nac3/nac3core/src/top_level.rs

446 lines
19 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;
2021-08-11 14:37:26 +08:00
use inkwell::context::Context;
use parking_lot::{Mutex, 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>,
// symbol resolver of the module defined the class, none if it is built-in type
2021-08-11 15:18:21 +08:00
resolver: Option<Arc<Mutex<dyn SymbolResolver + Send>>>,
2021-08-03 13:38:27 +08:00
},
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)>,
// symbol resolver of the module defined the class
2021-08-11 15:18:21 +08:00
resolver: Option<Arc<Mutex<dyn SymbolResolver + Send>>>,
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 TopLevelContext {
2021-08-05 14:55:23 +08:00
pub definitions: Arc<RwLock<Vec<RwLock<TopLevelDef>>>>,
2021-08-11 14:37:26 +08:00
pub unifiers: Arc<RwLock<Vec<(SharedUnifier, PrimitiveStore)>>>,
pub conetexts: Arc<RwLock<Vec<Mutex<Context>>>>,
}
// like adding some info on top of the TopLevelDef for
// later parsing the class bases, method, and function sigatures
2021-08-11 15:18:21 +08:00
pub struct TopLevelDefInfo {
// the definition entry
def: TopLevelDef,
// the entry in the top_level unifier
ty: Type,
// the ast submitted by applications, primitives and
// class methods will have None value here
ast: Option<ast::Stmt<()>>,
}
2021-08-10 21:57:31 +08:00
pub struct TopLevelComposer {
// list of top level definitions and their info
pub definition_list: RwLock<Vec<TopLevelDefInfo>>,
// primitive store
pub primitives: PrimitiveStore,
// start as a primitive unifier, will add more top_level defs inside
pub unifier: Unifier,
// mangled class method name to def_id
pub class_method_to_def_id: HashMap<String, DefinitionId>,
}
impl TopLevelComposer {
fn name_mangling(mut class_name: String, method_name: &str) -> String {
class_name.push_str(method_name);
class_name
}
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().into(),
});
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().into(),
});
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().into(),
});
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().into(),
});
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().into(),
});
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
2021-08-12 10:50:01 +08:00
/// return a composer and things to make a "primitive" symbol resolver, so that the symbol
/// resolver can later figure out primitive type definitions when passed a primitive type name
pub fn new() -> (Vec<(String, DefinitionId, Type)>, Self) {
let primitives = Self::make_primitives();
// the def list including the entries of primitive info
let definition_list: Vec<TopLevelDefInfo> = vec![
TopLevelDefInfo {
def: Self::make_top_level_class_def(0, None),
ast: None,
2021-08-10 21:57:31 +08:00
ty: primitives.0.int32,
},
TopLevelDefInfo {
def: Self::make_top_level_class_def(1, None),
ast: None,
2021-08-10 21:57:31 +08:00
ty: primitives.0.int64,
},
TopLevelDefInfo {
def: Self::make_top_level_class_def(2, None),
ast: None,
2021-08-10 21:57:31 +08:00
ty: primitives.0.float,
},
TopLevelDefInfo {
def: Self::make_top_level_class_def(3, None),
ast: None,
2021-08-10 21:57:31 +08:00
ty: primitives.0.bool,
},
TopLevelDefInfo {
def: Self::make_top_level_class_def(4, None),
ast: None,
2021-08-10 21:57:31 +08:00
ty: primitives.0.none,
},
];
let composer = TopLevelComposer {
definition_list: definition_list.into(),
primitives: primitives.0,
unifier: primitives.1,
class_method_to_def_id: Default::default(),
};
(vec![
("int32".into(), DefinitionId(0), composer.primitives.int32),
("int64".into(), DefinitionId(1), composer.primitives.int64),
("float".into(), DefinitionId(2), composer.primitives.float),
("bool".into(), DefinitionId(3), composer.primitives.bool),
("none".into(), DefinitionId(4), composer.primitives.none),
], composer)
}
2021-08-10 23:49:58 +08:00
/// already include the definition_id of itself inside the ancestors vector
2021-08-11 15:18:21 +08:00
pub fn make_top_level_class_def(
index: usize,
resolver: Option<Arc<Mutex<dyn SymbolResolver + Send>>>,
) -> 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)],
2021-08-11 15:18:21 +08:00
resolver,
}
}
2021-08-11 15:18:21 +08:00
pub fn make_top_level_function_def(
name: String,
ty: Type,
resolver: Option<Arc<Mutex<dyn SymbolResolver + Send>>>,
) -> TopLevelDef {
TopLevelDef::Function {
name,
signature: ty,
instance_to_symbol: Default::default(),
2021-08-10 21:57:31 +08:00
instance_to_stmt: Default::default(),
2021-08-11 15:18:21 +08:00
resolver,
}
}
2021-08-10 21:57:31 +08:00
pub fn register_top_level(
&mut self,
ast: ast::Stmt<()>,
2021-08-11 15:18:21 +08:00
resolver: Option<Arc<Mutex<dyn SymbolResolver + Send>>>,
2021-08-12 10:50:01 +08:00
) -> Result<(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();
2021-08-12 10:50:01 +08:00
let mut def_list = self.definition_list.write();
let class_def_id = def_list.len();
2021-08-11 15:18:21 +08:00
// 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(),
});
// add the class to the definition list
def_list.push(TopLevelDefInfo {
def: Self::make_top_level_class_def(class_def_id, resolver.clone()),
// NOTE: Temporarily none here since function body need to be read later
ast: None,
ty,
});
2021-08-12 10:50:01 +08:00
2021-08-10 23:49:58 +08:00
// parse class def body and register class methods into the def list
2021-08-12 10:50:01 +08:00
// 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
// by using the field `class_method_to_def_id`
2021-08-10 23:49:58 +08:00
for b in body {
2021-08-11 15:18:21 +08:00
if let ast::StmtKind::FunctionDef { name, .. } = &b.node {
let fun_name = Self::name_mangling(class_name.clone(), name);
2021-08-12 10:50:01 +08:00
let def_id = def_list.len();
2021-08-10 23:49:58 +08:00
// add to unifier
2021-08-11 15:18:21 +08:00
let ty = self.unifier.add_ty(TypeEnum::TFunc(
crate::typecheck::typedef::FunSignature {
args: Default::default(),
ret: self.primitives.none,
vars: Default::default(),
},
));
2021-08-12 10:50:01 +08:00
2021-08-10 23:49:58 +08:00
// add to the definition list
2021-08-12 10:50:01 +08:00
def_list.push(TopLevelDefInfo {
def: Self::make_top_level_function_def(fun_name.clone(), ty, resolver.clone()),
2021-08-11 15:18:21 +08:00
ty,
2021-08-12 10:50:01 +08:00
// since it is inside the class def body statments, the ast is None
ast: None,
2021-08-11 15:18:21 +08:00
});
2021-08-12 10:50:01 +08:00
// class method, do not let the symbol manager manage it, use our own map
self.class_method_to_def_id.insert(fun_name, DefinitionId(def_id));
2021-08-11 15:18:21 +08:00
// 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__" {
// FIXME: how can this later be fetched?
2021-08-12 10:50:01 +08:00
def_list.push(TopLevelDefInfo {
def: TopLevelDef::Initializer { class_id: DefinitionId(class_def_id) },
2021-08-12 10:50:01 +08:00
// arbitary picked one for the constructor
ty: self.primitives.none,
// it is inside the class def body statments, so None
ast: None,
2021-08-11 15:18:21 +08:00
})
2021-08-10 23:49:58 +08:00
}
}
2021-08-10 23:49:58 +08:00
}
// move the ast to the entry of the class in the def_list
def_list.get_mut(class_def_id).unwrap().ast = Some(ast);
// return
2021-08-12 10:50:01 +08:00
Ok((class_name, DefinitionId(class_def_id), ty))
},
2021-08-10 21:57:31 +08:00
ast::StmtKind::FunctionDef { name, .. } => {
let fun_name = name.to_string();
2021-08-12 10:50:01 +08:00
// add to the unifier
2021-08-12 10:50:01 +08:00
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
2021-08-12 10:50:01 +08:00
let mut def_list = self.definition_list.write();
def_list.push(TopLevelDefInfo {
2021-08-10 21:57:31 +08:00
def: Self::make_top_level_function_def(
name.into(),
2021-08-11 15:18:21 +08:00
self.primitives.none,
resolver,
2021-08-10 21:57:31 +08:00
),
ast: Some(ast),
ty,
});
2021-08-12 10:50:01 +08:00
Ok((fun_name, DefinitionId(def_list.len() - 1), 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> {
2021-08-12 10:50:01 +08:00
for d in self.definition_list.write().iter_mut() {
// only analyze those with ast, and class_method(ast in class def)
if let Some(ast) = &d.ast {
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 (
2021-08-10 23:49:58 +08:00
ancestors,
fields,
methods,
type_vars,
2021-08-12 10:50:01 +08:00
resolver,
2021-08-10 23:49:58 +08:00
) = if let TopLevelDef::Class {
object_id: _,
2021-08-10 23:49:58 +08:00
ancestors,
fields,
methods,
type_vars,
2021-08-12 10:50:01 +08:00
resolver: Some(resolver)
} = &mut d.def {
(ancestors, fields, methods, type_vars, resolver.lock())
2021-08-10 23:49:58 +08:00
} else { unreachable!() };
2021-08-10 23:49:58 +08:00
// try to get mutable reference of the entry in the unification table, get the `TypeEnum`
let (params,
fields
) = if let TypeEnum::TObj {
2021-08-12 10:50:01 +08:00
params,
2021-08-10 23:49:58 +08:00
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 {
2021-08-12 10:50:01 +08:00
// typevars bounded to the class, only support things like `class A(Generic[T, V])`,
// things like `class A(Generic[T, V, ImportedModule.T])` is not supported
// i.e. only simple names are allowed in the subscript
2021-08-10 23:49:58 +08:00
// 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 {
2021-08-12 10:50:01 +08:00
// `class Foo(Generic[T, V, P]):`
2021-08-10 23:49:58 +08:00
ast::ExprKind::Tuple {elts, ..} => {
for e in elts {
2021-08-12 14:48:11 +08:00
// let ty_def_id = resolver.
2021-08-10 23:49:58 +08:00
}
},
// `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())?); FIXME:
2021-08-10 23:49:58 +08:00
unimplemented!()
},
2021-08-12 10:50:01 +08:00
_ => return Err("not supported, only simple names are allowed in the subscript".into())
2021-08-10 23:49:58 +08:00
};
},
2021-08-12 10:50:01 +08:00
/* // base class, name directly available inside the
2021-08-10 23:49:58 +08:00
// 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); FIXME:
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()) } FIXME:
2021-08-10 23:49:58 +08:00
} 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!()
2021-08-12 10:50:01 +08:00
}, */
// base class is possible in other cases, we parse for thr base class
_ => 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 { }
2021-08-10 23:49:58 +08:00
// 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!()
}
2021-08-12 10:50:01 +08:00
node => {
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
}