Per Bothner scripsit:

> I don't think any implementation that allows over-writing literals is
> a good candidate for "Scheme-large".

See my previous post.  However, I have added this to the MandatoryDocket
so that it can be voted on in due course.

> I know Fortran allowed it in the 50s, but that has not been considered
> kosher since then.

For the record, a *bug* in one particular Fortran compiler allowed
overwriting of *atomic* function arguments that were literals, actually,
with possible knock-on effects due to hidden sharing of those literals.

> So is the goal for WG2 to only add features that can be implemented as
> pure-library additions?  If so, why is SRFI-124 (ephemerons) even in
> contention?

No, it isn't, although it is an informal goal for the Red Edition, on
the assumption that it's easier to get agreement on things early when
the barriers to adoption are low.  Later editions will have features
that cannot have R7RS-small portable implementations, though I hope that
they will be easily implemented on many Schemes.

Ephemerons in particular would certainly have been postponed if Will
Clinger had not pointed out that an implementation using strong references
is legal.  This is because there is never any guarantee that the GC will
break specific, or any, ephemerons.  The JVM doesn't have ephemerons,
but an implementation using weak references is valid: it simply means
that some garbage that should be collected will not be collected, just
as conservative GCs do.

--
John Cowan          http://www.ccil.org/~cowan        xxxxxx@ccil.org
Thor Heyerdahl recounts his attempt to prove Rudyard Kipling's theory
that the mongoose first came to India on a raft from Polynesia.
        --blurb for Rikki-Kon-Tiki-Tavi