Wollantine

Channel about Elm, JS, and React best practices.

React developer. Functional Programming advocate.

[Wollantine @wollantine]

I'm going to follow in the footsteps of @Aron_Adler and uninstall Twitter from my phone.
I hope this helps me focus more on the important stuff, both in here and out.

[undefined behavior @undebeha]

on my language wishlist is a language that makes it super easy to describe type level checklists. i think this sort of exists already but im not sure how clean it is.

like im thinking about opengl rn. shader programs will often have some global values, like positions for lights or cameras. each time you switch shaders you gotta configure whatever uniform data the program needs before you start using it to render objects. you can modify this data, but it is usually an error to not initialize it.

the naive fix is to wrap up all of opengl in something functional. each invocation of a shader program could have these values as parameters. which would totally work and make everything type safe but now you are sending data to the GPU for every single draw call. if you only need to configure the shader once and are gonna reuse the config on 1000 different objects then you are sending a lot of redundant info on the wire.

you could alternatively try to do some sort of reactive programming system, where you have the trad functional API but under the hood the program checks if there has been a state transition and skips the data upload if there isn't a change to the shader config. but this is pretty hard to get right. and i suspect it is really hard to get the compiler to eliminate all the checks for changed data at comptime, so there is still some runtime overhead

what seems like a nice lightweight alternative would be some kind of type level checklist. in order to run a shader program you need to provide evidence of initialization of each of the relevant uniform data fields. the signature for the shader execution would be something like `runShader : ShaderProgramID -> ProofThatLightingIsConfigured -> ProofThatCoordinateTransformIsConfigured -> IO ()`. in order to obtain evidence you'd have to run other programs, like `configureLighting : LightData -> IO ProofThatLightingIsConfigured`.

in principle all this evidence should be erasable at compile time. so no overhead. but it also guarantees that the shader is configured before it is used.

also opens the door to things like skipping the data upload. it is technically legit to skip configuration; opengl initializes unconfigured uniform buffers to predictable default values. but imo it's still good practice to be explicit. so you could add a `useDefaultLightingConfiguration : ProofThatLightingIsConfigured` and now it is clearly documented and checked by the type system that you wanna use the defaults.

i know there are things out there like this. idk ATS at all but i get the vibe that this is the sort of thing ATS is all about.

also seems probably pretty easy to implement a crude version of this with linear type systems + phantom types

otoh i don't see a lot of implementations of this sort of system and so that makes me wonder if there is a reason for that. given it doesn't seem terribly sophisticated i'd expect to see it show up places if it were actually useful

maybe the devex isn't that great? it kind of seems like modeling a state machine at the type level with a million phantom types could feel very cluttered. maybe with the right syntax support this would be useful?

or maybe its the kind of thing that you can get stuck admiring and improving your fancy type level model instead of getting work done. maybe its the sort of thing where the amount of time spent building the evidence management system eclipses the time spent debugging and so economy suggests not doing it

so idk maybe this is a dumb idea. but i really love checklists and so i imagine id love a good system for representing them at the type level

[Wollantine @wollantine]

@Aron_Adler @kerckhove_ts Make the illegal state unrepresentable democratically as well.

[Wollantine @wollantine]

@kerckhove_ts This might come as a surprise for C and bash developers, but 0 is not a boolean, it's an integer.

[Wollantine @wollantine]

@DjMolehill @nicbarkeragain But it's not "the actual concepts", it's "the C concepts", and it's good that people stop learning the C concepts.

[Wollantine @wollantine]

@itsfarseen A hidden storage. Like a concealed box of weapons, or a hole where a squirrel would hide the nuts it found so it wouldn't need to go searching when it gets hungry.

[Wollantine @wollantine]

@kai_fall Wow, Rust is miles ahead.

[Wollantine @wollantine]

@itsfarseen Computer science does not have the monopoly on the meaning of words either. Have you lost your mind? You are implying "cache" now means *exclusively* a memory optimisation, so fuck people who study rodents I guess.

[Wollantine @wollantine]

@kmdrfx Lol "manure" instead of "demure". Spot on!

[Wollantine @wollantine]

@iamtexture @zack_overflow Yes. We circle back to my first post. What if their personal beliefs include ceding the language to other maintainers once it gains traction?

[Wollantine @wollantine]

@iamtexture @zack_overflow I think the sample size is way too small and we're doomed to shoot our subjective impressions.
I propose we wait until a thousand more autists have made popular PLs.

[Wollantine @wollantine]

@iamtexture @zack_overflow Nothing stops that autistic man's manifest ideology from being "a language should cater to its audience".

[Wollantine @wollantine]

@mattpocockuk @dillon_mulroy @FlavioCorpa If you want to make FP palatable, you need to make FP palatable, not an approximation of it. FP is great because of functions, not despite them.

[Wollantine @wollantine]

@mattpocockuk @dillon_mulroy @FlavioCorpa Not having .ap() means you can't apply an Effect of a function to an Effect of a value. In that case, Effect would be designed as a wrapper of values exclusively, and not treating functions as a first class citizen. Thus, it wouldn't be FP.

[Wollantine @wollantine]

@mattpocockuk @dillon_mulroy @FlavioCorpa It doesn't matter if the course of action must be the same for both: Effect must be FP.

[Wollantine @wollantine]

@davesnx To hell with abbreviations, I say!
And don't get me started with single letter variables!

[IroncladDev @IroncladDev]

Anyone who prefers M*crosoft Teams over google meet doesn't deserve employment

[Wollantine @wollantine]

@awesomekling @ladybirdbrowser

[Wollantine @wollantine]

@Aron_Adler Well, they can't offer help with Lean, or you would be sure they are phishing, can they?

[Wollantine @wollantine]

Addendum: Slop is Senior Level Objective Programming.

[Wollantine @wollantine]

To make goop you just need Go and Objective Programming.