Re: master 68c09c6: Better CPS conversion of multi-binding `let`

Re: master 68c09c6: Better CPS conversion of multi-binding `let`

[Stefan Monnier]

> -    ;; Process `let' in a helper function that transforms it into a
> -    ;; let* with temporaries.
> +    (`(,(or 'let 'let*) () . ,body)
> +      (cps--transform-1 `(progn ,@body) next-state))
> +
> +    (`(let (,binding) . ,body)
> +      (cps--transform-1 `(let* (,binding) ,@body) next-state))
> +
> +    ;; Transform multi-variable `let' into `let*':
> +    ;;    (let ((v1 e1) ... (vN eN)) BODY)
> +    ;; -> (let* ((t1 e1) ... (tN eN) (v1 t1) (vN tN)) BODY)

I think this optimization can be generalized to the multi-var case by
noticing that the transformation of (vN eN) into a pair (tN eN) ... (vN tN)
is not necessary because the order of the (v1 t1) ... (vN tN) doesn't matter
so we can move (vN tN) to the beginning, right next to (tN eN) and then
merge them back.


        Stefan

Re: master 68c09c6: Better CPS conversion of multi-binding `let`

[Mattias Engdegård]

30 nov. 2021 kl. 14.19 skrev Stefan Monnier <monnier@iro.umontreal.ca>:

> I think this optimization can be generalized to the multi-var case by
> noticing that the transformation of (vN eN) into a pair (tN eN) ... (vN tN)
> is not necessary because the order of the (v1 t1) ... (vN tN) doesn't matter
> so we can move (vN tN) to the beginning, right next to (tN eN) and then
> merge them back.

Not sure how that would work unless you imply some kind of dependency analysis. Maybe you have an example in mind?

(Sorry about the incorrect git summary line, by the way -- 'multi-binding' should have been 'single-binding'.)

Re: master 68c09c6: Better CPS conversion of multi-binding `let`

[Stefan Monnier]

>> I think this optimization can be generalized to the multi-var case by
>> noticing that the transformation of (vN eN) into a pair (tN eN) ... (vN tN)
>> is not necessary because the order of the (v1 t1) ... (vN tN) doesn't matter
>> so we can move (vN tN) to the beginning, right next to (tN eN) and then
>> merge them back.
>
> Not sure how that would work unless you imply some kind of dependency
> analysis. Maybe you have an example in mind?

No dependency analysis needed, just always turn

    (let ((x1 e1) ... (xN eN))
      BODY)

into

    (let* ((t1 e1) ... (tN-1 eN-1) (xN eN) (x1 t1) .. (xN-1 tN-1))
      BODY)


-- Stefan

Re: master 68c09c6: Better CPS conversion of multi-binding `let`

[Mattias Engdegård]

30 nov. 2021 kl. 14.42 skrev Stefan Monnier <monnier@iro.umontreal.ca>:

>    (let ((x1 e1) ... (xN eN))
>      BODY)
> 
> into
> 
>    (let* ((t1 e1) ... (tN-1 eN-1) (xN eN) (x1 t1) .. (xN-1 tN-1))
>      BODY)

Sorry, I should have read your original comment more carefully; I thought you meant some way of eliminating more than one temporary. Yes, that's probably a good idea. I'll see what I can do.

(That function, cps--transform-1, suffers severely from our lack of TCO by the way.)

Re: master 68c09c6: Better CPS conversion of multi-binding `let`

[Stefan Monnier]

> (That function, cps--transform-1, suffers severely from our lack of TCO by the way.)

But we can wrap it within `named-let`, right?


        Stefan

Re: master 68c09c6: Better CPS conversion of multi-binding `let`

[Mattias Engdegård]

30 nov. 2021 kl. 14.54 skrev Stefan Monnier <monnier@iro.umontreal.ca>:

> But we can wrap it within `named-let`, right?

The thought did cross my mind, but I had to draw the line for my involvement in that code somewhere.