Thank you very much for your response!

Am Do., 9. Okt. 2025 um 14:34 Uhr schrieb Daphne Preston-Kendal
<xxxxxx@nonceword.org>:
>
> > Am So., 3. Aug. 2025 um 17:29 Uhr schrieb Marc Nieper-Wißkirchen
> > <xxxxxx@gmail.com>:
> >>
> >> Maybe I don't understand the semantics of your ephemeron-case form.
> >> Reading your first post, there does not seem to be a reference barrier
> >> on the key in the broken path. Can you give a syntax-case definition
> >> of ephemeron case?
>
>
> (define-syntax ephemeron-case
>   (syntax-rules (else)
>     ((_ eph-expr
>         ((k-id . v-id) expr_0 expr_1 ...)
>         (else broken-expr_0 broken-expr_1 ...))
>      (let* ((eph eph-expr)
>             (k (ephemeron-key eph))
>             (v (ephemeron-value eph))) ; why has this name been changed? grr
>        (if (ephemeron-broken? eph)
>            (begin broken-expr_0 broken-expr_1 ...)
>            (let-syntax ((k-id (identifier-syntax k))
>                         (v-id (identifier-syntax v)))
>              (dynamic-wind
>                (lambda () #f)
>                (lambda () expr_0 expr_1 ...)
>                (lambda () (reference-barrier k)))))))))
>
> > On 3 Aug 2025, at 17:30, Marc Nieper-Wißkirchen <xxxxxx@gmail.com> wrote:
> >
> > PS And maybe give a simple example of how a user would typically use it.
>
> (define (ephemeron-alist-ref ephal key)
>   (let loop ((more ephal))
>     (if (null? more)
>         (assertion-violation 'ephemeron-alist-ref "nothing found for key" ephal)
>         (ephemeron-case (car ephal)
>           ((found-key . value)
>            (if (eqv? found-key value)
>                value
>                (loop (cdr more))))
>           (else (loop (cdr more)))))))

[You likely mean "(eqv? found-key key)" here, I guess.]

Thank you for the example code. Unfortunately, the suggested
definition of "ephemeron-case" is useless. If there is no future call
of reference-barrier on K (in the case of ephemeron-case) or KEY (in
case of ephemeron-alist-ref), the compiler may elide the whole
not-broken-branch of ephemeron case (which is, a posteriori, correct
because reference-barrier won't be called in that case). But there is
a future call of reference-barrier on K/KEY, the call within
ephemeron-case is superfluous.

This is the semantics: "The GC can break an ephemeron if, under the
assumption that it is broken, the key will not be kept alive."

To correctly work with ephemerons, the user has to explicitly call
reference-barrier or procedures that document that they keep certain
values alive (unconditionally). See the example here:
https://srfi-email.schemers.org/srfi-254/msg/27973433/

Marc