Alice ✨
@welltypedwit.ch
There’s no sense crying over every mistake. You just keep on trying ’til you run out of cake
- GLaDOS
https://welltypedwit.ch
- GLaDOS
https://welltypedwit.ch
"hmm, i don't have a way yet to look up the representation of a data constructor from core. i wonder how i did that during core generation"
![Vega.DataConstructor _ name -> do
-- TODO: THIS IS WRONG. It's just a temporary fix to get Nil working.
-- To do this correctly, we need a GraphPersistence hook to look up the arity of a data constructor
-- and desugar this to a lambda taking that many parameters
pure ([], Core.DataConstructorApplication (Core.UserDefinedConstructor name) [], undefined)](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:j7ri4agbraq6g3dfmyccxep2/bafkreicwiglvqroo2lo34hnfi4jhumlerqo3z35ejnrqwuga3v4tn2lx7a@jpeg)
December 26, 2025 at 9:48 PM
"hmm, i don't have a way yet to look up the representation of a data constructor from core. i wonder how i did that during core generation"
"huh, i wonder what kind of code ghc's llvm backend generates here"
ah
ah
December 26, 2025 at 12:24 AM
"huh, i wonder what kind of code ghc's llvm backend generates here"
ah
ah
i really feel like gadts are a local maximum (if even that)
i'm not entirely sure what i would prefer instead but they're just such a messy indirect combination of a few very different features
maybe decoupling them would make them less dangerous so you don't immediately code yourself into a corner
i'm not entirely sure what i would prefer instead but they're just such a messy indirect combination of a few very different features
maybe decoupling them would make them less dangerous so you don't immediately code yourself into a corner
December 25, 2025 at 5:13 AM
i really feel like gadts are a local maximum (if even that)
i'm not entirely sure what i would prefer instead but they're just such a messy indirect combination of a few very different features
maybe decoupling them would make them less dangerous so you don't immediately code yourself into a corner
i'm not entirely sure what i would prefer instead but they're just such a messy indirect combination of a few very different features
maybe decoupling them would make them less dangerous so you don't immediately code yourself into a corner
for how much they talk about it functional programmers really don't understand hindley milner very well
December 25, 2025 at 5:02 AM
for how much they talk about it functional programmers really don't understand hindley milner very well
Reposted by Alice ✨
i'm really not convinced that having more than one compiler is automatically good for a language
December 23, 2025 at 4:12 PM
i'm really not convinced that having more than one compiler is automatically good for a language
til that HLS shows type definitions inline if you hover over them while holding ctrl! this is so nice omg since when does it do that?
December 23, 2025 at 12:35 AM
til that HLS shows type definitions inline if you hover over them while holding ctrl! this is so nice omg since when does it do that?
can programmers please behave like adults for once
December 22, 2025 at 5:36 PM
can programmers please behave like adults for once
does anyone know any other examples of using higher rank types for regions/lifetimes like with runST that aren't ST or bluefin? especially ones related to allocating something off-heap or similar (but anything really. doesn't even have to be in haskell)
December 21, 2025 at 8:28 PM
does anyone know any other examples of using higher rank types for regions/lifetimes like with runST that aren't ST or bluefin? especially ones related to allocating something off-heap or similar (but anything really. doesn't even have to be in haskell)
Reposted by Alice ✨
The Subtle Footgun of TVar (Map _ _)
How coarse-grained STM containers can livelock under load
www.parsonsmatt.org
December 18, 2025 at 9:39 PM
everyone already has an intuition for the pipe operator! it's just like unix pipes except for how it's not actually like unix pipes at all, has different semantics and serves a completely different purpose
December 18, 2025 at 12:34 AM
everyone already has an intuition for the pipe operator! it's just like unix pipes except for how it's not actually like unix pipes at all, has different semantics and serves a completely different purpose
okay, imagine this: ocaml with just basic data types, required mli files (and a type checker that uses the types from signatures), the same type syntax in signatures and implementations, only locally abstract types (no 'a or _ silliness)...
December 17, 2025 at 4:07 AM
okay, imagine this: ocaml with just basic data types, required mli files (and a type checker that uses the types from signatures), the same type syntax in signatures and implementations, only locally abstract types (no 'a or _ silliness)...
i genuinely think ocaml would be a dramatically better language if most of its advanced type features did not exist
December 17, 2025 at 3:28 AM
i genuinely think ocaml would be a dramatically better language if most of its advanced type features did not exist
i just love it when i write down the type ocaml infers for a let binding and get a type error because `constraint` is just so incredibly well thought out omg
December 17, 2025 at 3:00 AM
i just love it when i write down the type ocaml infers for a let binding and get a type error because `constraint` is just so incredibly well thought out omg
i hate eio's api
great implementation, horrible api
great implementation, horrible api
![File "src/pipe.ml", line 1:
Error: The implementation src/pipe.pp.ml
does not match the interface src/pipe.pp.mli:
Values do not match:
val compose :
sw:Eio__core.Switch.t ->
mgr:[> ([> `Generic ] as 'a) Eio.Process.mgr_ty ] Eio.Resource.t ->
env:string array ->
(string * string list) list -> 'a Eio.Process.ty Eio.Resource.t
is not included in
val compose :
sw:Eio__core.Switch.t ->
mgr:[> [> `Generic ] Eio.Process.mgr_ty ] Eio.Resource.t ->
env:string array ->
(string * string list) list ->
[> [> `Generic ] Eio.Process.ty ] Eio.Resource.t
The type
sw:Eio__core.Switch.t ->
mgr:[> `Platform of [> `Generic ] as 'a | `Process_mgr ]
Eio.Resource.t ->
env:string array ->
(string * string list) list -> 'a Eio.Process.ty Eio.Resource.t
is not compatible with the type
sw:Eio__core.Switch.t ->
mgr:[> [> `Generic ] Eio.Process.mgr_ty ] Eio.Resource.t ->
env:string array ->
(string * string list) list ->
[> [> `Generic ] Eio.Process.ty ] Eio.Resource.t
Type
([> `Generic ] as 'a) Eio.Process.ty =
[ `Platform of 'a | `Process ]
is not compatible with type
[> ([> `Generic ] as 'b) Eio.Process.ty ] =
[> `Platform of 'b | `Process ]
The second variant type is open and the first is not
File "src/pipe.mli", lines 3-8, characters 0-17: Expected declaration
File "src/pipe.ml", line 32, characters 4-11: Actual declaration](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:j7ri4agbraq6g3dfmyccxep2/bafkreihnzw6i4loyugou6fx2wrgk5jkgjzoonpx3rate3tya5yjd7lc4s4@jpeg)
December 17, 2025 at 2:43 AM
i hate eio's api
great implementation, horrible api
great implementation, horrible api
use this ONE WEIRD TRICK to implement a runtime feature by only changing the type checker (compilers HATE her)
(you just need to have implemented it in the runtime already at least 2 years ago but have forgotten to implement the two lines of type checker logic oops)
(you just need to have implemented it in the runtime already at least 2 years ago but have forgotten to implement the two lines of type checker logic oops)
December 17, 2025 at 2:26 AM
use this ONE WEIRD TRICK to implement a runtime feature by only changing the type checker (compilers HATE her)
(you just need to have implemented it in the runtime already at least 2 years ago but have forgotten to implement the two lines of type checker logic oops)
(you just need to have implemented it in the runtime already at least 2 years ago but have forgotten to implement the two lines of type checker logic oops)
> printf "a\nb\nc" | wc -l
2
welp
2
welp
December 17, 2025 at 2:12 AM
> printf "a\nb\nc" | wc -l
2
welp
2
welp
why do you need a separate ocaml compiler for flambda anyway? couldn't you just have one that supports both backends?
December 17, 2025 at 1:47 AM
why do you need a separate ocaml compiler for flambda anyway? couldn't you just have one that supports both backends?
i understand that minecraft is not exactly the most realistic game but like, how do you spend the same amount of wood on a single shelf as on 8 cubic meters of planks
December 16, 2025 at 11:49 PM
i understand that minecraft is not exactly the most realistic game but like, how do you spend the same amount of wood on a single shelf as on 8 cubic meters of planks
Reposted by Alice ✨
Extremely disappointed in the Rust community on this one. Changing your name online should be a very understandable thing to want to do in a trans and neurodivergent inclusive community. For example, having a publicly visible link between your name and deadname can cause real harm. 1/7
December 16, 2025 at 10:52 PM
Extremely disappointed in the Rust community on this one. Changing your name online should be a very understandable thing to want to do in a trans and neurodivergent inclusive community. For example, having a publicly visible link between your name and deadname can cause real harm. 1/7
it's interesting how often the `open (struct ... end)` trick comes up on the ocaml discourse considering that it's totally broken and unsound lol
December 16, 2025 at 5:16 AM
it's interesting how often the `open (struct ... end)` trick comes up on the ocaml discourse considering that it's totally broken and unsound lol
i would like a 4-list-colorable apple please. they have played us for absolute fools
December 15, 2025 at 3:07 AM
i would like a 4-list-colorable apple please. they have played us for absolute fools
i'm feeling very validated in only monomorphizing on representations by default in vega right now.
it's actually really not great that newtypes have a compile time and (more complicated) runtime cost in rust!
it's actually really not great that newtypes have a compile time and (more complicated) runtime cost in rust!
thanks! okay this took me a while (in part because woofwoof didn't build. i opened an issue) but i wrote a script that replaces every type with its representation and that took it down from 50 to 33 unique instantiations (38 if we don't include that String has the same representation as Vec)
count-representation-instantiations.pls
GitHub Gist: instantly share code, notes, and snippets.
gist.github.com
December 14, 2025 at 11:10 PM
i'm feeling very validated in only monomorphizing on representations by default in vega right now.
it's actually really not great that newtypes have a compile time and (more complicated) runtime cost in rust!
it's actually really not great that newtypes have a compile time and (more complicated) runtime cost in rust!
man, <REDACTED> might be the only programming language i'm physically unable to take seriously (and like, i know about v)
like, i'm trying but literally everything i learn about it is just so dumb omg
like, i'm trying but literally everything i learn about it is just so dumb omg
December 13, 2025 at 11:09 AM
man, <REDACTED> might be the only programming language i'm physically unable to take seriously (and like, i know about v)
like, i'm trying but literally everything i learn about it is just so dumb omg
like, i'm trying but literally everything i learn about it is just so dumb omg
Reposted by Alice ✨
your next game should be written in #haskell and represent life points and money and other state in peano numeral. this makes your game cheat engine proof cause who the heck knows what a haskell heap looks like
December 12, 2025 at 11:54 PM
your next game should be written in #haskell and represent life points and money and other state in peano numeral. this makes your game cheat engine proof cause who the heck knows what a haskell heap looks like