[RFC] assign Px, Py

During conversion of the MMD functions the last attacked group 
(logical_or, _and, _xor) turned out to be the toughest. Not because of 
the functionality per se, but due to the fact that a passed in 
destination PMC ought to get the value of the result.

1) semantics of assign

This is achieved by VTABLE_assign_pmc( [INTERP, SELF,] value ) [1], 
which is desribed in pdd06_pasm roughly as:

     assign actually does a value assignment, as:

                $foo = $bar;

     X's assign vtable method is invoked and it does whatever is 
appropriate.

The current implementation in classes/scalar.pmc does basically this:

   assign Px, Py:

    if y.isa("Integer")
       val = y.get_integer
       x.set_integer_native(val)
    else if y.sa("Float")
       val = y.get_number
       x.set_number_native(val)
    ...

The "set_<type>_native" functions morph themselves to the appropriate 
PMC and then set the value. Works fine for Integer and Float. But e.g. a 
Boolean PMC doesn't morph itself to the value, it sets it's boolean 
value according to the passed in value:

     void set_integer_native (INTVAL value) {
         PMC_int_val(SELF) = (value != 0);
     }

String was similar, it did set self's value to a string representaion of 
the value.

These differing semantics cause a lot of exceptions in C<assign_pmc> and 
it's error prone as all unhandled types might do it differently.

As a side-node: it seems to me that C<assign_pmc> is easier to 
implement, when the vtable is called on the value, which then morphs the 
destination according to it's own type and then sets the value. I.e.

    assign_pmc(PMC* dest)

set C<dest>'s value to C<SELF>'s value. Or:

   assign Px, Py     <=>   Py->vtable->assign_pmc(Px)

With this implemetation only one type has to be considered, e.g.

   Integer.assign_pmc(dest) {
      v = SELF.get_integer
      dest.morph(Integer_type)
      dest.set_integer_native(v)
   }


2) semantics of logical_*

Currently (if *dest == NULL) a clone of the result PMC is returned. I'm 
not sure, if HLL's wouldn't need the original PMC (i.e. the reference). 
I tried this but a few tests broke, which didn't cope with this behavior.

3) PIR syntax

It was already discussed a few times that we might change PIR syntax:

current:

   Px = Py            set Px, Py          alias Px and Py
   Px = assign Py     assign Px, Py       copy values

future:

   Px := Py           set Px, Py
   Px = Py            assign Px, Py

This would much more resemble the HLL's (and programmers) POV.

Comments very welcome,
leo

[1] it used to be C<set_pmc>, which is still called as a default, if 
C<assign_pmc> isn't implemented.

• [RFC] assign Px, Py by Leopold Toetsch
• Re: [RFC] assign Px, Py by Brent 'Dax' Royal-Gordon
• Re: [RFC] assign Px, Py by Leopold Toetsch