thunderseethe
@thunderseethe.dev
There's a neat bonus post included because I realized I need to rewrite type inference to support the full compiler 🥹
November 12, 2025 at 2:48 PM
There's a neat bonus post included because I realized I need to rewrite type inference to support the full compiler 🥹
It also ties into the Handle (previously Claim) work: smallcultfollowing.com/babysteps/bl...
The Handle trait · baby steps
smallcultfollowing.com
October 7, 2025 at 8:35 PM
It also ties into the Handle (previously Claim) work: smallcultfollowing.com/babysteps/bl...
I cant wait for them to update the book. I waste so much time reading through the proc macro source to figure out how the new salsa works. Interned structs now have a 'db lifetime. Guess who fixed up their whole codebase before realizing they also have a `no_lifetime` attribute 😭
October 3, 2025 at 4:16 PM
I cant wait for them to update the book. I waste so much time reading through the proc macro source to figure out how the new salsa works. Interned structs now have a 'db lifetime. Guess who fixed up their whole codebase before realizing they also have a `no_lifetime` attribute 😭
That's a fair clarification. I shouldn't say easy to box anything, but easy to copy anything.
September 25, 2025 at 5:53 PM
That's a fair clarification. I shouldn't say easy to box anything, but easy to copy anything.
Usually it means the payload of the Sum type is laid out adjacent to the discriminate, rather than behind a pointer. A lot of languages autobox sum type payloads because it makes layout easy. An enum is always a discriminant and a pointer
September 25, 2025 at 5:22 PM
Usually it means the payload of the Sum type is laid out adjacent to the discriminate, rather than behind a pointer. A lot of languages autobox sum type payloads because it makes layout easy. An enum is always a discriminant and a pointer
Swift! Sort of. It will layout sum types unboxed, but it's very easy to accidentally box anything in Swift, if you hold it wrong.
September 25, 2025 at 5:04 PM
Swift! Sort of. It will layout sum types unboxed, but it's very easy to accidentally box anything in Swift, if you hold it wrong.
No you don't understand. My Rust isn't gonna have a borrow checker. Instead, it'll have something that does 80% of what a borrow checker does and is memory safe 80% of the time. We trust programmers to handle the 20% safely
September 19, 2025 at 6:15 PM
No you don't understand. My Rust isn't gonna have a borrow checker. Instead, it'll have something that does 80% of what a borrow checker does and is memory safe 80% of the time. We trust programmers to handle the 20% safely
I love to hear that! But it doesn't do a great job actually explaining ML modules. It's mostly about how we keep reinventing ML module-like features because folks are intimidated by ML modules
September 6, 2025 at 8:38 PM
I love to hear that! But it doesn't do a great job actually explaining ML modules. It's mostly about how we keep reinventing ML module-like features because folks are intimidated by ML modules
Could be one of a couple things:
1) ml structs and signatures add verbosity (over "normal" modules)
2) applying functors explicitly everywhere
3) excessive proliferation of type sharing constraints
1) ml structs and signatures add verbosity (over "normal" modules)
2) applying functors explicitly everywhere
3) excessive proliferation of type sharing constraints
September 6, 2025 at 8:20 PM
Could be one of a couple things:
1) ml structs and signatures add verbosity (over "normal" modules)
2) applying functors explicitly everywhere
3) excessive proliferation of type sharing constraints
1) ml structs and signatures add verbosity (over "normal" modules)
2) applying functors explicitly everywhere
3) excessive proliferation of type sharing constraints
Its crazy how much syntax sugar WAT has for a target that is for the most part only ever generated (either from wasm, or from a compiler). Also crazy how the sugar is mostly undocumented. I didn't discover WAT could nest expressions until I saw `wasmprinter` has a fold instructions option
September 2, 2025 at 2:53 PM
Its crazy how much syntax sugar WAT has for a target that is for the most part only ever generated (either from wasm, or from a compiler). Also crazy how the sugar is mostly undocumented. I didn't discover WAT could nest expressions until I saw `wasmprinter` has a fold instructions option
I've never needed it, so quite possible you just don't. Maybe it's helpful when `<array>` is a compound expression and you don't want to have to let bind it. I could see that if the map was in the middle of an iterator chain
August 30, 2025 at 2:39 PM
I've never needed it, so quite possible you just don't. Maybe it's helpful when `<array>` is a compound expression and you don't want to have to let bind it. I could see that if the map was in the middle of an iterator chain
A caveat is that I'd only recommend the 10in+ models and only if your usecase is reading/writing PDFs. If you're looking for a more typical epub usage, their smaller models are pricey and more than what you need. I hear good things about Kobo in that space and that they support sideloading via USB
August 27, 2025 at 5:47 PM
A caveat is that I'd only recommend the 10in+ models and only if your usecase is reading/writing PDFs. If you're looking for a more typical epub usage, their smaller models are pricey and more than what you need. I hear good things about Kobo in that space and that they support sideloading via USB
I've really enjoyed Boox, they're pricey but run android so you can do whatever on them and aren't locked to any store. Folks have mentioned durabilityissues but I haven't experienced that personally and I've dropped them on concrete.
August 27, 2025 at 5:45 PM
I've really enjoyed Boox, they're pricey but run android so you can do whatever on them and aren't locked to any store. Folks have mentioned durabilityissues but I haven't experienced that personally and I've dropped them on concrete.
Its like they wanted the veneer of a modern feature set but at every fork in the road they veered towards C. I am quite biased though
August 25, 2025 at 3:12 PM
Its like they wanted the veneer of a modern feature set but at every fork in the road they veered towards C. I am quite biased though
Swift is the poster child of this for me. We'll give you enums but only a switch statement, no switch expression. Years later we'll add half a switch expression that's only allowed in specific positions and no one will use it because "explicit switch statements are more readable" 😩
August 25, 2025 at 3:10 PM
Swift is the poster child of this for me. We'll give you enums but only a switch statement, no switch expression. Years later we'll add half a switch expression that's only allowed in specific positions and no one will use it because "explicit switch statements are more readable" 😩
I would expect A in a lazy or strict language but would expect optimizations to float the let in a lazy lang (or a strict languages if launchMissiles() is side effect free somehow)
August 25, 2025 at 2:56 PM
I would expect A in a lazy or strict language but would expect optimizations to float the let in a lazy lang (or a strict languages if launchMissiles() is side effect free somehow)
Does this mean datalog is about "and" and "or" or "and" or "or"
August 13, 2025 at 3:39 PM
Does this mean datalog is about "and" and "or" or "and" or "or"
But the main idea is that modules have a concept of a typed hole and you can mix two modules together to fill each other's holes. This allows for a low friction combining of modules in a kind of duck-tyoed fashion where two modules mix when the types line up
August 12, 2025 at 8:52 PM
But the main idea is that modules have a concept of a typed hole and you can mix two modules together to fill each other's holes. This allows for a low friction combining of modules in a kind of duck-tyoed fashion where two modules mix when the types line up
people.mpi-sws.org/~rossberg/mi... is the seminal paper on it. But I don't love how it only covers generative functors.
plv.mpi-sws.org/backpack/ covers a system with applicative modules but is overly tied into Haskell's existing module system.
plv.mpi-sws.org/backpack/ covers a system with applicative modules but is overly tied into Haskell's existing module system.
MixML - Mixin' Up the ML Module System
Be declarative. Be functional. Just be.
people.mpi-sws.org
August 12, 2025 at 8:50 PM
people.mpi-sws.org/~rossberg/mi... is the seminal paper on it. But I don't love how it only covers generative functors.
plv.mpi-sws.org/backpack/ covers a system with applicative modules but is overly tied into Haskell's existing module system.
plv.mpi-sws.org/backpack/ covers a system with applicative modules but is overly tied into Haskell's existing module system.
It's a downer. I believe you can recover the commutativity with dynamic label generation. Then each effect has a unique name and disjointedness comes for free. A very heavy handed solution though
August 12, 2025 at 3:09 AM
It's a downer. I believe you can recover the commutativity with dynamic label generation. Then each effect has a unique name and disjointedness comes for free. A very heavy handed solution though
This looks to be a link to Gat example that breaks cycle checking. I see what you're saying though. Infinite types are still allowed under `Ind<T: Sized>` if the field is a `Box<T>`.
July 28, 2025 at 7:27 PM
This looks to be a link to Gat example that breaks cycle checking. I see what you're saying though. Infinite types are still allowed under `Ind<T: Sized>` if the field is a `Box<T>`.