[V] Handle macros, autoloads correctly in symbol-file. Add an incomplete TYPE arg.

[Stephen J. Turnbull]

Aidan Kehoe writes:

 >  > You broke the union build, and it's a real bug in the disunion build
 >  > too (`fun = Fcdr_safe(fun)' in doc.c:523 is a Lisp object and will
 >  > never be zero).
 > 
 > That was intentional

Ooh, yuck!  That code just "looks wrong" to me, specifically the naked
Lisp_Object in the midst of Lisp properly clothed in macros.

You could either make the assignment expression the argument of the
SUBRP (but you'd have to check the macro definition for argument
safety), or use the comma operator to sequence.

 >  > BTW, it looks to me like this probably should be an assert, but I
 >  > don't know where to look it up offhand.
 > 
 > No, if I understand you correctly that would make it trivial to crash XEmacs
 > from Lisp with:

Fine.  But shouldn't

 > (progn
 >   (fset 'whatever '(macro . not-a-subr))
 >   (built-in-symbol-file 'whatever))

should be caught and handled somehow?  And ... hm ...

 >  >  > 	Our load-history doesn't have the relevant metadata.

Do you plan to do anything about the lack of metadata?