Px = n_add Py, Pz    # create new Px as sum(Py, Pz)

and that overloaded infix functions have to return the result.

For Python, Lisp and probably more HLLs the same is of course needed for
unary opcodes:

   abs Px, Py           # use existing abs, Px exists
   Px = n_abs Py        # create new abs result PMC

This is of course just a fancy way to write

   n_abs Px, Py         # same

but it makes it more clear that a new value is assigned.

For the existing unary opcodes with PMC variants:

   abs, neg, not, nots (don't we have more?)

the same scheme as to infix opcodes apply: if the passed in C<dest> PMC
is NULL, a new result PMC is created. I don't know, if we should have
distinct inplace variants too:

   abs Px               # VTABLE_i_absolute()

I'd like additionally to have PMC variants of mostly existing vtables
that currently return native types only:

   PMC *get_str()        # string context
   PMC *get_num()        # numeric context
   PMC *get_int()        # get integer (maybe)
   PMC *get_list()       # list context
   PMC *hash()           # get hash value - we have INTVAL hash()

The exsiting C<INTVAL get_bool()> should better be C<get_bool_native> so
that we have too:

   PMC *get_bool()
   PMC *get_bignum()     # exists
   PMC *get_complex()    # return a new complex with SELF.value

I think these all should just return a new result PMC.

  abs, neg, not, bnot, bnots

and new:

  n_abs, n_neg, n_not, n_bnot, n_bnots