(SETF (CAR x) y) in tests about jscl HOT 2 CLOSED

I had a look. The reason is that setf macroexpands the place forms using the host get-setf-expansion function, instead of the target one. Why? Because get-setf-expansion is not available to call yet, of course.. so you get a call to %rplaca, which is the internal SBCL function for (setf car).

(get-setf-expansion `(car x))
;; =>
(#:X29814)
(X)
(#:NEW29813)
(SB-KERNEL:%RPLACA #:X29814 #:NEW29813)
(CAR #:X29814)

It should be possible to fix it hacking the setf implementation a little bit to keep setf-expanders at compile-time in our own structure. Something similar what we do with macros itself.

This change would look like moving setf implementation to its own file, which should be loaded both in host and in target, prefixing with a bang character to avoid clashes and defining aliases to the current names in the target.

from jscl.

I implemented a basic eval-when (30f1522). It allowed to fix the semantic of define-setf-expander at compile-time (317e7d8). In particular it fixes this ticket.

from jscl.