assq compatibility with Maclisp

[Glenn S. Burke <GSB at MIT-ML>]

    Date: Saturday, 10 April 1982, 06:04-EST
    From: JCMA@MIT-AI
    Why do assq and assoc have to be incompatible with maclisp?  That is,
    why aren't assoc-careful and assq-careful the defaults?
----
It happens that Maclisp does not actually do assq-carefully, at least
not deliberately.  It simply fails to check, and so can be fooled.
---- [transition to bug-lisp report] ----
uh oh...  Even though i "knew" what maclisp was doing, i figured that
i would test it to see.  The little cretin does assq-carefully in
(*RSET T) mode, and doesn't in (*RSET NIL) mode.  What this means is
that the following happens in (*rset nil)
  (assq nil '(0)) => 0
and in (*rset t) it yields NIL.  The reason for this is that the CAR of
the fixnum 0 gives NIL.  Similarly one can create other fixnums (and flonums)
which will fool assq for any other object.
So much for features.  This sounds like a totally useless, misleading,
and inconsistent Maclisp feature.  It is a waste of the address space
which someone has been trying hard to conserve.  If it is going to check,
it should bitch, it shouldn't conditionally do the wrong thing.