match-lambda from (chibi match) (? ...) defines guard for every match

[SirDroun]

when using a guard in a match-lambda expression i expect the guard to be defined once, when i define the pattern, and not every time the procedure the match-lambda defines is called. The guard should be part of the closure created by the match-lambda. A simple example:

(map (match-lambda* (((? (begin (display 'a)(newline)(lambda (x) #t)) x) x))(y y)) '(1 2 3 4))
a
a
a
a
((1) (2) (3) (4))

[ashinn]

Thanks for the report!

The nature of the match compiler is that it expands as it goes in a single pass. It would require a large rewrite or a whole new pass just for this one optimization.

I'll think about it but I'm inclined not to fix this, and to just add to the documentation that the guard form will be evaluated every time so is generally expected to be a lambda or reference.

[SirDroun]

I understand, unfortunately it's not just about efficiency, it's also about side-effects. I'll se if i can have a go on it when i get some spare time.

[ashinn]

Do you have a real-world use case where you want a side-effecting guard?

[SirDroun]

Had one where guard used a filesystem cache, but refactoring made it go away. When you start messing with files it can get interesting.

[ashinn]

The guard is evaluated inside the enclosing match scope, so I don't think it's actually possible to define it just once.