Benjamin Goldberg <[EMAIL PROTECTED]> wrote:
> ... I'd rather have two versions of add, one
> with set/create semantics and the other with assign/mutate semantics.
As already outlined, this leads to some kind of duplicating of our
opcodes, vtables, penalty on JIT implementators and so on. We (/me)
don't need this.
Its up to the HLL, how to program a CPU (virtual or not). You generate
a sequence of opcodes that do the Right Thing[1] (Fast)[2]. This may
imply to use a "assign", "clone", "set" or whatever opcode and an
"add" opcode, depending on the semantics or whatever the language
might need (references or values or ...)[3]
[1] there must be a minimal set of opcodes, that doesn't prohibit the
programmar ... If Something is really missing, we'll add some
wires to the (VM) hardware.
[2] If a sequence of ops is heavily used and some shortcut (micro
op) comes out to be a lot faster, we'll rewire the engine and
implement it in core.
[3] most of that stuff can be handled by the vtable implementing e.g.
the C<add> method. It can do everything with a (yes - existing)
destination PMC.
just my 2c
leo
