On Mon, Jun 07, 2021 at 05:45:16PM +0200, Daphne Preston-Kendal wrote:
> I've written a long review of the status quo regarding named (keyword) and optional parameters in various Schemes which I've now published online. It includes proposals for things we can do right now to get keyword arguments interoperably and/or portably, albeit less than satisfactory in some cases, pending further input. (I had hoped to release it as a SRFI and eventually produce a final version, but it’s not a solid enough library proposal for a SRFI yet.) You can read the review at <https://dpk.gitlab.io/presrfis/arguments.html>.

Hi!

This is quite the tour de force.  I've skimmed it and seems to make
sense.  One nitpick: at the end you mention:
"in Chicken [keywords] are symbols, but in DSSSL and other
 implementations they are a disjoint type"

This used to be true, but since 5.1.0 (released in July 2019) we've
successfully managed to turn them into a disjoint type.  The conflation
of symbols and keywords was made possible due to a godawful hack that
interfered with read/write invariance which we've now thankfully dropped
completely by making them disjoint.  This surfaced several bugs in macros
and also pointed out that people were using forms like (define foo: ...)
which would result in a keyword being bound, so that it would no longer
self-evaluate.

In the section "keyword datums" you correctly group CHICKEN with the
schemes in which keywords are disjoint from symbols, so maybe this
was a remnant from an earlier iteration?

Also, you state that in Gauche, keywords respond #t to symbol?, but
I do not see this in 0.9.6 on Debian.  However, if I invoke "gosh"
with an unrecognised switch like "-h", I see GAUCHE_KEYWORD_IS_SYMBOL
and GAUCHE_KEYWORD_DISJOINT in the list of environment variables.
It looks like this influences the behaviour of keywords, and it also
mentions that the former environment variable is to make it "fully R7RS
conformant".  Maybe worth a second look.

Cheers,
Peter