Add syntax sugar that translates ? to Result.try

[smores56]

Issue following from discussion in Zulip chat.

We want to add Rust's try operator, the ?, to Roc, making early returns on Err cleaner. This will work just like ! works when it desugars to Task.await.

There are many benefits to having ? added to Roc, including "flatter" code, better readability, and better concision, but these could also be achieved if ! was extended to work for Results or even anything implementing a hypothetical AndThen ability. However, the mental overhead of understanding that a return type's AndThen implementation is being leveraged, as well as the teach-ability damage to the language, make generalizing the ! operator cost-prohibitive. By having ! just for Tasks and ? just for Results, we get the cleanliness benefits of each operator without making the language much harder to learn, and we prevent potential exploitation of a generalized ! operator.

This also paves the way for the eventual deprecation and removal of backpassing from the language.

Some things we should consider when implementing this operator:

• It will work almost exactly like the ! operator, but will instead translate to Result.try every time
• The error messages we generate on type errors will not only need to handle when the value ? is used on isn't a tag union, but also if there are extra or missing variants from [Ok _, Err _]

[sekerez]

Would love to have a go at this! Looking into it, will provide an update

[lukewilliamboswell]

I can give you some good pointers on this.

First step is to implement a new variant in crates/compiler/parse/src/ast.rs, probably something like ResultTryQuestion or something, that is almost identical to TaskAwaitBang(&'a Expr<'a>).

You should add some snapshot tests in crates/compiler/test_syntax/tests/snapshots/pass and confirm it parses correctly.

That is probably one PR.

Then the next step is to desugar it, which I've nicknamed "unwrap" because the best place I think to do this is directly after all the Expr's get desugared but before the rest of canonicalisation. Specifically, in crates/compiler/can/src/suffixed.rs, you will find the unwrap_suffixed_expression which is responsible for desugaring the ! operator. So we can modify this to also handle the ? suffix. You can find the tests for this in crates/compiler/can/tests/test_suffixed.rs

If you need any assistance with parsing @joshuawarner32 is definitely the expert. For more general things @kdziamura and myself have been working in this part of the compiler recently. Hope this helps. 😍

[sekerez]

Thanks @lukewilliamboswell!