On 12.09.2021 19:05, Damien Mattei wrote:
> Hello Zelphir,
>
> condx evaluate all code(s) in the 'exec' block until a conditional is true,
> it then evaluate the consequent code of course.
> So ,yes your true it saves a lot of nesting parenthesis as in this example:
>
Interesting macro. I think I've occasionally felt the need for such a thing.
Two comments:
1. Using 'begin' instead of 'exec' might feel more familiar to Schemers, and
since the 'literals' of syntax-rules are matched hygienically starting
from Guile 2.2 (I think), it might be a good idea to use a common core
identifier instead of an unbound one like 'exec'. (Or you could bind it.)
A little elaboration, in case you don't know what I'm talking about: the
"literals" of a syntax-rules macro are not matched by "name" (the string
representing the symbol seen in code) but by "variable binding."
For example:
(let ((else #f))
(cond
((= 1 0) 'foo)
(else 'bar)
(#t 'qux)))
; => qux
The code returns qux, because the 'cond' macro doesn't recognize the 'else'
as the 'else' that it knows of, since it's been rebound via let.
An example that would work in a conforming R7RS-small implementation; don't
know if it works with Guile, it's just to explain the principle:
;; Import core bindings, but renaming 'else' to 'otherwise'
(import (rename (scheme base) (else otherwise)))
(cond
((= 1 0) 'foo)
((= 2 3) 'bar)
(otherwise 'qux))
; => qux
The (scheme base) library defined by R7RS-small exports the identifier 'else'
which is used by 'cond' for matching. This allows the programmer to rename
the 'else' used by 'cond' while importing the base library. (The 'else' is
not bound to anything useful, it's just bound at all so it can be renamed.)
Likewise you might want to bind 'exec' to anything and export it along with
the 'condx' identifier, so if some Schemer uses the identifier 'exec' for
something different in their code, they can still use your macro by renaming
your 'exec' to something else. Otherwise there's no way to make it work.
If 'exec' was unbound during the definition of 'condx' then it must remain
unbound for 'condx' to recognized it again, meaning it can't be renamed.
Or (IMO better) you could reuse the 'begin' binding (in Guile's case, from
boot-9, in R7RS-small, from (scheme base)), because it's very unlikely that
someone will use 'begin' for something else in their code, and it would force
them to rename the core 'begin' to something else and then that would work
with your code automatically. E.g. if someone renames 'begin' to 'start' it
will automatically work in your macro if you had defined it with 'begin' in
the literals list of syntax-rules.
2. You might be interested in let/ec, which lets you bind a variable to an
"escape continuation" i.e. a way to "return" from a block of code. Here's
your code using condx rewritten to use let/ec instead:
(define (ssigma-proto-condx L t)
(set! cpt {cpt + 1})
(define ls (length L))
(define dyn (array-ref dyna ls t))
;; dyna[ls][t] means 0: unknown solution, 1: solution found, 2: no solution
(let/ec return
(when (not (zero? dyn))
(return (one? dyn)))
(when (null? L)
(array-set! dyna 2 ls t)
(return #f))
(define c (first L))
(when {c = t}
(array-set! dyna 1 ls t)
(return #t))
(define R (rest L))
;; continue searching a solution in the rest
(when {c > t}
(define s (ssigma-proto R t))
(array-set! dyna (one-two s) ls t)
(return s))
;; else :
;; c < t at this point
;; c is part of the solution or his approximation
;; or c is not part of solution
(define s {(ssigma-proto R {t - c}) or (ssigma-proto R t)})
(array-set! dyna (one-two s) ls t)
(return s)))
I've turned all 'condx' branches that weren't exec to use 'when' because it
felt more natural with the imperative style. Of course you could still use
the regular 'cond' here and there or the regular 'if'. 'When' is just short
for an 'if' without an else part.
The last '(return s)' could just be 's' but I find it more consistent and
readable to use 'return' there as well.
--
Taylan
