Re: ice-9 match penalty depending on pattern?

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

Hi,

On mer., 07 févr. 2024 at 10:41, Carlo Zancanaro <carlo@zancanaro.id.au> wrote:

>> Why not?  Do I miss something in the implementation of ’match’?
>
> The only reason I can think of would be if these matches are sometimes
> provided improper lists, which need to fail these match conditions. That
> seems unlikely to me, but it should be clear from looking at the other
> match clauses in each case.

Well, I have not pruned the list returned by just grepping. :-)  And I
have just grepped with the term ’head’, ’tail’ and ’\.\.\.’

Somehow, my question is twofold:

1. Is the “expensive” check worth for such case:

      (match paths
        ((head tail ...)
         (if (visited? head)
             (loop tail visited result)
             (call-with-values
                 (lambda ()
                   (loop (references store head)
                         (visit head)
                         result))
               (lambda (visited result)
                 (loop tail
                       visited
                       (cons head result))))))
        (()
         (values visited result)))))

seen in ’topologically-sorted’ procedure from (guix store) module.

2. Is the “expensive” check worth for such multi-cases:

      (match sexp
        ((? string? str)
         (let ((prefix "swh:1:dir:"))
           (if (string-prefix? prefix str)
               (cons (string-drop str (string-length prefix)) ids)
               ids)))
        ((head tail ...)
         (loop tail (loop head ids)))
        (_ ids))
        
seen in ’lookup-disarchive-spec’ from (guix lint).

Well, I am not saying to rely on ’car’ and ’cdr’.  Instead, I am asking
what is the idiomatic Guile pattern matching for Guile?

My main concern is about chasing the unnecessary checks for making Guix
a bit faster. :-)


Cheers,
simon