Probably we should move to srfi-135 to clarify this, but my take that srfi-135 allowed mapper to return
a string is that you do need to treat a sequence of codepoints as a unit in Unicode world.  Like a base character plus
combining character(s).   And it doesn't make sense to reverse its order.


On Mon, Mar 2, 2020 at 1:45 PM Per Bothner <xxxxxx@bothner.com> wrote:
Like with string-split (see separate reply) the specification for string-unfold-right
was copied (without much thought) from SRFI-135's text-unfold-right:
https://srfi.schemers.org/srfi-135/srfi-135.html#text-unfold-right

I don't think reversing the characters in a string is a useful operation,
so I don't think we should reverse the sub-string returned by the mapper.
I.e. the Kawa implementation is correct.

However, I could be convinced if reversing makes the procedure more useful.

On 3/2/20 3:12 PM, Arthur A. Gleckler wrote:
> Per, when you have a chance, would you please reply to Chris's comments below?
>
> Thanks.
>
> On Sun, Nov 17, 2019 at 6:34 PM Chris Hanson <xxxxxx@chris-hanson.org <mailto:xxxxxx@chris-hanson.org>> wrote:
>
>     I'm implementing this SRFI and have noticed an ambiguity in the
>     specification.
>
>     string-unfold-right specifies that the results of the mapper procedure
>     are accumulated from right to left in the result.  When the mapper
>     returns a character, that's not a problem, and the specification is
>     clear.
>
>     However, when the mapper returns a string, it's not specified how the
>     characters of the string are accumulated.  One might imagine the
>     characters being prepended to the result one at a time in reverse order,
>     as if it was returned by multiple calls to the mapper in the string
>     order, or the string might be simply prepended to the output being
>     built.
>
>     I see that Kawa implements the latter.
>
>     I don't care what the actual semantics are, as long as they are clearly
>     specified.
>
>     I suppose it's worth spelling out for string-unfold too, though in that
>     case it seems like the obvious solution is to just append the string,
>     since the two orders are congruent.
>
>     Another concern would be whether someone expects that string-unfold and
>     string-unfold-right have some kind of deep symmetry, in which the
>     resulting strings are the reverse of one another, even when some mapper
>     values are strings.  In that case it might be more elegant to prepend
>     the string's characters in reverse order rather forward order.  I'm not
>     making that argument, though I can understand the appeal.  I just want a
>     clear specification.
>


--
        --Per Bothner
xxxxxx@bothner.com   http://per.bothner.com/