{register int i, *p; p = &i;}

[Wayne Throop]

> markb at sdcrdcf.UUCP (Mark Biggar)
> IT still doesn't work for REGISTER variables!

True, true.  Both K&R and the draft standard explicitly deem this
illegal.  But it is worth noting what Harbison and Steele have to say
about it:

    7.4.6 Address Operator

        [blah, blah, blah, blah, and then the interesting part]

    The original description of C specified that it was illegal to apply
    the unary address operator '&' to a register variable, and some
    compilers still enforce this restriction.  However, since register
    is treated only as a hint to the compiler and not a mandatory
    requirement, it seems appropriate to allow such usage, inasmuch as
    on some computers the registers really are addressable as if they
    were memory locations.  On the other hand, when the target computer
    does not have addressable registers, applying '&' to a register
    variable may simply defeat the declaration of the variable as
    register, forcing it to be of class auto instead.  We recommend that
    new compilers take this latter approach and perhaps also issue a
    warning message.  In any case, such usage should be regarded as
    nonportable.

This makes three points:

  - 'register' is "only" a hint to the compiler, and the declaration of
    a variable as register does not ensure that the variable *will* be
    in a register
  - therefore, '&' of register variables "ought" to be allowed, and
    simply be one of those cases where the compiler might be compelled
    to ignore the register hint
  - (despite this "ought", it is certainly non-portable, given current
     compilers and draft standards)

And to conclude, remember that the original posting in this series was a
request to make a macro "swap" such that

        swap( *a++, *b++ );

would work, and the posted examples would work even if a and b were
declared register.

--
"Eschew obfuscation."
"Giesundheit!"
-- 
Wayne Throop      <the-known-world>!mcnc!rti-sel!dg_rtp!throopw