Stefan Monnier <monnier@iro.umontreal.ca> writes:

> This can't be right because it presumes the CLOS semantics which we
> don't have yet: `:initform x` will use the symbol `x` rather than the
> value of variable `x` as the default value (and it will emit a warning
> because we don't want code to rely on this non-CLOS-compatible
> semantics).

Right.  I had to do it in Emacs 27 but it's implemented differently in
28, as should have been evident to me given the current code in
initialize-instance.

> we want a test that fails on the current code
> and succeeds after we install your patch.

OK; I replaced the tests.