iiinova — where your digital self keeps evolving

Basile Henry @basilehenry.bsky.social · 11/07/2026

My main use case for the either crate is to return an impl Iterator 😅

000

Good question! I have never tried PR stacking across forks, and it very well might not be supported by GitHub 🤔 Hopefully they think about that use case in their upcoming tool/ui updates 🤞

Combined with `git rebase --interactive --update-refs` (jj is probably even better), I can move commits between PRs, or add/remove commits halfway down the stack quite easily 😊

000

Right, with my suggested workflow (that I/my coworkers have been using for a while) "the first N commits are actually from another PR" is not true, each PR only shows the commits and diff for that PR. Each PR "targets" the previous PR instead of main, but only gets merged if/when it targets main 😊

200

Break large changes into small, reviewable, stacked pull requests with first-class GitHub support.

The part that confused me is that GitHub already lets you do stacked PRs 🤔 If you set the base branch of each PR to the previous PR in the stack, it'll even change the base to main automatically after the previous PR gets merged. Looks like the UI might get improved: github.github.com/gh-stack/

github.github.com

GitHub Stacked PRs

100

Basile Henry @basilehenry.bsky.social · 03/07/2026

I thought update-refs was the missing piece for your stacked PR workflow, but maybe it's on the GitHub/other side? 🤔

100

Basile Henry @basilehenry.bsky.social · 03/07/2026

You might be interested in `git rebase --update-refs` if you don't already know about it. Very useful to rebase a stack of PRs 😊 git-scm.com/docs/git-reb...

git-scm.com

Git - git-rebase Documentation

100

Basile Henry @basilehenry.bsky.social · 02/07/2026

There are multiple tools that do it automatically with different trade offs (you probably already know). I found crate2nix works fairly well on a big monorepo with many crates. Good luck with the non-nix option! I'm also interested 😊

100

Basile Henry @basilehenry.bsky.social · 26/06/2026

return void 🤔 But to your point, isn't that what protected does compared to private? Or is there another use case for it? (I haven't used one of the OOP lang in a while...)

100

Basile Henry @basilehenry.bsky.social · 03/06/2026

flakeless but still pinned? 😅 I definitely still have a bunch of niv based repos...

Basile Henry @basilehenry.bsky.social · 26/05/2026

But if you put it as the last argument you get to use the "reader" monad instance: `Monad ((->) r)` so you can pass around the continuation silently and look smart 🤓

Basile Henry @basilehenry.bsky.social · 21/05/2026

The data structure: `Maybe a` 😅

Basile Henry @basilehenry.bsky.social · 07/05/2026

A while ago I made one which gave semantics to git conflict markers 😈

040

One tool call that amazed me a while ago went something like `print "y\ny\nn\ny\nn\n" | git add -p` 😅

030

Sounds similar to what antithesis does under the hood as described on the Signal and Threads podcast 🤔

Right 👍 So the minimal feature that makes row polymorphism is the ability to write code generic on the rest of the fields that your function doesn't care about? In a monomorphising compiler that would give you one type per unique record 🤔

000

Basile Henry @basilehenry.bsky.social · 05/05/2026

I think row polymorphism let's you add (and sometimes remove) fields from a generic record: { r } -> { r | a: T } But most of the time you don't actually need this

100

Fair! I thought maybe they'd release their weights and the differences would show up on most inputs to give such a different error rate. Sometimes you can also witness differences on random weights, but then you don't really know if they matter as much once trained

000

Basile Henry @basilehenry.bsky.social · 05/05/2026

So it was a CPU vs GPU issue 😅 bsky.app/profile/basi... At least it wasn't something as subtle as f32 operations aren't commutative and different platforms compute in different order 🙃

100

Basile Henry @basilehenry.bsky.social · 03/05/2026

I'd be interested to see pytorch on the CPU 🤔

000

Basile Henry @basilehenry.bsky.social · 19/04/2026

That is pretty messed up 🤯 So if you go sail around the world you better be back under 3 years?

100

Basile Henry @basilehenry.bsky.social · 19/04/2026

In all seriousness, I don't know anything about boat VAT in the EU and I understand it might be messed up even if you don't own/plan on owning one 🤔

100

Basile Henry @basilehenry.bsky.social · 19/04/2026

Why do you care about VAT on boats if you don't have any? 😅

100

Basile Henry @basilehenry.bsky.social · 18/04/2026

Is that... one of the rare use cases for a bloom filter? 🤔

Basile Henry @basilehenry.bsky.social · 07/04/2026

Maybe you could have a go to definition for the impl function itself go to the class definition? 🤔 So from any use site you're one hop away from 2) and two hops away from 1)

Basile Henry @basilehenry.bsky.social · 29/03/2026

I'm pretty sure there will always be something more useful to spend the energy on, even if highly inefficient, like desalination 🤔

000

Basile Henry @basilehenry.bsky.social · 16/03/2026

With Nat you can encode constraints without extra comparison support: head :: Vec (n + 1) a -> a But if you support comparison constraints I think this is nicer to use in practice anyway: head :: (1 <= n) => Vec n a -> a

041

Basile Henry @basilehenry.bsky.social · 05/03/2026

Is that because for B you could take a reference to b.1 and the layout has to match (u8, u16)? 🤔

Basile Henry @basilehenry.bsky.social · 18/02/2026

You could learn Haskell while employed: myrtle.ai/about/careers/ 😄 We use Rust, Haskell, nix (and Python, Verilog, ...) for compilers and customs FPGA circuits

myrtle.ai

Careers

We are always looking for brilliant minds to join our team. If you have experience in Computer Science, Maths, or ML, we’d love to hear from you!

030

Basile Henry @basilehenry.bsky.social · 08/02/2026

So maybe we should be vibe coding in C to be consistent in our use of UB throughout the abstractions 😅

Basile Henry @basilehenry.bsky.social · 19/01/2026

Probably a rhetorical "what" 🤔 It basically says that `[a]` is equivalent to a function `Int -> a`. It's effectively the indexing function. If you want to set a value in the "array", you create a new function that checks if the input is the index you're setting and use the old function otherwise.

Basile Henry @basilehenry.bsky.social · 17/12/2025

Lots of GADTs when writing Clash Haskell 😭 clash-lang.org

clash-lang.org

Home

Basile Henry @basilehenry.bsky.social · 10/12/2025

"Linux in the Kernel" 👀 How dare they! 😅

090

Basile Henry @basilehenry.bsky.social · 09/12/2025

That is a pretty bad error message 😅 I assumed the Generic instance was a given and was already reaching for orphan instances to explain the lack of import bsky.app/profile/basi...

000

Basile Henry @basilehenry.bsky.social · 09/12/2025

Maybe you're also supposed to derive Generic for that type in the shame module (so it's not an orphan instance), and that would force you to add the import 🤔

000

Basile Henry @basilehenry.bsky.social · 09/12/2025

Ouch 😬 That would make sense to me for TemplateHaskell generated code, but for this type of instance deriving/default impl it feels like the extension should manage its own imports 😅

100

Basile Henry @basilehenry.bsky.social · 08/12/2025

M1 not in scope is weird 🤔 missing a language extension?

Basile Henry @basilehenry.bsky.social · 30/11/2025

I don't know how much AI they ended up using, but watching some of their dev talks a while ago they had some really impressive procedural generation using advanced ML techniques (i.e. AI), it really didn't strike me (at that time) like the plagiarism machine you hear about in more recent game dev 🤷‍♂️

Basile Henry @basilehenry.bsky.social · 24/11/2025

I mean yeah theoretically 😅 It's a practical approach, and I'm struggling to find ways to exploit it to create a function that would do radically different things for different types 🤔 There might be some tricks where you detect the type using broken `==` for NaN, but that's also the case in Haskell

Basile Henry @basilehenry.bsky.social · 24/11/2025

I think it might depend if you have "magic" (no constraints) type classes à la Elm for things like `==`, `+`, or `>=` 🤔

Basile Henry @basilehenry.bsky.social · 18/11/2025

That's probably the most reliable method (especially since the source path could be generated from nix)! I guess I'm not trusting my AI enough to do the run build step for me yet 😅