1 unstable release
0.1.0 | Mar 27, 2023 |
---|
#945 in Rust patterns
46KB
788 lines
Long story short, this library solves the function colouring problem and some other stuff that the Keyword Generics Initiative is looking at. It's an experiment in how an effect system could fit into Rust.
effing-mad, an effect library for Rust
This library brings algebraic effects and effect handlers to Rust, by providing traits and macros
that allow writing effectful code in more or less the same style as Rust's existing async
functions.
what does that mean
Effectful functions that use this library can explicitly suspend (yield
) their execution and pass
control back to their callers, who later resume the function. Data can be carried through both of
these transitions, and this data is strongly typed in the same way as one would expect in Rust.
The data passed out of an effectful function specifies an action. It is given to an effect handler, which performs the action and passes the result back into the effectful function. Then, execution of the effectful function continues. It's like calling functions, but upside down.
As it turns out, calling functions upside down has some fantastic advantages. Since effects are
handled inside the caller instead of the callee, there are no bounds on whether they are handled
asynchronously, optionally, fallibly, or a bunch of other adverbs. On top of that, different
callers can use different handlers on the same effectful function, meaning an effectful function
that performs I/O can be called from either a regular fn
or an async fn
, and do the I/O in the
"natural" way in both. No more distinction between popular_crate
and popular_crate::blocking
!*
* effing-mad
not guaranteed to become popular. Use at own risk.
This whole mess can be seen in action in the examples/
directory. Check out the
"basic" example first, unless you're really smart or brave or something.
why
TL;DR: API experimentation and fun
I saw a post about the
recent efforts to make functions be usable from both async and sync contexts. The authors also
wrote about the desire for higher-order functions such as Option::map
to be able to be
asynchronous, optional, fallible, or a bunch of other adjectives - without having to write a
specialised function like try_async_map
for each set of effects. This is the problem I was talking
about last paragraph, where I also explain how an effect system could solve it.
Knowing this, I was surprised to see that in the FAQ of the post they answered "are you building an effect system?" with "not really". They never explained why not! So I did it instead, because I wanted to see what would happen.
how cool is it?
effing-mad is cool. Rad, even. Check out all these cool things it has:
- unstable compiler features (generators, generator_trait)
- a function with 22 type arguments
#![no_std]
- scary category theory words (coproduct! 👻)
- procedural macros
declarative macros(used to have them, but I needed them to generate idents which they can't do :( )unsafe
- raw pointers
Pin
, which means I'm really good at Rust- a dependency called "frunk". hehehehe
a market analysis conducted in 2022 determined that no other library has as many cool things as effing-mad, except maybe core. I mean, core invented Pin, which is cheating, so really I win.
some other facts
In pure functional languages like Haskell, side effects such as I/O are traditionally encoded in the language and its type system using monads. These are types which hold the effects and/or the result of the pure computation. So, if you want to have multiple types of effect, things get tricky because all your effect types want to hold the pure thing inside themselves, but you only have one pure thing going on. This means you have to use monad transformers. I don't really understand monad transformers, therefore they are bad.
On the other hand, I do understand algebraic effects, and therefore they are good. Also they compose far more easily and in my opinion are more intuitive. Download effing-mad today!
more facts - to do with rust this time
effing-mad uses generators. The implementation of it was made far easier by the lang team already needing functions that can be suspended and resumed, because that caused them to invent generators. Even though they were made for the compiler to be able to compile async fns, they're way more general than async fns are.
Generators are used by the compiler, but directly using them has not been stabilised yet because no one really needs it to be. That's why you need to use a nightly compiler to use effing-mad.
Dependencies
~2MB
~41K SLoC