Implementation of reduce-right in SRFI 1 TAKIZAWA Yozo (28 Dec 2021 08:50 UTC)
Re: Implementation of reduce-right in SRFI 1 Arthur A. Gleckler (08 Jan 2022 16:47 UTC)
Re: Implementation of reduce-right in SRFI 1 Alex Shinn (09 Jan 2022 23:20 UTC)
Re: Implementation of reduce-right in SRFI 1 John Cowan (10 Jan 2022 04:13 UTC)
Re: Implementation of reduce-right in SRFI 1 Alex Shinn (10 Jan 2022 04:43 UTC)
Re: Implementation of reduce-right in SRFI 1 Marc Nieper-Wißkirchen (10 Jan 2022 12:02 UTC)
Re: Implementation of reduce-right in SRFI 1 Arthur A. Gleckler (10 Jan 2022 16:31 UTC)
Re: Implementation of reduce-right in SRFI 1 Arthur A. Gleckler (23 Oct 2022 01:45 UTC)
Re: Implementation of reduce-right in SRFI 1 TAKIZAWA Yozo (10 Jan 2022 11:54 UTC)

Re: Implementation of reduce-right in SRFI 1 Alex Shinn 10 Jan 2022 04:42 UTC

Although I agree that would be preferable, it's not the
implementations that matter but the uses.  If there are any existing
programs passing non-associative procedures to reduce(-right), then
your proposal would break them.

--
Alex

On Mon, Jan 10, 2022 at 1:13 PM John Cowan <xxxxxx@ccil.org> wrote:
>
> +1.
>
> I note that SRFI 1, unlike some versions of reduce(-right), doesn't require _f_ to be associative.  If a non-associative _f_ were made an error, it would allow reduction, unlike folding, to be done in parallel.  I don't know if it's too late to make this change or not.  It wouldn't affect any existing implementations.
>
> On Sun, Jan 9, 2022 at 6:21 PM Alex Shinn <xxxxxx@gmail.com> wrote:
>>
>> This was from an issue first reported in Chibi.
>>
>> The definition of `reduce` is quite clear that it only uses the
>> `ridentity` in the empty list case.
>> It further provides a detailed note explaining this and the motivation.
>>
>> The definition of `reduce-right` both in the expansions given and the summary:
>>
>>   "... in other words, we compute (fold-right f ridentity list)"
>>
>> seems to make it clear that the ridentity is used even in the
>> non-empty list case.
>>
>> However, this makes no sense, because then there is no difference
>> between fold-right and reduce-right.
>> I think it's clear the same note was intended for reduce-right as for reduce.
>>
>> Also, the reduce-right definition incorrectly expands into reduce:
>>
>>   (reduce-right f ridentity '(e1 e2 ...)) =
>>     (f e1 (reduce f ridentity (e2 ...)))
>>
>> At least there should be an errata to replace that expansion with
>> `reduce-right`.
>> Arguably it should be modified further:
>>
>>   (reduce-right f ridentity '(e1 e2 ...)) =
>>     (fold-right f e1 (e2 ...))
>>
>> updating the summary to something like
>>
>>   "... in other words, we compute (fold-right f ridentity list), but
>> as with reduce only use ridentity in the empty list case."
>>
>> --
>> Alex
>>
>> On Sun, Jan 9, 2022 at 1:48 AM Arthur A. Gleckler <xxxxxx@speechcode.com> wrote:
>> >
>> > On Tue, Dec 28, 2021 at 12:50 AM TAKIZAWA Yozo <xxxxxx@nbk.bz> wrote:
>> >>
>> >> Although this is not a specification topic but implementation, I send
>> >> it to the list as a related issue to confirm or clarify.
>> >
>> >
>> > Thanks for the question.  Would you mind showing some examples to make the question clearer?  It would be great if you could show examples that differ between Scheme implementations.
>> >
>> > Thanks, and happy new year.