Pattern matching with Kleene star on structure type in Racket

Question

I started playing with Racket pattern matching system recently and got into a problem i can't understand.

If i do:

(match (list 1 2 3 4 5 6 7 8 9 10 11 12)
    [(list _ x y z ...) (list y ': x)]) 

in REPL i get

'(3 : 2)

as my desired result.

If i do:

(match (current-date)
    [(date* _ x y z ...) (list y ': x)])

or

(match (date* 5 18 13 18 11 2011 5 321 #f 3600 0 "W. Europe Standard Time")
    [(date* _ x y z ...) (list y ': x)])

i get this error:

match: wrong number for fields for structure date*: expected 12 but got 5 in: (_ x y z ...)

i suspect that the Kleene star ... does not work with struct type for some reason. Why is it so?

Answer 1

It seems that what you really want to do is match a subset of the fields in a struct, rather than actually bind the rest of the fields in the struct to (z ...). In this case, you may want to try using the struct* match pattern instead.

Here's an example:

(match (date* 5 18 13 18 11 2011 5 321 #f 3600 0 "W. Europe Standard Time")
   [(struct* date ([minute x] [hour y])) (list y ': x)])

Answer 2

i suspect that the Kleene star ... does not work with struct type for some reason.

Correct. The ... notation, and the related ..k and ___ and __k notations, aren't really a general feature of matching, but rather a specific feature of list-matching and hash-table-matching and so on. If you examine the formal production in the documentation, you'll see that some types of subpatterns accept lvp (which is defined as either pat, or pat followed by ... or one of its friends) in various places, whereas others only accept pat.

Why is it so?

I guess it's simply that ... is used to translate part of a list or vector or hash-table or whatnot into its own list, and there's no sensible way to do that for a fixed-length struct. It may be worth pointing out that ... behaves differently for different types — for example, it can be used to extract a list of the keys in a hash-table — so the only reason to support it for structs would be if there were a specific analogous operation that would make sense for structs. I don't think there is one.