ice-9 match penalty depending on pattern?

[Simon Tournier <zimon.toutoune@gmail.com>]

Hi,

From Ludo’s mastodon message [1]:

    Re ‘match’ penalty: when using ellipses in patterns, the generated
    code checks for “proper lists”, which is O(n).  The trick is to
    instead match a pair:

    ✔ (match lst ((head . tail) …))
    ❎ (match lst ((head tail ...) …))

Therefore I have confused by some patterns.

--8<---------------cut here---------------start------------->8---
28 candidates:
./gnu/services/monitoring.scm:249:    ((head tail ...)
./gnu/system/file-systems.scm:249:          ((head1 tail1 ...)
./gnu/system/file-systems.scm:251:             ((head2 tail2 ...)
./gnu/packages.scm:116:                              ((_ file head tail ...)
./gnu/build/activation.scm:85:      ((head tail ...)
./gnu/build/linux-boot.scm:225:      ((head tail ...)
./guix/http-client.scm:255:          ((head tail ...)
./guix/records.scm:604:      ((_ record field offset ((head normal) tail ...))
./guix/records.scm:607:      ((_ record field offset ((head delayed) tail ...))
./guix/records.scm:610:      ((_ record field offset ((head thunked) tail ...))
./guix/read-print.scm:482:      ((head tail ...)
./guix/read-print.scm:750:      ((head tail ...)
./guix/lint.scm:1614:        ((head tail ...)
./guix/scripts/package.scm:809:                     ((head tail ...) head))))
./guix/self.scm:160:      ((head tail ...)
./guix/store.scm:1556:        ((head tail ...)
./guix/store.scm:1764:      ((head tail ...)
./guix/ui.scm:295:      ((head tail ...)
./guix/ui.scm:2258:               ((head tail ...) head)))
./guix/docker.scm:261:      (((head ...) (tail ...) id)
./guix/build/store-copy.scm:72:        ((head tail ...)
./guix/build/gremlin.scm:336:      ((head tail ...)
./guix/build/graft.scm:330:      ((head tail ...)
./guix/build/utils.scm:236:      ((head tail ...)
./guix/build/utils.scm:405:      ((head tail ...)
./guix/utils.scm:910:          ((head1 tail1 ...)
./guix/utils.scm:913:             ((head2 tail2 ...)
./build-aux/compile-all.scm:71:           ((head tail ...)
--8<---------------cut here---------------end--------------->8---

Maybe for some of them, it changes nothing for the user-visible
performances.  Maybe it does. :-)

Well, I do not know the length of each match.  However, some are part of
some loop.  Therefore, if we could save some cycles by simply replacing
the ellipsis, as

    ((head tail ...) stuff that use head or tail)

by

    ((head . tail) stuff that use head or tail)

Why not?  Do I miss something in the implementation of ’match’?

Cheers,
simon

1:
https://social.sciences.re/@civodul@toot.aquilenet.fr/111885442823194970