This evening, I decided to take another look at SRFI 147/148.

I was able to fix the problem reported by Daphne.  It has two root causes:

- Some internal library of Chibi takes precedence over what the
implementation provides unless one runs Chibi with an explicit "-I.".
I guess that Chibi's (trivial) implementation of SRFI 147 doesn't do
the things expected by SRFI 148.

- As Chibi changed how it matches an unbound identifier against a
bound one (now R7RS-compatible), I had to insert a syntax-quote in the
er-macro-transformer wrapper.

Arthur: I am going to send you a pull request soon.

Am Mi., 20. Apr. 2022 um 10:53 Uhr schrieb Daphne Preston-Kendal
<xxxxxx@nonceword.org>:
>
> On 20 Apr 2022, at 10:17, Marc Nieper-Wißkirchen <xxxxxx@gmail.com> wrote:
>
> > With Chibi 0.9, all tests pass; with Chibi 0.10 it's broken. What
> > change between 0.9 and 0.10 may be responsible?
>
>
> I ran Git bisect between the 0.9 tag and HEAD, rebuilding Chibi from scratch each time and then running the sample implementation test suite. It identified this commit as the culprit:
>
> > 42aab7905c8420e7c5475b67f286e87a730a7873 is the first bad commit
> > commit 42aab7905c8420e7c5475b67f286e87a730a7873
> > Author: Alex Shinn <xxxxxx@gmail.com>
> > Date:   Mon Apr 19 10:27:46 2021 +0900
> >
> >     enable SEXP_USE_STRICT_TOPLEVEL_BINDINGS by default (issue #699)
> >
> >  include/chibi/features.h | 16 ++++++++++++++--
> >  1 file changed, 14 insertions(+), 2 deletions(-)
>
> I should note that the commits without the crash fail one test anyway:
>
> > %%%% Starting test SRFI 148  (Writing full log to "SRFI 148.log")
> > FAIL em-cute: without ...
> > # of expected passes      141
> > # of unexpected failures  1
>
> I marked them good nonetheless since that’s not the bug I was investigating.
>
>
> Daphne
>