[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: why I love scheme
From: |
Zelphir Kaltstahl |
Subject: |
Re: why I love scheme |
Date: |
Wed, 15 Dec 2021 20:30:46 +0000 |
I did not know, that lambdas are allocated on the heap. I have a few questions
now:
How does this affect using fibers? (And how do fibers work better in that case?)
The unrolling you mentioned. Would same not be possible for the
naive-but-not-tail-recursive version? Is the idea, that the continuation tail
recursive version does work better, because the compiler is somehow able to
optimize it better? If so, why?
I am asking, because I once had the same kind of problem and then read, that
instead of growing stack levels, I am growing the continuation, so not winning
anything. But perhaps that was wrong and I should have gone for the continuation
solution. I would like to be able to make an educated decision when next meeting
such a problem.
Best regards,
Zelphir
On 12/15/21 1:59 PM, Stefan Israelsson Tampe wrote:
> I believe that the lambda closures will be allocated from the heap and hence
> this procedure will
> be perfect if you are using fibers.. Also the compiler can do magic if it
> want's and unroll
> and untangle a few iterations, so it can be very fast as well.My point is that
> the named let
> is such a nice looping construct (try to do this with a for loop in java). I
> use it all the time
> and only sometimes I need to move to even more advanced constructs like
> letrec.
>
> On Wed, Dec 15, 2021 at 10:38 AM Zelphir Kaltstahl <zelphirkaltstahl@posteo.de
> <mailto:zelphirkaltstahl@posteo.de>> wrote:
>
> Hello Stefan!
>
> This translates a recursive tree function into a tail recursive function.
> However, in this case, I am not sure it is really worth doing, in
> comparison to
> the naive (+ first-branch other-branch) solution. The reason is, that
> instead of
> a call stack for multiple branches, you are only moving that stuff into a
> longer
> and longer continuation, which will be on the stack in each recursive
> call.
>
> However, I think you or other people on the list probably know more about
> this
> than I do and about how the approaches compare in terms of memory and
> time.
> Maybe the stack frames are more in size than the memory consumed by the
> overhead
> of the continuation, or the other way around.
>
> Regards,
> Zelphir
>
> On 12/15/21 12:44 AM, Stefan Israelsson Tampe wrote:
> > Maybe you think the below program is trivial, but I adore named let's so
> > much that I just cannot fathom that when people go functional they
> totally
> > miss this beauty
> >
> >
> > (define (count tree)
> >
> > ;; s = total sum up to now
> >
> > ;; t = tree of the type (car = child . cdr = siblings)
> >
> > ;; cont is the continuation, (cont 10) will continue
> >
> > ;; the calculation with the sum=10 see how we initiate
> >
> > ;; with a continuation that evaluates returns it's argument
> >
> >
> > (let loop ((s 0) (t tree) (cont (lambda (s) s)))
> >
> > (if (pair? t)
> >
> > (loop s (car t) (lambda (s) (loop s (cdr t) cont)))
> >
> > (cont (if (number? t) t 0))))
>
> --
> repositories: https://notabug.org/ZelphirKaltstahl
> <https://notabug.org/ZelphirKaltstahl>
>
--
repositories: https://notabug.org/ZelphirKaltstahl