2019-10-04 16:38:51 +02:00
|
|
|
# Slot Language
|
|
|
|
|
|
|
|
This is a new layer / approach to create code for the next level, the slot_machine.
|
|
|
|
|
|
|
|
In essence the slot_language lets us code the slot_machine.
|
|
|
|
|
|
|
|
## Problem
|
|
|
|
|
2019-12-01 11:09:24 +01:00
|
|
|
The Problem with the current way is that some of the slot_machine instructions are
|
|
|
|
really quite complex. They are really more functions than instructions.
|
2019-10-04 16:38:51 +02:00
|
|
|
|
|
|
|
This is especially true for everything around the dynamic call. Dynamic call itself
|
2019-10-05 13:49:45 +02:00
|
|
|
is still ok, but resolve_method is too much. And it even uses method_missing, another
|
2019-10-04 16:38:51 +02:00
|
|
|
instruction that is too much, which in turn should use raise and now we really see
|
|
|
|
the point.
|
|
|
|
|
2019-10-05 13:49:45 +02:00
|
|
|
I thought about making those "super" instruction real methods and just calling them,
|
2019-10-04 16:38:51 +02:00
|
|
|
but the calling overhead is just too much, and really it is the wrong tool for the
|
|
|
|
job. Calling implies switching of context, while resolve_method and raise and mm
|
|
|
|
really all operate on the same context.
|
|
|
|
|
|
|
|
## The Slot Machine
|
|
|
|
|
|
|
|
The Slot Machine is a kind of memory based, oo abstraction of the risc machine, that in
|
|
|
|
turn mirrors a cpu relatively closely. The machine "knows" the message in the way a
|
2019-10-05 13:49:45 +02:00
|
|
|
cpu knows its registers. And the oo part means it also knows the parfait object
|
2019-10-04 16:38:51 +02:00
|
|
|
model.
|
|
|
|
|
|
|
|
While Ruby/Sol code only ever assumes the type of self, the Slot Machine assumes types
|
|
|
|
of the whole of Parfait. The main instruction after which the machine is named is
|
|
|
|
the SlotLoad, which moves one instance variable to another.
|
|
|
|
|
|
|
|
## Code for the Machine
|
|
|
|
|
|
|
|
Since the Slot and SlotMachine have proven useful abstractions, this introduces the
|
|
|
|
SlotLanguage as a way to create code for the SlotMachine.
|
|
|
|
|
|
|
|
The SlotMachine defines no methods on objects and passes no objects. While it has call
|
|
|
|
and return, these are defined in terms of jumps and use, like all Slot instructions,
|
|
|
|
the message.
|
|
|
|
|
|
|
|
## Syntax (projection)
|
|
|
|
|
2019-12-01 11:09:24 +01:00
|
|
|
Since we are not defining methods, there is no separate scope. We create objects that
|
2019-10-05 13:49:45 +02:00
|
|
|
will transform to SlotMachine Instructions _in_ the scope of the current method.
|
|
|
|
In other words they will have access to the compiler and the callable, when transforming
|
2019-12-01 11:09:24 +01:00
|
|
|
to SlotMachine (similar to Sol in that way). This means at compile time we
|
2019-10-05 13:49:45 +02:00
|
|
|
can use the frame type and constants, while we can always assume the Message (and not
|
|
|
|
much else) at runtime.
|
|
|
|
|
|
|
|
As the scope is "fixed", we will use the file scope, ie one file defines one
|
|
|
|
instruction/macro, by convention of the same name.
|
2019-10-04 16:38:51 +02:00
|
|
|
|
|
|
|
For starters we will use ruby syntax, with these semantics:
|
|
|
|
- only globals and message (the literal) are valid variable names
|
|
|
|
- dot will mean pointer traversal, sort of like c (no calling)
|
|
|
|
- names on right hand of dot must be instance variables of left
|
|
|
|
- equal will mean assignment in the SlotLoad kind of sense
|
|
|
|
- some macro mechanism is called for (tbd)
|
|
|
|
- maybe labels have to be created (sort of like the risc dsl)
|
|
|
|
|
|
|
|
The result of a compilation will be a SlotMachine Macro
|