Files
not-python-rust/src/vm.rs

417 lines
14 KiB
Rust
Raw Normal View History

Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
use std::collections::HashMap;
use std::rc::Rc;
use crate::obj::function::*;
use crate::obj::*;
#[derive(Debug, Clone, Copy, PartialEq)]
pub enum Op {
// Stack functions
Pop,
PushConstant(LongOpArg),
// Variables
GetLocal(LocalIndex),
SetLocal(LocalIndex),
GetGlobal(GlobalId),
SetGlobal(GlobalId),
// Attributes
GetAttr(ConstantId),
SetAttr(ConstantId),
// Jumps
Jump(JumpOpArg),
JumpFalse(JumpOpArg),
JumpTrue(JumpOpArg),
// Functions
Call(Argc),
Return,
CloseOver { depth: ShortOpArg, slot: ShortOpArg },
// VM control
Nop,
Halt,
}
pub type LineRange = (usize, usize);
type ShortOpArg = u16;
type LongOpArg = u32;
pub type JumpOpArg = i32;
pub type LocalIndex = LongOpArg;
pub type LocalSlot = ShortOpArg;
pub type ConstantId = LongOpArg;
pub type GlobalId = LongOpArg;
pub type Argc = LongOpArg;
pub type FrameDepth = ShortOpArg;
#[derive(Debug, Clone)]
pub struct Local {
pub(crate) slot: LocalSlot,
pub(crate) index: LocalIndex,
pub(crate) name: String,
}
#[derive(Debug, Default, Clone)]
pub struct Chunk {
pub(crate) code: Vec<Op>,
pub(crate) lines: Vec<LineRange>,
pub(crate) locals: Vec<Local>,
}
#[derive(Debug)]
pub(crate) enum Function {
Chunk(Rc<Chunk>),
Builtin(BuiltinFunctionPtr, FunctionState),
}
#[derive(Debug)]
pub struct Frame {
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
pub(crate) name: Rc<String>,
pub(crate) function: Function,
pub(crate) ip: usize,
pub(crate) stack_base: usize,
}
impl Frame {
pub fn new(name: Rc<String>, function: Function, stack_base: usize) -> Self {
Self {
name,
function,
ip: 0,
stack_base,
}
}
}
pub struct Vm {
constants: Vec<ObjP>,
//global_names: Vec<String>,
globals: Vec<ObjP>,
stack: Vec<ObjP>,
frames: Vec<Frame>,
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
builtins: HashMap<String, ObjP>,
}
impl Vm {
/// Create a new virtual machine with the given chunk, constants, and global names.
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
pub fn new(
chunk: Rc<Chunk>,
constants: Vec<ObjP>,
global_names: Vec<String>,
builtins: HashMap<String, ObjP>,
) -> Self {
// set up globals
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
let nil = builtins.create_nil();
let mut globals: Vec<_> = global_names.iter().map(|_| ObjP::clone(&nil)).collect();
let mut register_global = |name: &str, value: ObjP| {
let index = global_names
.iter()
.position(|global| global == name)
.expect("could not find global");
globals[index] = value;
};
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
for (name, builtin) in builtins.iter() {
register_global(&name, ObjP::clone(&builtin));
}
// stack and frames
let stack = Vec::new();
let frames = vec![Frame::new(
"__main__".to_string().into(),
Function::Chunk(chunk),
0,
)];
Vm {
constants,
//global_names,
globals,
stack,
frames,
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
builtins,
}
}
/// Get the stack.
pub fn stack(&self) -> &Vec<ObjP> {
&self.stack
}
/// Get the stack, mutably.
pub fn stack_mut(&mut self) -> &mut Vec<ObjP> {
&mut self.stack
}
/// Gets the current stack, starting at the frame's stack base.
pub fn frame_stack(&self) -> &[ObjP] {
&self.stack()[self.frame().stack_base..]
}
/// Current stack frame.
pub fn frame(&self) -> &Frame {
self.frames.last().unwrap()
}
/// Current stack frame, mutably.
pub fn frame_mut(&mut self) -> &mut Frame {
self.frames.last_mut().unwrap()
}
/// Push a new stack frame.
pub fn push_frame(&mut self, frame: Frame) {
self.frames.push(frame);
}
/// Pop the current stack frame.
pub fn pop_frame(&mut self) -> Frame {
self.frames.pop().expect("no frame")
}
/// Gets the chunk of the currently executing frame.
pub fn chunk(&self) -> Option<&Chunk> {
if let Function::Chunk(chunk) = &self.frame().function {
Some(chunk)
} else {
None
}
}
/// Instruction pointer of the current frame.
pub fn ip(&self) -> usize {
self.frame().ip
}
/// Update the current instruction pointer.
pub fn set_ip(&mut self, ip: usize) {
self.frame_mut().ip = ip;
}
/*
/// Gets the line of the current instruction.
fn line(&self, offset: isize) -> LineRange {
let index = (((self.ip() as isize) + offset).max(0) as usize).min(self.chunk().lines.len());
self.chunk().lines[index]
}
*/
/// Get the current instruction and advance the IP.
///
/// This may actually end up calling a function on top of the stack, if it's a builtin function.
fn dispatch(&mut self) -> Op {
let mut ip = self.ip();
let op = match &self.frame().function {
Function::Chunk(chunk) => {
let op = chunk.code[ip];
ip += 1;
op
}
Function::Builtin(function, state) => {
// keep track of where the current frame index is in case we need to yield
let frame_index = self.frames.len() - 1;
let result = (function)(self, *state);
match result {
FunctionResult::ReturnPush(value) => {
// push value to the stack and let the VM handle return protocols
self.push(value);
Op::Return
}
// value is already on top of the stack, let the VM handle return protocols
FunctionResult::Return => Op::Return,
// new stack frame has been pushed, yield control while keeping track of the
// old state
FunctionResult::Yield(resume_state) => {
// update the current state
if let Function::Builtin(_function, ref mut state) =
&mut self.frames[frame_index].function
{
*state = FunctionState::Resume(resume_state)
} else {
panic!("function stack got really messed up - function stack was changed under us");
}
// inject a no-op so the VM can load a new instruction or dispatch a new
// instruction.
Op::Nop
}
}
}
};
self.set_ip(ip);
op
}
/// Pop a value from the stack.
pub fn pop(&mut self) -> ObjP {
self.stack.pop().expect("stack empty")
}
/// Peek the top value of the stack.
pub fn peek(&self) -> ObjP {
self.stack.last().map(Ptr::clone).expect("stack empty")
}
/// Push a value to the stack.
pub fn push(&mut self, value: ObjP) {
self.stack.push(value);
}
pub fn run(&mut self) {
// Cached pointers that we just always want to have on hand
let method_type = self.builtins().get("Method").unwrap().clone();
loop {
match self.dispatch() {
Op::Pop => {
self.pop();
}
Op::PushConstant(constant_id) => {
let constant = Ptr::clone(&self.constants[constant_id as usize]);
self.push(constant);
}
Op::GetLocal(local_index) => {
let local = &self.chunk().expect("no chunk").locals[local_index as usize];
let value =
Ptr::clone(&self.stack[self.frame().stack_base + local.slot as usize]);
self.push(value);
}
Op::SetLocal(local_index) => {
let value = self.pop();
let local = &self.chunk().expect("no chunk").locals[local_index as usize];
let index = self.frame().stack_base + local.slot as usize;
self.stack[index] = value;
}
Op::GetGlobal(global_index) => {
let value = Ptr::clone(&self.globals[global_index as usize]);
self.push(value);
}
Op::SetGlobal(global_index) => {
let value = self.pop();
self.globals[global_index as usize] = value;
}
Op::GetAttr(constant_id) => {
// need both declarations to borrow cell value
let name_obj = Ptr::clone(&self.constants[constant_id as usize]);
let name =
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
with_obj_downcast(name_obj, |name: &StrInst| Rc::clone(&name.str_value()));
let owner = self.pop();
let value =
owner
.borrow_mut()
.get_attr_lazy(owner.clone(), method_type.clone(), &name);
if let Some(value) = value {
self.push(value);
} else {
// TODO Vm::run, Op::GetAttr - throw an exception when the attribute
// doesn't exist
// BLOCKED-ON: exceptions
todo!(
"throw an error because we couldn't read attr '{}' on '{}'",
name,
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
owner.borrow(),
);
}
}
Op::SetAttr(constant_id) => {
let name_obj = Ptr::clone(&self.constants[constant_id as usize]);
let name =
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
with_obj_downcast(name_obj, |name: &StrInst| Rc::clone(&name.str_value()));
let value = self.pop();
let target = self.pop();
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
target.borrow_mut().set_attr(&name, value);
}
Op::Jump(offset) => {
let base = (self.ip() - 1) as JumpOpArg;
assert!(base + offset > 0, "tried to jump to negative IP");
self.set_ip((base + offset) as usize);
}
Op::JumpFalse(offset) => {
let base = (self.ip() - 1) as JumpOpArg;
let value = self.peek();
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
if !value.borrow().is_truthy() {
self.set_ip((base + offset) as usize);
}
}
Op::JumpTrue(offset) => {
let base = (self.ip() - 1) as JumpOpArg;
let value = self.peek();
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
if value.borrow().is_truthy() {
self.set_ip((base + offset) as usize);
}
}
Op::Call(argc) => {
let argc = argc;
let index = self.stack.len() - (argc as usize) - 1;
let fun_ptr = Ptr::clone(&self.stack[index]);
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
let arity = if let Some(arity) = fun_ptr.borrow().arity() {
arity
} else {
// TODO Vm::run, Op::Call - throw an exception when the value isn't
// callable
// BLOCKED-ON: exceptions
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
todo!(
"throw an error because we couldn't call {}",
fun_ptr.borrow()
);
};
// Check arity
if arity != argc {
// TODO Vm::run, Op::Call - throw an exception when the number of arguments
// does not match the function's arity
// BLOCKED-ON: exceptions
todo!(
"throw an error because we passed the wrong number of arguments to {}",
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
fun_ptr.borrow()
);
}
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
fun_ptr.borrow().call(self, argc as Argc);
}
Op::Return => {
let return_value = self.pop();
let old_frame = self.pop_frame();
// stack_base is always going to be <= current stack size
self.stack
.resize_with(old_frame.stack_base, || unreachable!());
// also pop the function object off of the stack
self.stack.pop();
self.push(return_value);
}
Op::CloseOver { depth, slot } => {
// since we're closing over a value, and functions ultimately come from
// constants, we want to deep-clone this object so we don't alter any live
// objects.
// there is some room for optimization here so we aren't cloning the entire
// UserFunctionInst for every individual capture in a function.
let fun_ptr = self.pop();
let mut fun: UserFunctionInst =
with_obj_downcast(fun_ptr, UserFunctionInst::clone);
let frame_index = self.frames.len() - (depth as usize) - 1;
let stack_base = self.frames[frame_index].stack_base;
let value = Ptr::clone(&self.stack[stack_base + (slot as usize)]);
fun.push_capture(value);
self.push(make_ptr(fun));
}
Op::Nop => {
continue;
}
Op::Halt => {
break;
}
}
}
}
}
Revamp object system, start using `gc` crate Wow, what a ride. I think everything should be working now. In short: * Objects use the `gc` crate, which as a `Gc` garbage-collected pointer type. I may choose to implement my own in contiguous memory in the future. We will see. * The type system is no longer global. This is a bit of a burden, because now, whenever you want to create a new object, you need to pass its type object into the `Obj::instantiate` method, as well as its `::create` static method. * This burden is somewhat alleviated by the `ObjFactory` trait, which helps create new objects as long as you have access to a `builtins` hashmap. So something that would normally look like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = upcast_obj(BuiltinFunctionInst::create( ObjP::clone(&builtins.get("BuiltinFunction").unwrap()), "print", print, 1 ); builtins.insert("print".to_string(), print_builtin) // other builtins inserted here... } now looks like this: fn init_builtins(builtins: &mut HashMap<String, ObjP>) { let print_builtin = builtins.create_builtin_function("print", print, 1); builtins.insert("print".to_string(), print_builtin); } (turns out, if all you need is a HashMap<String, ObjP>, you can implement ObjFactory for HashMap<String, ObjP> itself(!)) Overall, I'm happier with this design, and I think this is what is going to get merged. It's a little weird to be querying type names that are used in the language itself to get those type objects, but whatever works, I guess. Next up is vtables. Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
2024-09-23 18:12:32 -07:00
impl ObjFactory for Vm {
fn builtins(&self) -> &HashMap<String, ObjP> {
&self.builtins
}
}