nac3/nac3core/src/top_level.rs

1150 lines
48 KiB
Rust
Raw Normal View History

use std::borrow::BorrowMut;
use std::ops::{Deref, DerefMut};
use std::{collections::HashMap, collections::HashSet, sync::Arc};
2021-08-03 13:38:27 +08:00
2021-08-23 02:52:54 +08:00
use self::top_level_type_annotation_info::*;
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};
2021-08-23 02:52:54 +08:00
use crate::symbol_resolver::SymbolResolver;
2021-08-23 11:13:45 +08:00
use crate::typecheck::typedef::{FunSignature, FuncArg};
use itertools::{izip, Itertools};
2021-08-11 14:37:26 +08:00
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
2021-08-23 02:52:54 +08:00
#[derive(PartialEq, Eq, PartialOrd, Ord, Clone, Copy)]
pub struct DefinitionId(pub usize);
2021-08-03 13:38:27 +08:00
2021-08-23 02:52:54 +08:00
pub mod top_level_type_annotation_info {
use super::*;
#[derive(Clone)]
pub enum TypeAnnotation {
PrimitiveKind(Type),
ConcretizedCustomClassKind {
id: DefinitionId,
// can not be type var, others are all fine
2021-08-23 11:13:45 +08:00
params: Vec<TypeAnnotation>,
2021-08-23 02:52:54 +08:00
},
// can only be ConcretizedCustomClassKind
VirtualKind(Box<TypeAnnotation>),
TypeVarKind(Type),
SelfTypeKind(DefinitionId),
}
pub fn parse_ast_to_type_annotation_kinds<T>(
resolver: &dyn SymbolResolver,
top_level_defs: &[Arc<RwLock<TopLevelDef>>],
unifier: &mut Unifier,
primitives: &PrimitiveStore,
expr: &ast::Expr<T>,
) -> Result<TypeAnnotation, String> {
let results = vec![
2021-08-23 11:13:45 +08:00
parse_ast_to_concrete_primitive_kind(
resolver,
top_level_defs,
unifier,
primitives,
expr,
),
parse_ast_to_concretized_custom_class_kind(
resolver,
top_level_defs,
unifier,
primitives,
expr,
),
2021-08-23 02:52:54 +08:00
parse_ast_to_type_variable_kind(resolver, top_level_defs, unifier, primitives, expr),
2021-08-23 11:13:45 +08:00
parse_ast_to_virtual_kind(resolver, top_level_defs, unifier, primitives, expr),
2021-08-23 02:52:54 +08:00
];
let results = results.iter().filter(|x| x.is_ok()).collect_vec();
if results.len() == 1 {
results[0].clone()
} else {
Err("cannot parsed the type annotation without ambiguity".into())
2021-08-23 02:52:54 +08:00
}
}
pub fn get_type_from_type_annotation_kinds(
top_level_defs: &[Arc<RwLock<TopLevelDef>>],
unifier: &mut Unifier,
primitives: &PrimitiveStore,
2021-08-23 11:13:45 +08:00
ann: &TypeAnnotation,
2021-08-23 02:52:54 +08:00
) -> Result<Type, String> {
match ann {
TypeAnnotation::ConcretizedCustomClassKind { id, params } => {
let class_def = top_level_defs[id.0].read();
2021-08-23 11:13:45 +08:00
if let TopLevelDef::Class { fields, methods, type_vars, .. } = &*class_def {
2021-08-23 02:52:54 +08:00
if type_vars.len() != params.len() {
Err(format!(
"unexpected number of type parameters: expected {} but got {}",
type_vars.len(),
params.len()
))
} else {
let param_ty = params
.iter()
2021-08-23 11:13:45 +08:00
.map(|x| {
get_type_from_type_annotation_kinds(
top_level_defs,
unifier,
primitives,
x,
)
})
2021-08-23 02:52:54 +08:00
.collect::<Result<Vec<_>, _>>()?;
let subst = type_vars
.iter()
.map(|x| {
if let TypeEnum::TVar { id, .. } = unifier.get_ty(*x).as_ref() {
*id
} else {
unreachable!()
}
})
.zip(param_ty.into_iter())
.collect::<HashMap<u32, Type>>();
let mut tobj_fields = methods
.iter()
.map(|(name, ty, _)| {
let subst_ty = unifier.subst(*ty, &subst).unwrap_or(*ty);
(name.clone(), subst_ty)
})
.collect::<HashMap<String, Type>>();
2021-08-23 11:13:45 +08:00
tobj_fields.extend(fields.iter().map(|(name, ty)| {
let subst_ty = unifier.subst(*ty, &subst).unwrap_or(*ty);
(name.clone(), subst_ty)
}));
2021-08-23 02:52:54 +08:00
Ok(unifier.add_ty(TypeEnum::TObj {
obj_id: *id,
fields: tobj_fields.into(),
2021-08-23 11:13:45 +08:00
params: subst.into(),
2021-08-23 02:52:54 +08:00
}))
}
} else {
unreachable!("should be class def here")
}
}
TypeAnnotation::SelfTypeKind(obj_id) => {
let class_def = top_level_defs[obj_id.0].read();
2021-08-23 11:13:45 +08:00
if let TopLevelDef::Class { fields, methods, type_vars, .. } = &*class_def {
2021-08-23 02:52:54 +08:00
let subst = type_vars
.iter()
.map(|x| {
if let TypeEnum::TVar { id, .. } = unifier.get_ty(*x).as_ref() {
(*id, *x)
} else {
unreachable!()
}
})
.collect::<HashMap<u32, Type>>();
let mut tobj_fields = methods
.iter()
2021-08-23 11:13:45 +08:00
.map(|(name, ty, _)| (name.clone(), *ty))
2021-08-23 02:52:54 +08:00
.collect::<HashMap<String, Type>>();
tobj_fields.extend(fields.clone().into_iter());
2021-08-23 02:52:54 +08:00
Ok(unifier.add_ty(TypeEnum::TObj {
obj_id: *obj_id,
fields: tobj_fields.into(),
2021-08-23 11:13:45 +08:00
params: subst.into(),
2021-08-23 02:52:54 +08:00
}))
} else {
unreachable!("should be class def here")
}
}
TypeAnnotation::PrimitiveKind(ty) => Ok(*ty),
TypeAnnotation::TypeVarKind(ty) => Ok(*ty),
TypeAnnotation::VirtualKind(ty) => {
let ty = get_type_from_type_annotation_kinds(
top_level_defs,
unifier,
primitives,
2021-08-23 11:13:45 +08:00
ty.as_ref(),
2021-08-23 02:52:54 +08:00
)?;
Ok(unifier.add_ty(TypeEnum::TVirtual { ty }))
}
}
}
fn parse_ast_to_concrete_primitive_kind<T>(
_resolver: &dyn SymbolResolver,
_top_level_defs: &[Arc<RwLock<TopLevelDef>>],
_unifier: &mut Unifier,
primitives: &PrimitiveStore,
expr: &ast::Expr<T>,
) -> Result<TypeAnnotation, String> {
match &expr.node {
ast::ExprKind::Name { id, .. } => match id.as_str() {
"int32" => Ok(TypeAnnotation::PrimitiveKind(primitives.int32)),
"int64" => Ok(TypeAnnotation::PrimitiveKind(primitives.int64)),
"float" => Ok(TypeAnnotation::PrimitiveKind(primitives.float)),
2021-08-23 11:13:45 +08:00
"bool" => Ok(TypeAnnotation::PrimitiveKind(primitives.bool)),
"None" => Ok(TypeAnnotation::PrimitiveKind(primitives.none)),
_ => Err("not primitive".into()),
},
2021-08-23 02:52:54 +08:00
2021-08-23 11:13:45 +08:00
_ => Err("not primitive".into()),
2021-08-23 02:52:54 +08:00
}
}
pub fn parse_ast_to_concretized_custom_class_kind<T>(
resolver: &dyn SymbolResolver,
top_level_defs: &[Arc<RwLock<TopLevelDef>>],
unifier: &mut Unifier,
primitives: &PrimitiveStore,
expr: &ast::Expr<T>,
) -> Result<TypeAnnotation, String> {
match &expr.node {
ast::ExprKind::Name { id, .. } => match id.as_str() {
2021-08-23 11:13:45 +08:00
"int32" | "int64" | "float" | "bool" | "None" => {
Err("expect custom class instead of primitives here".into())
}
2021-08-23 02:52:54 +08:00
x => {
let obj_id = resolver
.get_identifier_def(x)
.ok_or_else(|| "unknown class name".to_string())?;
let def = top_level_defs[obj_id.0].read();
if let TopLevelDef::Class { .. } = &*def {
2021-08-23 11:13:45 +08:00
Ok(TypeAnnotation::ConcretizedCustomClassKind {
id: obj_id,
params: vec![],
})
2021-08-23 02:52:54 +08:00
} else {
Err("function cannot be used as a type".into())
}
}
},
ast::ExprKind::Subscript { value, slice, .. } => {
if let ast::ExprKind::Name { id, .. } = &value.node {
2021-08-23 11:13:45 +08:00
if vec!["virtual", "Generic"].contains(&id.as_str()) {
return Err("keywords cannot be class name".into());
}
2021-08-23 02:52:54 +08:00
let obj_id = resolver
.get_identifier_def(id)
.ok_or_else(|| "unknown class name".to_string())?;
let def = top_level_defs[obj_id.0].read();
if let TopLevelDef::Class { .. } = &*def {
2021-08-23 11:13:45 +08:00
let param_type_infos =
2021-08-23 02:52:54 +08:00
if let ast::ExprKind::Tuple { elts, .. } = &slice.node {
elts.iter()
.map(|v| {
parse_ast_to_type_annotation_kinds(
resolver,
top_level_defs,
unifier,
primitives,
2021-08-23 11:13:45 +08:00
v,
2021-08-23 02:52:54 +08:00
)
})
.collect::<Result<Vec<_>, _>>()?
} else {
vec![parse_ast_to_type_annotation_kinds(
2021-08-23 11:13:45 +08:00
resolver,
top_level_defs,
unifier,
primitives,
slice,
2021-08-23 02:52:54 +08:00
)?]
};
2021-08-23 11:13:45 +08:00
if param_type_infos
.iter()
.any(|x| matches!(x, TypeAnnotation::TypeVarKind(..)))
{
return Err(
"cannot apply type variable to class generic parameters".into()
);
2021-08-23 02:52:54 +08:00
}
2021-08-23 11:13:45 +08:00
Ok(TypeAnnotation::ConcretizedCustomClassKind {
id: obj_id,
params: param_type_infos,
})
2021-08-23 02:52:54 +08:00
} else {
Err("function cannot be used as a type".into())
}
} else {
Err("unsupported expression type for class name".into())
2021-08-23 02:52:54 +08:00
}
2021-08-23 11:13:45 +08:00
}
2021-08-23 02:52:54 +08:00
_ => Err("unsupported expression type for concretized class".into()),
2021-08-23 02:52:54 +08:00
}
}
pub fn parse_ast_to_virtual_kind<T>(
resolver: &dyn SymbolResolver,
top_level_defs: &[Arc<RwLock<TopLevelDef>>],
unifier: &mut Unifier,
primitives: &PrimitiveStore,
expr: &ast::Expr<T>,
) -> Result<TypeAnnotation, String> {
match &expr.node {
ast::ExprKind::Subscript { value, slice, .. }
2021-08-23 11:13:45 +08:00
if { matches!(&value.node, ast::ExprKind::Name { id, .. } if id == "virtual") } =>
{
2021-08-23 02:52:54 +08:00
let def = parse_ast_to_concretized_custom_class_kind(
resolver,
top_level_defs,
unifier,
primitives,
2021-08-23 11:13:45 +08:00
slice.as_ref(),
2021-08-23 02:52:54 +08:00
)?;
if !matches!(def, TypeAnnotation::ConcretizedCustomClassKind { .. }) {
unreachable!("must be concretized custom class kind in the virtual")
2021-08-23 02:52:54 +08:00
}
Ok(TypeAnnotation::VirtualKind(def.into()))
}
2021-08-23 11:13:45 +08:00
_ => Err("virtual type annotation must be like `virtual[ .. ]`".into()),
2021-08-23 02:52:54 +08:00
}
}
pub fn parse_ast_to_type_variable_kind<T>(
resolver: &dyn SymbolResolver,
_top_level_defs: &[Arc<RwLock<TopLevelDef>>],
unifier: &mut Unifier,
primitives: &PrimitiveStore,
expr: &ast::Expr<T>,
) -> Result<TypeAnnotation, String> {
if let ast::ExprKind::Name { id, .. } = &expr.node {
let ty = resolver
.get_symbol_type(unifier, primitives, id)
.ok_or_else(|| "unknown type variable name".to_string())?;
if let TypeEnum::TVar { .. } = unifier.get_ty(ty).as_ref() {
Ok(TypeAnnotation::TypeVarKind(ty))
} else {
Err("not a type variable identifier".into())
}
2021-08-23 02:52:54 +08:00
} else {
Err("unsupported expression for type variable".into())
}
}
}
2021-08-03 13:38:27 +08:00
pub enum TopLevelDef {
Class {
// name for error messages and symbols
name: String,
2021-08-03 13:38:27 +08:00
// 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.
2021-08-23 02:52:54 +08:00
ancestors: Vec<TypeAnnotation>,
// symbol resolver of the module defined the class, none if it is built-in type
resolver: Option<Arc<Mutex<dyn SymbolResolver + Send + Sync>>>,
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
resolver: Option<Arc<Mutex<dyn SymbolResolver + Send + Sync>>>,
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 {
pub definitions: Arc<RwLock<Vec<Arc<RwLock<TopLevelDef>>>>>,
2021-08-11 14:37:26 +08:00
pub unifiers: Arc<RwLock<Vec<(SharedUnifier, PrimitiveStore)>>>,
}
pub struct TopLevelComposer {
2021-08-13 02:38:29 +08:00
// list of top level definitions, same as top level context
pub definition_ast_list: Vec<(Arc<RwLock<TopLevelDef>>, Option<ast::Stmt<()>>)>,
2021-08-13 02:38:29 +08:00
// start as a primitive unifier, will add more top_level defs inside
2021-08-17 14:01:18 +08:00
pub unifier: Unifier,
// primitive store
2021-08-23 02:52:54 +08:00
pub primitives_ty: PrimitiveStore,
// keyword list to prevent same custom def class name
2021-08-23 02:52:54 +08:00
pub keyword_list: Vec<String>,
}
impl TopLevelComposer {
2021-08-23 02:52:54 +08:00
pub fn make_top_level_context(self) -> TopLevelContext {
2021-08-13 02:38:29 +08:00
TopLevelContext {
2021-08-23 11:13:45 +08:00
definitions: RwLock::new(
self.definition_ast_list.into_iter().map(|(x, ..)| x).collect::<Vec<_>>(),
)
.into(),
2021-08-13 02:38:29 +08:00
// FIXME: all the big unifier or?
unifiers: Default::default(),
}
}
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
/// return a composer and things to make a "primitive" symbol resolver, so that the symbol
2021-08-12 10:50:01 +08:00
/// resolver can later figure out primitive type definitions when passed a primitive type name
2021-08-23 02:52:54 +08:00
// TODO: add list and tuples?
pub fn new() -> (Vec<(String, DefinitionId, Type)>, Self) {
let primitives = Self::make_primitives();
2021-08-13 02:38:29 +08:00
let top_level_def_list = vec![
Arc::new(RwLock::new(Self::make_top_level_class_def(0, None, "int32"))),
Arc::new(RwLock::new(Self::make_top_level_class_def(1, None, "int64"))),
Arc::new(RwLock::new(Self::make_top_level_class_def(2, None, "float"))),
Arc::new(RwLock::new(Self::make_top_level_class_def(3, None, "bool"))),
Arc::new(RwLock::new(Self::make_top_level_class_def(4, None, "none"))),
2021-08-13 02:38:29 +08:00
];
let ast_list: Vec<Option<ast::Stmt<()>>> = vec![None, None, None, None, None];
let composer = TopLevelComposer {
definition_ast_list: izip!(top_level_def_list, ast_list).collect_vec(),
2021-08-23 02:52:54 +08:00
primitives_ty: primitives.0,
unifier: primitives.1,
// class_method_to_def_id: Default::default(),
// to_be_analyzed_class: Default::default(),
2021-08-23 02:52:54 +08:00
keyword_list: vec![
"Generic".into(),
"virtual".into(),
"list".into(),
"tuple".into(),
"int32".into(),
"int64".into(),
"float".into(),
"bool".into(),
"none".into(),
"None".into(),
2021-08-23 11:13:45 +08:00
],
};
2021-08-13 02:38:29 +08:00
(
vec![
2021-08-23 02:52:54 +08:00
("int32".into(), DefinitionId(0), composer.primitives_ty.int32),
("int64".into(), DefinitionId(1), composer.primitives_ty.int64),
("float".into(), DefinitionId(2), composer.primitives_ty.float),
("bool".into(), DefinitionId(3), composer.primitives_ty.bool),
("none".into(), DefinitionId(4), composer.primitives_ty.none),
2021-08-13 02:38:29 +08:00
],
composer,
)
}
2021-08-10 23:49:58 +08:00
/// already include the definition_id of itself inside the ancestors vector
/// when first regitering, the type_vars, fields, methods, ancestors are invalid
2021-08-11 15:18:21 +08:00
pub fn make_top_level_class_def(
index: usize,
resolver: Option<Arc<Mutex<dyn SymbolResolver + Send + Sync>>>,
name: &str,
2021-08-11 15:18:21 +08:00
) -> TopLevelDef {
TopLevelDef::Class {
name: name.to_string(),
object_id: DefinitionId(index),
type_vars: Default::default(),
fields: Default::default(),
methods: Default::default(),
2021-08-23 02:52:54 +08:00
ancestors: vec![TypeAnnotation::SelfTypeKind(DefinitionId(index))],
2021-08-11 15:18:21 +08:00
resolver,
}
}
/// when first registering, the type is a invalid value
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 + Sync>>>,
2021-08-11 15:18:21 +08:00
) -> 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,
}
}
2021-08-23 02:52:54 +08:00
fn make_class_method_name(mut class_name: String, method_name: &str) -> String {
class_name.push_str(method_name);
class_name
}
fn extract_def_list(&self) -> Vec<Arc<RwLock<TopLevelDef>>> {
2021-08-23 11:13:45 +08:00
self.definition_ast_list.iter().map(|(def, ..)| def.clone()).collect_vec()
}
/// step 0, register, just remeber the names of top level classes/function
2021-08-10 21:57:31 +08:00
pub fn register_top_level(
&mut self,
ast: ast::Stmt<()>,
resolver: Option<Arc<Mutex<dyn SymbolResolver + Send + Sync>>>,
) -> Result<(String, DefinitionId), String> {
let mut defined_class_name: HashSet<String> = HashSet::new();
let mut defined_class_method_name: HashSet<String> = HashSet::new();
let mut defined_function_name: HashSet<String> = HashSet::new();
match &ast.node {
2021-08-10 21:57:31 +08:00
ast::StmtKind::ClassDef { name, body, .. } => {
2021-08-23 02:52:54 +08:00
if self.keyword_list.contains(name) {
2021-08-23 11:13:45 +08:00
return Err("cannot use keyword as a class name".into());
2021-08-23 02:52:54 +08:00
}
if !defined_class_name.insert(name.clone()) {
return Err("duplicate definition of class".into());
}
2021-08-23 11:13:45 +08:00
let class_name = name.to_string();
let class_def_id = self.definition_ast_list.len();
2021-08-11 15:18:21 +08:00
2021-08-13 02:38:29 +08:00
// since later when registering class method, ast will still be used,
// here push None temporarly, later will move the ast inside
let mut class_def_ast = (
2021-08-18 16:33:50 +08:00
Arc::new(RwLock::new(Self::make_top_level_class_def(
class_def_id,
resolver.clone(),
name.as_str(),
2021-08-18 16:33:50 +08:00
))),
None,
);
2021-08-13 13:55:44 +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
2021-08-18 16:33:50 +08:00
let mut class_method_name_def_ids: Vec<(
2021-08-23 02:52:54 +08:00
// the simple method name without class name
2021-08-18 16:33:50 +08:00
String,
2021-08-23 02:52:54 +08:00
// in this top level def, method name is prefixed with the class name
2021-08-18 16:33:50 +08:00
Arc<RwLock<TopLevelDef>>,
DefinitionId,
2021-08-23 11:13:45 +08:00
Type,
2021-08-18 16:33:50 +08:00
)> = Vec::new();
let mut class_method_index_offset = 0;
2021-08-10 23:49:58 +08:00
for b in body {
if let ast::StmtKind::FunctionDef { name: method_name, .. } = &b.node {
let global_class_method_name =
Self::make_class_method_name(class_name.clone(), method_name);
if !defined_class_method_name.insert(global_class_method_name.clone()) {
return Err("duplicate class method definition".into());
}
let method_def_id = self.definition_ast_list.len() + {
class_method_index_offset += 1;
class_method_index_offset
};
// dummy method define here
let dummy_method_type = self.unifier.get_fresh_var();
class_method_name_def_ids.push((
method_name.clone(),
RwLock::new(Self::make_top_level_function_def(
global_class_method_name,
// later unify with parsed type
dummy_method_type.0,
resolver.clone(),
2021-08-18 16:33:50 +08:00
))
.into(),
DefinitionId(method_def_id),
2021-08-23 11:13:45 +08:00
dummy_method_type.0,
));
} else {
// do nothing
2021-08-23 11:13:45 +08:00
continue;
}
2021-08-10 23:49:58 +08:00
}
2021-08-23 11:13:45 +08:00
2021-08-13 02:38:29 +08:00
// move the ast to the entry of the class in the ast_list
class_def_ast.1 = Some(ast);
// get the methods into the class_def
2021-08-23 02:52:54 +08:00
for (name, _, id, ty) in &class_method_name_def_ids {
let mut class_def = class_def_ast.0.write();
if let TopLevelDef::Class { methods, .. } = class_def.deref_mut() {
methods.push((name.clone(), *ty, *id))
2021-08-23 11:13:45 +08:00
} else {
unreachable!()
}
}
// now class_def_ast and class_method_def_ast_ids are ok, put them into actual def list in correct order
self.definition_ast_list.push(class_def_ast);
for (_, def, ..) in class_method_name_def_ids {
self.definition_ast_list.push((def, None));
}
// put the constructor into the def_list
self.definition_ast_list.push((
2021-08-18 16:33:50 +08:00
RwLock::new(TopLevelDef::Initializer { class_id: DefinitionId(class_def_id) })
.into(),
None,
));
Ok((class_name, DefinitionId(class_def_id)))
2021-08-13 02:38:29 +08:00
}
2021-08-10 21:57:31 +08:00
ast::StmtKind::FunctionDef { name, .. } => {
let fun_name = name.to_string();
if !defined_function_name.insert(name.to_string()) {
return Err("duplicate top level function define".into());
}
// add to the definition list
self.definition_ast_list.push((
2021-08-18 16:33:50 +08:00
RwLock::new(Self::make_top_level_function_def(
name.into(),
// unify with correct type later
self.unifier.get_fresh_var().0,
2021-08-18 16:33:50 +08:00
resolver,
))
.into(),
Some(ast),
));
2021-08-13 02:38:29 +08:00
// return
Ok((fun_name, DefinitionId(self.definition_ast_list.len() - 1)))
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()),
}
}
/// step 1, analyze the type vars associated with top level class
fn analyze_top_level_class_type_var(&mut self) -> Result<(), String> {
2021-08-23 02:52:54 +08:00
let def_list = &self.definition_ast_list;
let temp_def_list = self.extract_def_list();
let unifier = self.unifier.borrow_mut();
let primitives_store = &self.primitives_ty;
for (class_def, class_ast) in def_list {
// only deal with class def here
let mut class_def = class_def.write();
let (class_bases_ast, class_def_type_vars, class_resolver) = {
if let TopLevelDef::Class { type_vars, resolver, .. } = class_def.deref_mut() {
2021-08-16 20:17:08 +08:00
if let Some(ast::Located {
node: ast::StmtKind::ClassDef { bases, .. }, ..
}) = class_ast
{
(bases, type_vars, resolver)
2021-08-16 20:17:08 +08:00
} else {
unreachable!("must be both class")
}
} else {
continue;
}
};
let class_resolver = class_resolver.as_ref().unwrap().lock();
let class_resolver = class_resolver.deref();
2021-08-23 02:52:54 +08:00
let mut is_generic = false;
for b in class_bases_ast {
match &b.node {
// analyze 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
// should update the TopLevelDef::Class.typevars and the TypeEnum::TObj.params
ast::ExprKind::Subscript { value, slice, .. }
if {
2021-08-23 11:13:45 +08:00
matches!(&value.node, ast::ExprKind::Name { id, .. } if id == "Generic")
} =>
{
if !is_generic {
is_generic = true;
} else {
return Err("Only single Generic[...] can be in bases".into());
}
2021-08-23 02:52:54 +08:00
let mut type_var_list: Vec<&ast::Expr<()>> = vec![];
// if `class A(Generic[T, V, G])`
if let ast::ExprKind::Tuple { elts, .. } = &slice.node {
type_var_list.extend(elts.iter());
// `class A(Generic[T])`
} else {
type_var_list.push(slice.deref());
}
2021-08-16 20:17:08 +08:00
// parse the type vars
let type_vars = type_var_list
2021-08-23 02:52:54 +08:00
.into_iter()
.map(|e| {
class_resolver.parse_type_annotation(
&temp_def_list,
unifier,
primitives_store,
2021-08-23 11:13:45 +08:00
e,
2021-08-23 02:52:54 +08:00
)
})
.collect::<Result<Vec<_>, _>>()?;
// check if all are unique type vars
let mut occured_type_var_id: HashSet<u32> = HashSet::new();
let all_unique_type_var = type_vars.iter().all(|x| {
2021-08-23 02:52:54 +08:00
let ty = unifier.get_ty(*x);
if let TypeEnum::TVar { id, .. } = ty.as_ref() {
occured_type_var_id.insert(*id)
} else {
false
}
});
2021-08-23 02:52:54 +08:00
// NOTE: create a copy of all type vars for the type vars associated with class
let type_vars = type_vars
.into_iter()
.map(|x| {
let range = unifier.get_ty(x);
if let TypeEnum::TVar { range, .. } = range.as_ref() {
let range = &*range.borrow();
let range = range.as_slice();
unifier.get_fresh_var_with_range(range).0
} else {
unreachable!("must be type var here");
}
})
.collect_vec();
2021-08-23 11:13:45 +08:00
if !all_unique_type_var {
return Err("expect unique type variables".into());
}
2021-08-23 11:13:45 +08:00
// add to TopLevelDef
class_def_type_vars.extend(type_vars);
2021-08-12 10:50:01 +08:00
}
2021-08-16 20:17:08 +08:00
// if others, do nothing in this function
2021-08-16 20:17:08 +08:00
_ => continue,
}
}
2021-08-16 20:17:08 +08:00
}
Ok(())
}
2021-08-23 02:52:54 +08:00
/// step 2, base classes.
fn analyze_top_level_class_bases(&mut self) -> Result<(), String> {
2021-08-23 02:52:54 +08:00
let temp_def_list = self.extract_def_list();
for (class_def, class_ast) in self.definition_ast_list.iter_mut() {
let mut class_def = class_def.write();
2021-08-16 20:17:08 +08:00
let (class_bases, class_ancestors, class_resolver) = {
if let TopLevelDef::Class { ancestors, resolver, .. } = class_def.deref_mut() {
2021-08-16 20:17:08 +08:00
if let Some(ast::Located {
node: ast::StmtKind::ClassDef { bases, .. }, ..
}) = class_ast
{
(bases, ancestors, resolver)
2021-08-16 20:17:08 +08:00
} else {
unreachable!("must be both class")
}
} else {
continue;
}
};
let class_resolver = class_resolver.as_ref().unwrap().lock();
let class_resolver = class_resolver.deref();
let mut has_base = false;
for b in class_bases {
// type vars have already been handled, so skip on `Generic[...]`
if matches!(
&b.node,
ast::ExprKind::Subscript { value, .. }
if matches!(
&value.node,
ast::ExprKind::Name { id, .. } if id == "Generic"
)
2021-08-23 11:13:45 +08:00
) {
continue;
}
if has_base {
return Err("a class def can only have at most one base class \
2021-08-23 11:13:45 +08:00
declaration and one generic declaration"
.into());
}
2021-08-23 02:52:54 +08:00
has_base = true;
2021-08-23 02:52:54 +08:00
let base_ty = parse_ast_to_type_annotation_kinds(
class_resolver,
&temp_def_list,
self.unifier.borrow_mut(),
2021-08-23 02:52:54 +08:00
&self.primitives_ty,
2021-08-23 11:13:45 +08:00
b,
)?;
if let TypeAnnotation::ConcretizedCustomClassKind { .. } = &base_ty {
2021-08-23 02:52:54 +08:00
// TODO: check to prevent cyclic base class
class_ancestors.push(base_ty);
} else {
2021-08-23 11:13:45 +08:00
return Err(
"class base declaration can only be concretized custom class".into()
);
2021-08-23 02:52:54 +08:00
}
}
2021-08-16 20:17:08 +08:00
}
Ok(())
}
/// step 3, class fields and methods
fn analyze_top_level_class_fields_methods(&mut self) -> Result<(), String> {
2021-08-23 02:52:54 +08:00
let temp_def_list = self.extract_def_list();
let unifier = self.unifier.borrow_mut();
let primitives = &self.primitives_ty;
let def_ast_list = &self.definition_ast_list;
2021-08-23 11:13:45 +08:00
2021-08-23 02:52:54 +08:00
let mut type_var_to_concrete_def: HashMap<Type, TypeAnnotation> = HashMap::new();
for (class_def, class_ast) in def_ast_list {
Self::analyze_single_class(
class_def.clone(),
&class_ast.as_ref().unwrap().node,
&temp_def_list,
unifier,
primitives,
2021-08-23 11:13:45 +08:00
&mut type_var_to_concrete_def,
2021-08-23 02:52:54 +08:00
)?
}
2021-08-23 02:52:54 +08:00
// base class methods add and check
2021-08-23 11:13:45 +08:00
// TODO:
2021-08-23 02:52:54 +08:00
// unification of previously assigned typevar
for (ty, def) in type_var_to_concrete_def {
2021-08-23 11:13:45 +08:00
let target_ty =
get_type_from_type_annotation_kinds(&temp_def_list, unifier, primitives, &def)?;
2021-08-23 02:52:54 +08:00
unifier.unify(ty, target_ty)?;
}
2021-08-23 11:13:45 +08:00
2021-08-23 02:52:54 +08:00
Ok(())
}
2021-08-23 02:52:54 +08:00
/// step 4, after class methods are done
fn analyze_top_level_function(&mut self) -> Result<(), String> {
let def_list = &self.definition_ast_list;
let temp_def_list = self.extract_def_list();
let unifier = self.unifier.borrow_mut();
let primitives_store = &self.primitives_ty;
for (function_def, function_ast) in def_list {
let function_def = function_def.read();
let function_def = function_def.deref();
let function_ast = if let Some(function_ast) = function_ast {
function_ast
} else {
continue;
};
if let TopLevelDef::Function { signature: dummy_ty, resolver, .. } = function_def {
if let ast::StmtKind::FunctionDef { args, returns, .. } = &function_ast.node {
let resolver = resolver.as_ref();
let resolver = resolver.unwrap();
let resolver = resolver.deref().lock();
let function_resolver = resolver.deref();
2021-08-23 11:13:45 +08:00
let mut defined_paramter_name: HashSet<String> = HashSet::new();
let have_unique_fuction_parameter_name =
args.args.iter().all(|x| defined_paramter_name.insert(x.node.arg.clone()));
if !have_unique_fuction_parameter_name {
return Err("top level function have duplicate parameter name".into());
}
2021-08-23 02:52:54 +08:00
let arg_types = {
2021-08-23 11:13:45 +08:00
args.args
2021-08-23 02:52:54 +08:00
.iter()
.map(|x| -> Result<FuncArg, String> {
let annotation = x
.node
.annotation
.as_ref()
2021-08-23 11:13:45 +08:00
.ok_or_else(|| {
"function parameter type annotation needed".to_string()
})?
2021-08-18 16:33:50 +08:00
.as_ref();
2021-08-23 02:52:54 +08:00
Ok(FuncArg {
name: x.node.arg.clone(),
ty: function_resolver.parse_type_annotation(
temp_def_list.as_slice(),
unifier,
primitives_store,
2021-08-23 11:13:45 +08:00
annotation,
2021-08-23 02:52:54 +08:00
)?,
// TODO: function type var
2021-08-23 11:13:45 +08:00
default_value: Default::default(),
2021-08-23 02:52:54 +08:00
})
})
.collect::<Result<Vec<_>, _>>()?
};
2021-08-18 16:33:50 +08:00
2021-08-23 02:52:54 +08:00
let return_ty = {
let return_annotation = returns
.as_ref()
.ok_or_else(|| "function return type needed".to_string())?
.as_ref();
function_resolver.parse_type_annotation(
temp_def_list.as_slice(),
unifier,
primitives_store,
2021-08-23 11:13:45 +08:00
return_annotation,
2021-08-23 02:52:54 +08:00
)?
};
2021-08-23 11:13:45 +08:00
let function_ty = unifier.add_ty(TypeEnum::TFunc(
FunSignature {
args: arg_types,
ret: return_ty,
// TODO: handle var map
vars: Default::default(),
}
.into(),
));
2021-08-23 02:52:54 +08:00
unifier.unify(*dummy_ty, function_ty)?;
} else {
unreachable!("must be both function");
}
} else {
continue;
}
2021-08-23 11:13:45 +08:00
}
2021-08-23 02:52:54 +08:00
Ok(())
}
/// step 5, field instantiation?
fn analyze_top_level_field_instantiation(&mut self) -> Result<(), String> {
// TODO:
unimplemented!()
}
fn analyze_single_class(
class_def: Arc<RwLock<TopLevelDef>>,
class_ast: &ast::StmtKind<()>,
temp_def_list: &[Arc<RwLock<TopLevelDef>>],
unifier: &mut Unifier,
primitives: &PrimitiveStore,
2021-08-23 11:13:45 +08:00
type_var_to_concrete_def: &mut HashMap<Type, TypeAnnotation>,
2021-08-23 02:52:54 +08:00
) -> Result<(), String> {
let mut class_def = class_def.write();
let (
class_id,
2021-08-23 02:52:54 +08:00
_class_name,
_class_bases_ast,
class_body_ast,
_class_ancestor_def,
class_fields_def,
class_methods_def,
_class_type_vars_def,
class_resolver,
) = if let TopLevelDef::Class {
object_id,
ancestors,
fields,
methods,
resolver,
type_vars,
..
2021-08-23 11:13:45 +08:00
} = class_def.deref_mut()
{
2021-08-23 02:52:54 +08:00
if let ast::StmtKind::ClassDef { name, bases, body, .. } = &class_ast {
2021-08-23 11:13:45 +08:00
(
object_id,
name.clone(),
bases,
body,
ancestors,
fields,
methods,
type_vars,
resolver,
)
2021-08-23 02:52:54 +08:00
} else {
unreachable!("here must be class def ast");
}
} else {
unreachable!("here must be class def ast");
};
let class_resolver = class_resolver.as_ref().unwrap();
let mut class_resolver = class_resolver.lock();
let class_resolver = class_resolver.deref_mut();
2021-08-18 16:33:50 +08:00
2021-08-23 02:52:54 +08:00
for b in class_body_ast {
if let ast::StmtKind::FunctionDef { args, returns, name, body, .. } = &b.node {
2021-08-23 11:13:45 +08:00
let (method_dummy_ty, ..) =
Self::get_class_method_def_info(class_methods_def, name)?;
let mut defined_paramter_name: HashSet<String> = HashSet::new();
let have_unique_fuction_parameter_name =
args.args.iter().all(|x| defined_paramter_name.insert(x.node.arg.clone()));
if !have_unique_fuction_parameter_name {
return Err("class method have duplicate parameter name".into());
}
2021-08-23 02:52:54 +08:00
// TODO: handle parameter with same name
let arg_type: Vec<FuncArg> = {
let mut result = Vec::new();
2021-08-23 11:13:45 +08:00
for x in &args.args {
2021-08-23 02:52:54 +08:00
let name = x.node.arg.clone();
if name != "self" {
let type_ann = {
let annotation_expr = x
.node
.annotation
.as_ref()
.ok_or_else(|| "type annotation needed".to_string())?
.as_ref();
parse_ast_to_type_annotation_kinds(
class_resolver,
temp_def_list,
unifier,
primitives,
annotation_expr,
)?
};
if let TypeAnnotation::TypeVarKind(_ty) = &type_ann {
// TODO: need to handle to different type vars that are
// asscosiated with the class and that are not
}
let dummy_func_arg = FuncArg {
name,
ty: unifier.get_fresh_var().0,
// TODO: symbol default value?
default_value: None,
};
// push the dummy type and the type annotation
// into the list for later unification
type_var_to_concrete_def.insert(dummy_func_arg.ty, type_ann.clone());
result.push(dummy_func_arg)
} else {
// if the parameter name is self
// python does not seem to enforce the name
// representing the self class object to be
// `self`, but we do it here
let dummy_func_arg = FuncArg {
name: "self".into(),
ty: unifier.get_fresh_var().0,
default_value: None,
};
type_var_to_concrete_def
.insert(dummy_func_arg.ty, TypeAnnotation::SelfTypeKind(*class_id));
result.push(dummy_func_arg);
}
2021-08-23 02:52:54 +08:00
}
result
};
let ret_type = {
if name != "__init__" {
let result = returns
.as_ref()
.ok_or_else(|| "method return type annotation needed".to_string())?
.as_ref();
let annotation = parse_ast_to_type_annotation_kinds(
class_resolver,
temp_def_list,
unifier,
primitives,
result,
)?;
let dummy_return_type = unifier.get_fresh_var().0;
type_var_to_concrete_def.insert(dummy_return_type, annotation.clone());
dummy_return_type
} else {
// if is the "__init__" function, the return type is self
let dummy_return_type = unifier.get_fresh_var().0;
type_var_to_concrete_def
.insert(dummy_return_type, TypeAnnotation::SelfTypeKind(*class_id));
dummy_return_type
}
2021-08-23 02:52:54 +08:00
};
// TODO: handle var map, to create a new copy of type var
// while tracking the type var associated with class
let method_var_map: HashMap<u32, Type> = HashMap::new();
2021-08-23 11:13:45 +08:00
let method_type = unifier.add_ty(TypeEnum::TFunc(
FunSignature { args: arg_type, ret: ret_type, vars: method_var_map }.into(),
));
2021-08-23 02:52:54 +08:00
// unify now since function type is not in type annotation define
// which is fine since type within method_type will be subst later
unifier.unify(method_dummy_ty, method_type)?;
2021-08-18 16:33:50 +08:00
2021-08-23 02:52:54 +08:00
// class fields
if name == "__init__" {
for b in body {
let mut defined_fields: HashSet<String> = HashSet::new();
// TODO: check the type of value, field instantiation check
2021-08-23 11:13:45 +08:00
if let ast::StmtKind::AnnAssign { annotation, target, value: _, .. } =
&b.node
{
2021-08-23 02:52:54 +08:00
if let ast::ExprKind::Attribute { value, attr, .. } = &target.node {
2021-08-23 11:13:45 +08:00
if matches!(&value.node, ast::ExprKind::Name { id, .. } if id == "self")
{
2021-08-23 02:52:54 +08:00
if defined_fields.insert(attr.to_string()) {
let dummy_field_type = unifier.get_fresh_var().0;
class_fields_def.push((attr.to_string(), dummy_field_type));
let annotation = parse_ast_to_type_annotation_kinds(
class_resolver,
&temp_def_list,
unifier,
primitives,
2021-08-23 11:13:45 +08:00
annotation.as_ref(),
2021-08-23 02:52:54 +08:00
)?;
2021-08-23 11:13:45 +08:00
type_var_to_concrete_def
.insert(dummy_field_type, annotation);
2021-08-18 16:33:50 +08:00
} else {
2021-08-23 02:52:54 +08:00
return Err("same class fields defined twice".into());
2021-08-18 16:33:50 +08:00
}
}
}
}
}
}
2021-08-18 16:33:50 +08:00
} else {
2021-08-23 02:52:54 +08:00
continue;
}
2021-08-23 11:13:45 +08:00
}
Ok(())
}
2021-08-16 20:17:08 +08:00
2021-08-23 02:52:54 +08:00
fn get_class_method_def_info(
class_methods_def: &[(String, Type, DefinitionId)],
2021-08-23 11:13:45 +08:00
method_name: &str,
2021-08-23 02:52:54 +08:00
) -> Result<(Type, DefinitionId), String> {
for (name, ty, def_id) in class_methods_def {
if name == method_name {
return Ok((*ty, *def_id));
}
}
2021-08-23 02:52:54 +08:00
Err(format!("no method {} in the current class", method_name))
}
2021-08-23 11:13:45 +08:00
}