Strict pattern matching

[samhh]

What is the problem this feature would solve?

Walls of unneeded lambdas when pattern matching harm readability. For example:

import { Data } from "effect"

type Example = Data.TaggedEnum<{
  A: {},
  B: {},
  C: {},
  // etc
}>

const { $match } = Data.taggedEnum<Example>()

$match({
  A: () => "it is A",
  B: () => "this time it is B",
  C: () => "this is wild, it's C!",
  // etc
})

What is the feature you are proposing to solve the problem?

matchX simply makes each case strict. This has different performance characteristics, however in many cases it's viable and improves readability:

$matchX({
  A: "it is A",
  B: "this time it is B",
  C: "this is wild, it's C!",
  // etc
})

This is particularly relevant when there are potentially dozens of sum members.

What alternatives have you considered?

constant, tolerating the lambdas.

[mikearnaldi]

Given that JS has eager evaluation you generally don't want to compute the alternatives unless they match, this isn't something we would like to add for the time being

[samhh]

Given that JS has eager evaluation you generally don't want to compute the alternatives unless they match, this isn't something we would like to add for the time being

I'm really thinking primarily of the example where rather than expensive computations it's just strings or other primitives. We find in our codebase that this comes up quite a lot in pattern matching, and I perhaps naively wonder if this might be more performant than a redundant anonymous function call.

[mikearnaldi]

Given that JS has eager evaluation you generally don't want to compute the alternatives unless they match, this isn't something we would like to add for the time being

I'm really thinking primarily of the example where rather than expensive computations it's just strings or other primitives. We find in our codebase that this comes up quite a lot in pattern matching, and I perhaps naively wonder if this might be more performant than a redundant anonymous function call.

If that kind of perf is important to you probably you shouldn't even use a matcher in the first place, preventive optimization is usually bad

[samhh]

If that kind of perf is important to you probably you shouldn't even use a matcher in the first place, preventive optimization is usually bad

I agree, but wasn't performance your primary concern?

Maybe there's unease about further complicating the pattern matching API, which'd make sense. Just thought I'd raise this though as, in considering a migration path from the fp-ts ecosystem, we've found ourselves quite liking matchX in sum-types.

[mikearnaldi]

If that kind of perf is important to you probably you shouldn't even use a matcher in the first place, preventive optimization is usually bad

I agree, but wasn't performance your primary concern?

Maybe there's unease about further complicating the pattern matching API, which'd make sense. Just thought I'd raise this though as, in considering a migration path from the fp-ts ecosystem, we've found ourselves quite liking matchX in sum-types.

No, my primary concern was semantics, even in the Effect type we had to make matchers lazy because users expected branches not to be computed

[samhh]

No, my primary concern was semantics, even in the Effect type we had to make matchers lazy because users expected branches not to be computed

Gotcha, that makes holistic sense. Thanks for engaging. 🙂

[mikearnaldi]

Wondering if we could use the same api with an optional parameter though, as long as we don't duplicate functions and we can support it across the types I am not against

[aminnairi]

You could also create an identity function that takes an argument, and return a function that return that argument.

const always = <Value>(value: Value) => () => value;

match({
  A: always("it is A"),
  B: always("this time it is B"),
  C: always("this is wild, it's C!"),
});

Similar to what Lodash or Ramda do.

[FedericoBiccheddu]

You could also create an identity function that takes an argument, and return a function that return that argument. [...]

FYI there is const already.