--- Begin Message ---
Subject: |
[feature request] optimization of case-lambda |
Date: |
Mon, 20 Feb 2023 18:56:25 +0100 |
On 3.0.9
> ,optimize ((case-lambda (() 0)))
= 0
but
> ,optimize ((case-lambda (() 0) ((a) 1)))
= ((case-lambda (() 0) ((a) 1)))
The problem with this is that when the output of a macro contains case-lambda,
recursive application results in geometrical increase of code size. It seems
that it should be possible to resolve the application on the spot; at last to
reduce the case-lambda to a simple lambda when the arity is available (which
now only happens when the case-lambda contains a single clause).
Thanks
Daniel
--- End Message ---
--- Begin Message ---
Subject: |
Re: bug#61660: [feature request] optimization of case-lambda |
Date: |
Mon, 27 Feb 2023 18:19:42 +0100 |
> On 27 Feb 2023, at 11:11, Ludovic Courtès <ludo@gnu.org> wrote:
>
> Hi Daniel,
>
> lloda <lloda@sarc.name> skribis:
>
>> From 61ed612fb36108e395bdee4b1bbb46b49ef017b3 Mon Sep 17 00:00:00 2001
>> From: Daniel Llorens <lloda@sarc.name>
>> Date: Thu, 23 Feb 2023 17:38:10 +0100
>> Subject: [PATCH] peval reduces some inlined case-lambda calls
>>
>> * module/language/tree-il/peval.scm (peval): Reduce multiple case lambda
>> in <call> trees according to the number of arguments. Do not try to
>> reduce case-lambda using keyword arguments.
>> * test-suite/tests/peval.test: Tests.
>
> [...]
>
>> +++ b/module/language/tree-il/peval.scm
>> @@ -1668,6 +1668,29 @@ top-level bindings from ENV and return the resulting
>> expression."
>>
>> (log 'inline-end result exp)
>> result)))))
>> + (($ <lambda> src-proc meta orig-body)
>> + ;; If there are multiple cases and one matches nargs, omit all
>> the others.
>> + (or (and
>> + (lambda-case-alternate orig-body)
>> + (let ((nargs (length orig-args)))
>> + (let loop ((body orig-body))
>> + (match body
>> + (#f #f) ;; No matching case; an error.
>> + (($ <lambda-case> src-case req opt rest kw inits
>> gensyms case-body alt)
>> + (cond (kw
>> + ;; FIXME: Not handling keyword cases.
>> + #f)
>
> Maybe s/FIXME/XXX/ since it’s at most a limitation, certainly not a bug.
>
> It LGTM and Andy already approved it on IRC, so go ahead!
>
> Ludo’.
Apologies for not seeing this earlier. Pushed to
3b47f87618047ebb8812788c64a44877a4f2e0dd. Thanks!
- Daniel
--- End Message ---