Re: master d62766305a: Add `take` and `ntake` (bug#56521)

Re: master d62766305a: Add `take` and `ntake` (bug#56521)

[Robert Pluim]

>>>>> On Sun, 17 Jul 2022 11:47:25 -0400 (EDT), Mattias Engdegård <mattiase@acm.org> said:
 
    Mattias> +@defun ntake n list
    Mattias> +This is a version of @code{take} that works by destructively modifying
    Mattias> +the list structure of the argument.  That makes it faster, but the
    Mattias> +original value of @var{list} is lost.
    Mattias> +@end defun
    Mattias> +
    Mattias>  @defun last list &optional n
    Mattias>  This function returns the last link of @var{list}.  The @code{car} of
    Mattias>  this link is the list's last element.  If @var{list} is
    Mattias> null,

Does this need the usual blurb to do

(setq list-var (ntake 3 list-var))

to ensure `list-var' is updated?

    Mattias> +DEFUN ("take", Ftake, Stake, 2, 2, 0,
    Mattias> +       doc: /* Return the first N elements of LIST.
    Mattias> +If N is zero or negative, return nil.
    Mattias> +If LIST is no more than N elements long, return it (or a copy).  */)
    Mattias> +  (Lisp_Object n, Lisp_Object list)

I think this would be clearer as

"If N is >= the length of LIST, return LIST (or a copy)."

since weʼre talking about the effect of various values of N.

    Mattias> +
    Mattias> +DEFUN ("ntake", Fntake, Sntake, 2, 2, 0,
    Mattias> +       doc: /* Modify LIST to keep only the first N elements.
    Mattias> +If N is zero or negative, return nil.
    Mattias> +If LIST is no more than N elements long, return it.  */)

And similarly here.

Robert
-- 

Re: master d62766305a: Add `take` and `ntake` (bug#56521)

[Mattias Engdegård]

18 juli 2022 kl. 10.48 skrev Robert Pluim <rpluim@gmail.com>:

> Does this need the usual blurb to do
> 
> (setq list-var (ntake 3 list-var))
> 
> to ensure `list-var' is updated?

Sort of; I added some text to that effect. Using the return value is only needed if N can be nonpositive.

> I think this would be clearer as
> 
> "If N is >= the length of LIST, return LIST (or a copy)."
> 
> since weʼre talking about the effect of various values of N.

Right, it's now been changed to something like that. Thank you!

Now `seq-take` uses `take`, and so does `seq-subseq`. This made both faster, but the latter was boosted immensely by not formatting an error string up-front.

`seq-subseq` would be even faster if simply written as

  (take (- end start) (nthcdr start sequence))

but the interface requirements force a much more cumbersome implementation. (And does it really need `copy-sequence`? The docs are unclear.)

I suppose that `seq-subseq` should use `cl-defmethod` for dispatching on type but I didn't change that part.

Re: master d62766305a: Add `take` and `ntake` (bug#56521)

[Robert Pluim]

>>>>> On Mon, 18 Jul 2022 12:59:51 +0200, Mattias Engdegård <mattiase@acm.org> said:

    Mattias> 18 juli 2022 kl. 10.48 skrev Robert Pluim <rpluim@gmail.com>:
    >> Does this need the usual blurb to do
    >> 
    >> (setq list-var (ntake 3 list-var))
    >> 
    >> to ensure `list-var' is updated?

    Mattias> Sort of; I added some text to that effect. Using the return value is only needed if N can be nonpositive.

    >> I think this would be clearer as
    >> 
    >> "If N is >= the length of LIST, return LIST (or a copy)."
    >> 
    >> since weʼre talking about the effect of various values of N.

    Mattias> Right, it's now been changed to something like that. Thank you!

That looks good, thanks

    Mattias> Now `seq-take` uses `take`, and so does `seq-subseq`. This made both
    Mattias> faster, but the latter was boosted immensely by not formatting an
    Mattias> error string up-front.

(meta-comment: youʼve made a number of speedups recently, it would be
great if you could show some numbers of the effects)
(meta-meta-comment: this is not a criticism of your work, just a nice-to-have)

    Mattias> `seq-subseq` would be even faster if simply written as

    Mattias>   (take (- end start) (nthcdr start sequence))

    Mattias> but the interface requirements force a much more cumbersome
    Mattias> implementation. (And does it really need `copy-sequence`? The docs are
    Mattias> unclear.)

(info "(elisp) Sequence Functions") says:

       All functions defined in this library are free of side-effects; i.e.,
    they do not modify any sequence (list, vector, or string) that you pass
    as an argument.

It doesnʼt say anything about the result not sharing structure with
the inputs. Probably code in the wild already implicitly depends on
the `copy-sequence', though, so Iʼd think twice before modifying that
contract.

Robert
-- 

Re: master d62766305a: Add `take` and `ntake` (bug#56521)

[Michael Heerdegen]

Robert Pluim <rpluim@gmail.com> writes:

> (info "(elisp) Sequence Functions") says:
>
>        All functions defined in this library are free of side-effects; i.e.,
>     they do not modify any sequence (list, vector, or string) that you pass
>     as an argument.
>
> It doesnʼt say anything about the result not sharing structure with
> the inputs.

The doc of `cl-subseq' which is (now) implemented based on `seq-subseq'
does, however (info "(cl) Sequence Functions"):

| [...] The return value is always a copy; it does not share structure
| with SEQUENCE.

Michael.

Re: master d62766305a: Add `take` and `ntake` (bug#56521)

[Sam Steingold]

> * Mattias Engdegård <znggvnfr@npz.bet> [2022-07-17 11:47:25 -0400]:
>
> branch: master
> commit d62766305ad8fe6ca1695341c34b9836d051e3cb
> Author: Mattias Engdegård <mattiase@acm.org>
> Commit: Mattias Engdegård <mattiase@acm.org>
>
>     Add `take` and `ntake` (bug#56521)
>     
>     These are useful list primitives, complementary to `nthcdr`.

How are these better than the venerable butlast and nbutlast?
https://www.gnu.org/software/emacs/manual/html_node/elisp/List-Elements.html#index-butlast
note that butlast and nbutlast have a very long Lisp history, they
appear also in Common Lisp and many lisps before that.


-- 
Sam Steingold (http://sds.podval.org/) on darwin Ns 10.3.2113
http://childpsy.net http://calmchildstories.com http://steingoldpsychology.com
https://thereligionofpeace.com https://ffii.org https://jihadwatch.org
Democrats, get out of my wallet! Republicans, get out of my bedroom!

Re: master d62766305a: Add `take` and `ntake` (bug#56521)

[Philip Kaludercic]

Sam Steingold <sds@gnu.org> writes:

>> * Mattias Engdegård <znggvnfr@npz.bet> [2022-07-17 11:47:25 -0400]:
>>
>> branch: master
>> commit d62766305ad8fe6ca1695341c34b9836d051e3cb
>> Author: Mattias Engdegård <mattiase@acm.org>
>> Commit: Mattias Engdegård <mattiase@acm.org>
>>
>>     Add `take` and `ntake` (bug#56521)
>>     
>>     These are useful list primitives, complementary to `nthcdr`.
>
> How are these better than the venerable butlast and nbutlast?
> https://www.gnu.org/software/emacs/manual/html_node/elisp/List-Elements.html#index-butlast
> note that butlast and nbutlast have a very long Lisp history, they
> appear also in Common Lisp and many lisps before that.

The immediate advantage I recognise is that you don't need to determine
the length of the list if you are actually just interested in the first
n, not just not-interested in the last m.

Re: master d62766305a: Add `take` and `ntake` (bug#56521)

[Sam Steingold]

> * Philip Kaludercic <cuvyvcx@cbfgrb.arg> [2022-07-18 18:56:24 +0000]:
>
> Sam Steingold <sds@gnu.org> writes:
>
>>> * Mattias Engdegård <znggvnfr@npz.bet> [2022-07-17 11:47:25 -0400]:
>>>
>>> branch: master
>>> commit d62766305ad8fe6ca1695341c34b9836d051e3cb
>>> Author: Mattias Engdegård <mattiase@acm.org>
>>> Commit: Mattias Engdegård <mattiase@acm.org>
>>>
>>>     Add `take` and `ntake` (bug#56521)
>>>     
>>>     These are useful list primitives, complementary to `nthcdr`.
>>
>> How are these better than the venerable butlast and nbutlast?
>> https://www.gnu.org/software/emacs/manual/html_node/elisp/List-Elements.html#index-butlast
>> note that butlast and nbutlast have a very long Lisp history, they
>> appear also in Common Lisp and many lisps before that.
>
> The immediate advantage I recognise is that you don't need to determine
> the length of the list if you are actually just interested in the first
> n, not just not-interested in the last m.

good point, thank you!

in that case you might consider modifying (n)butlast to
call (n)take to make them faster.



-- 
Sam Steingold (http://sds.podval.org/) on darwin Ns 10.3.2113
http://childpsy.net http://calmchildstories.com http://steingoldpsychology.com
https://www.memritv.org https://fairforall.org https://ffii.org
Incorrect time synchronization.

Re: master d62766305a: Add `take` and `ntake` (bug#56521)

[Mattias Engdegård]

19 juli 2022 kl. 00.57 skrev Sam Steingold <sds@gnu.org>:

> you might consider modifying (n)butlast to
> call (n)take to make them faster.

Thank you, that was indeed the plan and has now been done (for `butlast`; it was difficult to see any benefit for `nbutlast`).