References ...

[Leopold Toetsch]

... are autogenrated sice some time. They delegete all but a few methods
to the refered PMC. [1]
But there are some pieces missing IMHO:
There is no means to get at the type of what the Ref refers too.
And we can't dereference the ref.

I'm thinking of 2 new ops:

deref Px, Py # set Px to what Ref Py refers to
ref S0, Py # := typeof S0, Py->referee
ref I0, Py # := typeof I0, Py->referee

The "deref" opocde could call vtable->get_pmc, which isn't covered by
any opcode yet (assign does a set_pmc - but we don't have the opposite).

This could be also useful for Keys. We can do:
new P0, .Key
new P1, .PerlString
set P1, "key"
assign P0, P1
But there is no opcode to get the PerlString out of the key.

Comments welcome,
leo

[1]
new P1, .PerlString
set P1, 42
new P0, .Ref, P1

print P0
print "\n"

inc P1 # or inc PO
print P0
print "\n"

typeof S0, P0
print S0
print "\n"
typeof S0, P1
print S0
print "\n"
end
42
43
Ref
PerlInt

[Leopold Toetsch]

Melvin Smith <mel...@us.ibm.com> wrote:

> I think if you have the op for dereferencing, you don't need the
> additional ops for
> getting the type of the reference.

Too true, thanks
leo

[Nicholas Clark]

Yes, but if you have to go via another PMC register then you may spill.
Whereas it seemed that the type retrieval ops went direct to S and I
registers, which may save spillage. (at a cost of more ops)

Nicholas Clark

[Melvin Smith]

Well, 1 extra register is required which shouldn't be enough to worry
about. Spillage is pretty rare with the number of registers we have
so I doubt it will make a difference.

-Melvin