well-foundedness?
2025-12-28 05:58:01 merijn joins (~merijn@host-cl.cgnat-g.v4.dfn.nl)
2025-12-28 05:58:38 <ski> for a strict order relation `<', a property like `forall P. (forall x. (forall y. y < x -> P y) -> P x) -> forall x. P x'. if we were talking about natural numbers, this would be called "strong induction" : you can prove a property `P' holds for all `x', if you can prove it holds for all `x', under the assumption it holds for all `y' that are strictly less than `x'
2025-12-28 05:59:14 <iqubic> Right I see.
2025-12-28 06:00:53 <ski> in terms of `nextMoves', instead of `forall y. y < x -> P y', we'd say `forall y. elem y (nextMoves x) = True -> P y'. iow, assuming that it holds for all the possible next states, it should be able to show it holds for the current state
2025-12-28 06:02:25 × merijn quits (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 245 seconds)
2025-12-28 06:02:39 <iqubic> forall a. let a' = nextMoves a in all [ length (nextMoves opt) < length a' | opt <- nextMoves a]
2025-12-28 06:02:51 <iqubic> Or something like that.
2025-12-28 06:03:27 <iqubic> Basically, if you keep iterating nextMoves function, you should keep getting less and less options.
2025-12-28 06:03:55 × jmcantrell quits (~weechat@user/jmcantrell) (Ping timeout: 244 seconds)
2025-12-28 06:07:06 <ski> not necessarily
2025-12-28 06:08:21 peterbecich joins (~Thunderbi@71.84.33.135)
2025-12-28 06:08:22 <ski> a tree that at each node at depth `n' has either zero or `n' branches of that node could still be well-founded
2025-12-28 06:09:17 <ski> the important part is that each state is getting closer to a minimal element (a leaf, one which has zero successors), on each step
2025-12-28 06:09:53 <ski> you could have the branching explode, and still have finite depth for every path
2025-12-28 06:10:16 <iqubic> Yeah. But in my case, I want you to always have strictly less moves as you go along. That's the main case I'm interested in exploring.
2025-12-28 06:10:47 <ski> (there's not even a need for the branching to be finite, it could still be well-founded. although, possibly, in your case, you're only interested in finite branching)
2025-12-28 06:11:02 <ski> ah, okay
2025-12-28 06:13:25 merijn joins (~merijn@host-cl.cgnat-g.v4.dfn.nl)
2025-12-28 06:18:24 × merijn quits (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 260 seconds)
2025-12-28 06:29:12 merijn joins (~merijn@host-cl.cgnat-g.v4.dfn.nl)
2025-12-28 06:34:07 × merijn quits (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 246 seconds)
2025-12-28 06:35:20 × synchromesh quits (~john@2406:5a00:2412:2c00:7842:6802:4767:2e5b) (Read error: Connection reset by peer)
2025-12-28 06:36:46 synchromesh joins (~john@2406:5a00:2412:2c00:7842:6802:4767:2e5b)
2025-12-28 06:40:06 <iqubic> Essentially, if a level of my tree has N members, then all children there should have strictly less than N descendants, except for the root.
2025-12-28 06:40:54 <iqubic> Except, it's more like a DAG, because there might be multiple possible ways to get from the current state to some future state.
2025-12-28 06:42:24 <iqubic> But this mathematical theory has little to do with Haskell at this point and I have gotten my orginal design question answered already.
2025-12-28 06:45:19 <hololeap> are there any language extensions now days that can let you write a Functor instance for Set? some way to inject a constraint requirement into an existing function, or something?
2025-12-28 07:02:17 merijn joins (~merijn@host-cl.cgnat-g.v4.dfn.nl)
2025-12-28 07:06:57 × merijn quits (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 244 seconds)
2025-12-28 07:13:20 × euphores quits (~SASL_euph@user/euphores) (Quit: Leaving.)
2025-12-28 07:18:05 merijn joins (~merijn@host-cl.cgnat-g.v4.dfn.nl)
2025-12-28 07:23:15 × merijn quits (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 252 seconds)
2025-12-28 07:29:18 euphores joins (~SASL_euph@user/euphores)
2025-12-28 07:31:12 × sroso quits (~sroso@user/SrOso) (Quit: Leaving :))
2025-12-28 07:33:52 merijn joins (~merijn@host-cl.cgnat-g.v4.dfn.nl)
2025-12-28 07:38:22 × merijn quits (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 244 seconds)
2025-12-28 07:45:32 <[exa]> hololeap: you can wrap the set in a newtype that imposes the constraint, with similar tradeoffs like with other newtype wraps (inconvenient but coerces well) and typeclass constraints on data types (generally hated)
2025-12-28 07:49:00 <[exa]> iqubic: people usually have these "operation records" for various backend implementations for whatever, e.g. here: https://hackage-content.haskell.org/package/tls-2.1.13/docs/Network-TLS.html#t:Backend or here: https://hackage.haskell.org/package/selda-0.5.2.0/docs/Database-Selda-Backend.html#t:SeldaBackend . TBH quite easy to go that way, and very easy to slap on a typeclass later, in case you
2025-12-28 07:49:02 <[exa]> decide you want to infer which backend to use where.
2025-12-28 07:49:15 <hololeap> hm, ok. I was looking into a different instance of Apply for Map (different than the one defined in semigroupoids) and ultimately it requires (Ord k, Semigroup a) => Map k a
2025-12-28 07:49:19 merijn joins (~merijn@host-cl.cgnat-g.v4.dfn.nl)
2025-12-28 07:50:04 <hololeap> but Apply is a class for Type -> Type
2025-12-28 07:50:19 <[exa]> hololeap: why though (kinda curious, and also usually this has been a common instance of xyz issues)
2025-12-28 07:51:15 <hololeap> mostly just out of curiosity. the current Apply instance feels very niche since it uses Map.intersectionWith under the hood
2025-12-28 07:53:26 <hololeap> I was looking into an instance that would use something like Map.unionWith (<>)
2025-12-28 07:53:39 <hololeap> though not exactly that
2025-12-28 07:53:47 <Leary> That sounds like `Semigroup` instance instead.
2025-12-28 07:53:58 × merijn quits (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 244 seconds)
2025-12-28 07:54:39 × ystael quits (~ystael@user/ystael) (Ping timeout: 260 seconds)
2025-12-28 07:55:04 <[exa]> hololeap: that can't do applicative though, right? (you'd need a default function and default parameter)
2025-12-28 07:55:26 <hololeap> right, Apply is Applicative sans `pure`,
2025-12-28 07:55:30 <[exa]> so yeah that would be a (specialized) semigroup
2025-12-28 07:55:47 <hololeap> just the <*> part of Applicative
2025-12-28 07:55:50 <[exa]> no in this case even `ap` would require the defaults
2025-12-28 07:56:47 ystael joins (~ystael@user/ystael)
2025-12-28 07:57:06 <hololeap> anyway, this wasn't really a problem I needed to solve, just an exploration of ideas
2025-12-28 07:58:58 <hololeap> @hackage quickcheck-classes -- makes it easy to verify the Apply laws
2025-12-28 07:58:58 <lambdabot> https://hackage.haskell.org/package/quickcheck-classes -- makes it easy to verify the Apply laws
2025-12-28 08:00:01 × tt1231607 quits (~tt1231@75.185.104.199) (Quit: The Lounge - https://thelounge.chat)
2025-12-28 08:03:17 merijn joins (~merijn@host-cl.cgnat-g.v4.dfn.nl)
2025-12-28 08:05:33 tt1231607 joins (~tt1231@75.185.104.199)
2025-12-28 08:08:07 × merijn quits (~merijn@host-cl.cgnat-g.v4.dfn.nl) (Ping timeout: 240 seconds)
2025-12-28 08:19:04 merijn joins (~merijn@host-cl.cgnat-g.v4.dfn.nl)

All times are in UTC.