nac3/nac3core/src/top_level.rs

493 lines
22 KiB
Rust
Raw Normal View History

2021-08-10 23:49:58 +08:00
use std::borrow::Borrow;
use std::{collections::HashMap, sync::Arc};
2021-08-03 13:38:27 +08:00
2021-08-05 14:55:23 +08:00
use super::typecheck::type_inferencer::PrimitiveStore;
2021-08-10 21:57:31 +08:00
use super::typecheck::typedef::{SharedUnifier, Type, TypeEnum, Unifier};
use crate::symbol_resolver::SymbolResolver;
2021-08-09 16:10:17 +08:00
use inkwell::{
basic_block::BasicBlock, builder::Builder, context::Context, module::Module,
types::BasicTypeEnum, values::PointerValue,
};
use parking_lot::RwLock;
2021-08-10 21:57:31 +08:00
use rustpython_parser::ast::{self, Stmt};
2021-08-03 13:38:27 +08:00
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Copy)]
pub struct DefinitionId(pub usize);
2021-08-03 13:38:27 +08:00
pub enum TopLevelDef {
Class {
// object ID used for TypeEnum
object_id: DefinitionId,
2021-08-03 13:38:27 +08:00
// type variables bounded to the class.
type_vars: Vec<Type>,
2021-08-07 15:06:39 +08:00
// class fields
2021-08-03 13:38:27 +08:00
fields: Vec<(String, Type)>,
// class methods, pointing to the corresponding function definition.
2021-08-07 15:06:39 +08:00
methods: Vec<(String, Type, DefinitionId)>,
2021-08-03 13:38:27 +08:00
// ancestor classes, including itself.
ancestors: Vec<DefinitionId>,
},
Function {
2021-08-07 15:06:39 +08:00
// prefix for symbol, should be unique globally, and not ending with numbers
name: String,
// function signature.
2021-08-03 13:38:27 +08:00
signature: Type,
/// Function instance to symbol mapping
/// Key: string representation of type variable values, sorted by variable ID in ascending
/// order, including type variables associated with the class.
/// Value: function symbol name.
instance_to_symbol: HashMap<String, String>,
/// Function instances to annotated AST mapping
/// Key: string representation of type variable values, sorted by variable ID in ascending
/// order, including type variables associated with the class. Excluding rigid type
/// variables.
/// Value: AST annotated with types together with a unification table index. Could contain
/// rigid type variables that would be substituted when the function is instantiated.
2021-08-05 14:55:23 +08:00
instance_to_stmt: HashMap<String, (Stmt<Option<Type>>, usize)>,
2021-08-03 13:38:27 +08:00
},
Initializer {
2021-08-10 21:57:31 +08:00
class_id: DefinitionId,
},
2021-08-03 13:38:27 +08:00
}
pub struct CodeGenTask {
pub subst: HashMap<usize, Type>,
pub symbol_name: String,
2021-08-05 14:55:23 +08:00
pub body: Stmt<Option<Type>>,
2021-08-03 13:38:27 +08:00
pub unifier: SharedUnifier,
}
pub struct TopLevelContext {
2021-08-05 14:55:23 +08:00
pub definitions: Arc<RwLock<Vec<RwLock<TopLevelDef>>>>,
pub unifiers: Arc<RwLock<Vec<SharedUnifier>>>,
2021-08-03 13:38:27 +08:00
}
2021-08-05 14:55:23 +08:00
pub struct CodeGenContext<'ctx> {
pub ctx: &'ctx Context,
pub builder: Builder<'ctx>,
pub module: Module<'ctx>,
pub top_level: &'ctx TopLevelContext,
pub unifier: Unifier,
2021-08-05 14:55:23 +08:00
pub resolver: Box<dyn SymbolResolver>,
pub var_assignment: HashMap<String, PointerValue<'ctx>>,
2021-08-07 15:06:39 +08:00
pub type_cache: HashMap<Type, BasicTypeEnum<'ctx>>,
pub primitives: PrimitiveStore,
2021-08-09 16:37:28 +08:00
// stores the alloca for variables
pub init_bb: BasicBlock<'ctx>,
2021-08-09 16:10:17 +08:00
// where continue and break should go to respectively
// the first one is the test_bb, and the second one is bb after the loop
pub loop_bb: Option<(BasicBlock<'ctx>, BasicBlock<'ctx>)>,
}
2021-08-10 23:49:58 +08:00
pub fn name_mangling(mut class_name: String, method_name: &str) -> String { // need to further extend to more name mangling like instantiations of typevar
class_name.push_str(method_name);
class_name
}
pub struct TopLevelDefInfo<'a> {
2021-08-10 21:57:31 +08:00
// like adding some info on top of the TopLevelDef for later parsing the class bases, method,
// and function sigatures
2021-08-10 23:49:58 +08:00
def: TopLevelDef, // the definition entry
ty: Type, // the entry in the top_level unifier
ast: Option<ast::Stmt<()>>, // the ast submitted by applications, primitives and class methods will have None value here
resolver: Option<&'a dyn SymbolResolver> // the resolver
}
2021-08-10 21:57:31 +08:00
pub struct TopLevelComposer<'a> {
pub definition_list: Vec<TopLevelDefInfo<'a>>,
pub primitives: PrimitiveStore,
pub unifier: Unifier,
}
impl<'a> TopLevelComposer<'a> {
pub fn make_primitives() -> (PrimitiveStore, Unifier) {
let mut unifier = Unifier::new();
let int32 = unifier.add_ty(TypeEnum::TObj {
2021-08-10 21:57:31 +08:00
obj_id: DefinitionId(0),
fields: HashMap::new().into(),
params: HashMap::new(),
});
let int64 = unifier.add_ty(TypeEnum::TObj {
2021-08-10 21:57:31 +08:00
obj_id: DefinitionId(1),
fields: HashMap::new().into(),
params: HashMap::new(),
});
let float = unifier.add_ty(TypeEnum::TObj {
2021-08-10 21:57:31 +08:00
obj_id: DefinitionId(2),
fields: HashMap::new().into(),
params: HashMap::new(),
});
let bool = unifier.add_ty(TypeEnum::TObj {
2021-08-10 21:57:31 +08:00
obj_id: DefinitionId(3),
fields: HashMap::new().into(),
params: HashMap::new(),
});
let none = unifier.add_ty(TypeEnum::TObj {
2021-08-10 21:57:31 +08:00
obj_id: DefinitionId(4),
fields: HashMap::new().into(),
params: HashMap::new(),
});
let primitives = PrimitiveStore { int32, int64, float, bool, none };
crate::typecheck::magic_methods::set_primitives_magic_methods(&primitives, &mut unifier);
(primitives, unifier)
}
2021-08-10 21:57:31 +08:00
pub fn new() -> Self {
let primitives = Self::make_primitives();
let definition_list: Vec<TopLevelDefInfo<'a>> = vec![
TopLevelDefInfo {
def: Self::make_top_level_class_def(0),
ast: None,
resolver: None,
2021-08-10 21:57:31 +08:00
ty: primitives.0.int32,
},
TopLevelDefInfo {
def: Self::make_top_level_class_def(1),
ast: None,
resolver: None,
2021-08-10 21:57:31 +08:00
ty: primitives.0.int64,
},
TopLevelDefInfo {
def: Self::make_top_level_class_def(2),
ast: None,
resolver: None,
2021-08-10 21:57:31 +08:00
ty: primitives.0.float,
},
TopLevelDefInfo {
def: Self::make_top_level_class_def(3),
ast: None,
resolver: None,
2021-08-10 21:57:31 +08:00
ty: primitives.0.bool,
},
TopLevelDefInfo {
def: Self::make_top_level_class_def(4),
ast: None,
resolver: None,
2021-08-10 21:57:31 +08:00
ty: primitives.0.none,
},
]; // the entries for primitive types
2021-08-10 21:57:31 +08:00
TopLevelComposer { definition_list, primitives: primitives.0, unifier: primitives.1 }
}
2021-08-10 23:49:58 +08:00
/// already include the definition_id of itself inside the ancestors vector
pub fn make_top_level_class_def(index: usize) -> TopLevelDef {
TopLevelDef::Class {
object_id: DefinitionId(index),
type_vars: Default::default(),
fields: Default::default(),
methods: Default::default(),
2021-08-10 23:49:58 +08:00
ancestors: vec![DefinitionId(index)],
}
}
pub fn make_top_level_function_def(name: String, ty: Type) -> TopLevelDef {
TopLevelDef::Function {
name,
signature: ty,
instance_to_symbol: Default::default(),
2021-08-10 21:57:31 +08:00
instance_to_stmt: Default::default(),
}
}
// like to make and return a "primitive" symbol resolver? so that the symbol resolver
// can later figure out primitive type definitions when passed a primitive type name
pub fn get_primitives_definition(&self) -> Vec<(String, DefinitionId, Type)> {
vec![
("int32".into(), DefinitionId(0), self.primitives.int32),
2021-08-10 23:49:58 +08:00
("int64".into(), DefinitionId(1), self.primitives.int64),
("float".into(), DefinitionId(2), self.primitives.float),
("bool".into(), DefinitionId(3), self.primitives.bool),
("none".into(), DefinitionId(4), self.primitives.none),
]
}
2021-08-10 21:57:31 +08:00
pub fn register_top_level(
&mut self,
ast: ast::Stmt<()>,
resolver: &'a dyn SymbolResolver,
) -> Result<Vec<(String, DefinitionId, Type)>, String> {
match &ast.node {
2021-08-10 21:57:31 +08:00
ast::StmtKind::ClassDef { name, body, .. } => {
let class_name = name.to_string();
2021-08-10 23:49:58 +08:00
let class_def_id = self.definition_list.len();
// add the class to the unifier
let ty = self.unifier.add_ty(TypeEnum::TObj {
2021-08-10 23:49:58 +08:00
obj_id: DefinitionId(class_def_id),
fields: Default::default(),
2021-08-10 21:57:31 +08:00
params: Default::default(),
});
2021-08-10 23:49:58 +08:00
let mut ret_vector: Vec<(String, DefinitionId, Type)> = vec![(class_name.clone(), DefinitionId(class_def_id), ty)];
// parse class def body and register class methods into the def list
// NOTE: module's symbol resolver would not know the name of the class methods,
// thus cannot return their definition_id? so we have to manage it ourselves?
// or do we return the class method list of (method_name, def_id, type) to
// application to be used to build symbol resolver? <- current implementation
// FIXME: better do not return and let symbol resolver to manage the mangled name
2021-08-10 23:49:58 +08:00
for b in body {
if let ast::StmtKind::FunctionDef {name, ..} = &b.node {
let fun_name = name_mangling(class_name.clone(), name);
let def_id = self.definition_list.len();
// add to unifier
let ty = self.unifier.add_ty(TypeEnum::TFunc(crate::typecheck::typedef::FunSignature {
args: Default::default(),
ret: self.primitives.none,
vars: Default::default()
}));
// add to the definition list
self.definition_list.push(
TopLevelDefInfo {
def: Self::make_top_level_function_def(fun_name.clone(), ty),
resolver: Some(resolver),
ty,
ast: None // since it is inside the class def body statments
}
);
ret_vector.push((fun_name, DefinitionId(def_id), ty));
// if it is the contructor, special handling is needed. In the above
// handling, we still add __init__ function to the class method
if name == "__init__" {
2021-08-10 23:49:58 +08:00
self.definition_list.push(
TopLevelDefInfo {
def: TopLevelDef::Initializer {
class_id: DefinitionId(class_def_id)
2021-08-10 23:49:58 +08:00
},
ty: self.primitives.none, // arbitary picked one
ast: None, // it is inside the class def body statments
resolver: Some(resolver)
}
)
// FIXME: should we return this to the symbol resolver?, should be yes
2021-08-10 23:49:58 +08:00
}
} else { } // else do nothing
}
// add to the definition list
self.definition_list.push(
TopLevelDefInfo {
def: Self::make_top_level_class_def(class_def_id),
resolver: Some(resolver),
ast: Some(ast),
ty,
}
);
Ok(ret_vector)
},
2021-08-10 21:57:31 +08:00
ast::StmtKind::FunctionDef { name, .. } => {
let fun_name = name.to_string();
let def_id = self.definition_list.len();
// add to the unifier
2021-08-10 21:57:31 +08:00
let ty =
self.unifier.add_ty(TypeEnum::TFunc(crate::typecheck::typedef::FunSignature {
args: Default::default(),
2021-08-10 23:49:58 +08:00
ret: self.primitives.none,
vars: Default::default()
}));
// add to the definition list
2021-08-10 21:57:31 +08:00
self.definition_list.push(TopLevelDefInfo {
2021-08-10 23:49:58 +08:00
def: Self::make_top_level_function_def(
name.into(),
self.primitives.none
),
resolver: Some(resolver),
ast: Some(ast),
ty,
2021-08-10 21:57:31 +08:00
});
Ok(vec![(fun_name, DefinitionId(def_id), ty)])
2021-08-10 21:57:31 +08:00
}
2021-08-10 21:57:31 +08:00
_ => Err("only registrations of top level classes/functions are supprted".into()),
}
}
/// this should be called after all top level classes are registered, and will actually fill in those fields of the previous dummy one
pub fn analyze_top_level(&mut self) -> Result<(), String> {
for mut d in &mut self.definition_list {
if let (Some(ast), Some(resolver)) = (&d.ast, d.resolver) {
match &ast.node {
ast::StmtKind::ClassDef {
bases,
body,
..
} => {
2021-08-10 23:49:58 +08:00
// get the mutable reference of the entry in the definition list, get the `TopLevelDef`
let (_,
ancestors,
fields,
methods,
type_vars
) = if let TopLevelDef::Class {
object_id,
ancestors,
fields,
methods,
type_vars
} = &mut d.def {
(object_id, ancestors, fields, methods, type_vars)
} else { unreachable!() };
// try to get mutable reference of the entry in the unification table, get the `TypeEnum`
let (params,
fields
) = if let TypeEnum::TObj {
params, // FIXME: this params is immutable, even if this is mutable, what should the key be, get the original typevar's var_id?
fields,
..
} = self.unifier.get_ty(d.ty).borrow() {
(params, fields)
} else { unreachable!() };
2021-08-10 21:57:31 +08:00
// ancestors and typevars associate with the class are analyzed by looking
// into the `bases` ast node
for b in bases {
match &b.node {
2021-08-10 23:49:58 +08:00
// typevars bounded to the class, things like `class A(Generic[T, V, ImportedModule.T])`
// should update the TopLevelDef::Class.typevars and the TypeEnum::TObj.params
ast::ExprKind::Subscript {value, slice, ..} if {
if let ast::ExprKind::Name {id, ..} = &value.node {
id == "Generic"
} else { false }
} => {
match &slice.node {
// `class Foo(Generic[T, V, P, ImportedModule.T]):`
ast::ExprKind::Tuple {elts, ..} => {
for e in elts {
// TODO: I'd better parse the node to get the Type of the type vars(can have things like: A.B.C.typevar?)
match &e.node {
ast::ExprKind::Name {id, ..} => {
// the def_list
type_vars.push(resolver.get_symbol_type(id).ok_or_else(|| "unknown type variable".to_string())?);
// the TypeEnum of the class
// FIXME: the `params` destructed above is not mutable, even if this is mutable, what should the key be?
unimplemented!()
},
_ => unimplemented!()
}
}
},
// `class Foo(Generic[T]):`
ast::ExprKind::Name {id, ..} => {
// the def_list
type_vars.push(resolver.get_symbol_type(id).ok_or_else(|| "unknown type variable".to_string())?);
// the TypeEnum of the class
// FIXME: the `params` destructed above is not mutable, even if this is mutable, what should the key be?
unimplemented!()
},
// `class Foo(Generic[ImportedModule.T])`
ast::ExprKind::Attribute {value, attr, ..} => {
// TODO:
unimplemented!()
},
_ => return Err("not supported".into()) // NOTE: it is really all the supported cases?
};
},
// base class, name directly available inside the
// module, can use this module's symbol resolver
2021-08-10 21:57:31 +08:00
ast::ExprKind::Name {id, ..} => {
let def_id = resolver.get_identifier_def(id);
2021-08-10 23:49:58 +08:00
// the definition list
ancestors.push(def_id);
},
2021-08-10 23:49:58 +08:00
// base class, things can be like `class A(BaseModule.Base)`, here we have to get the
// symbol resolver of the module `BaseModule`?
2021-08-10 21:57:31 +08:00
ast::ExprKind::Attribute {value, attr, ..} => {
2021-08-10 23:49:58 +08:00
if let ast::ExprKind::Name {id, ..} = &value.node {
if let Some(base_module_resolver) = resolver.get_module_resolver(id) {
let def_id = base_module_resolver.get_identifier_def(attr);
// the definition list
ancestors.push(def_id);
} else { return Err("unkown imported module".into()) }
} else { return Err("unkown imported module".into()) }
},
2021-08-10 23:49:58 +08:00
// `class Foo(ImportedModule.A[int, bool])`, A is a class with associated type variables
2021-08-10 21:57:31 +08:00
ast::ExprKind::Subscript {value, slice, ..} => {
2021-08-10 23:49:58 +08:00
unimplemented!()
},
_ => return Err("not supported".into())
}
}
// class method and field are analyzed by
// looking into the class body ast node
for stmt in body {
2021-08-10 23:49:58 +08:00
if let ast::StmtKind::FunctionDef {
name,
args,
body,
returns,
..
} = &stmt.node {
} else { }
// do nothing. we do not care about things like this?
// class A:
// a = 3
// b = [2, 3]
}
},
2021-08-10 23:49:58 +08:00
// top level function definition
ast::StmtKind::FunctionDef {
name,
args,
body,
returns,
..
} => {
unimplemented!()
}
_ => return Err("only expect function and class definitions to be submitted here to be analyzed".into())
}
}
2021-08-10 21:57:31 +08:00
}
Ok(())
}
2021-08-10 21:57:31 +08:00
}
2021-08-10 23:49:58 +08:00
pub fn parse_type_var<T>(input: &ast::Expr<T>, resolver: &dyn SymbolResolver) -> Result<Type, String> {
match &input.node {
ast::ExprKind::Name {id, ..} => {
resolver.get_symbol_type(id).ok_or_else(|| "unknown type variable identifer".to_string())
},
ast::ExprKind::Attribute {value, attr, ..} => {
if let ast::ExprKind::Name {id, ..} = &value.node {
let next_resolver = resolver.get_module_resolver(id).ok_or_else(|| "unknown imported module".to_string())?;
next_resolver.get_symbol_type(attr).ok_or_else(|| "unknown type variable identifer".to_string())
} else {
unimplemented!()
// recursively resolve attr thing, FIXME: new problem: how do we handle this?
// # A.py
// class A:
// T = TypeVar('T', int, bool)
// pass
// # B.py
// import A
// class B(Generic[A.A.T]):
// pass
}
},
_ => Err("not supported".into())
}
}