forked from M-Labs/nac3
1
0
Fork 0
nac3/nac3core/src/toplevel/composer.rs

1813 lines
84 KiB
Rust
Raw Normal View History

use std::cell::RefCell;
use rustpython_parser::ast::fold::Fold;
2021-10-23 23:53:36 +08:00
use crate::{
symbol_resolver::SymbolValue,
typecheck::type_inferencer::{FunctionData, Inferencer},
};
use super::*;
type DefAst = (Arc<RwLock<TopLevelDef>>, Option<ast::Stmt<()>>);
pub struct TopLevelComposer {
// list of top level definitions, same as top level context
pub definition_ast_list: Vec<DefAst>,
// start as a primitive unifier, will add more top_level defs inside
pub unifier: Unifier,
// primitive store
pub primitives_ty: PrimitiveStore,
// keyword list to prevent same user-defined name
2021-09-22 17:19:27 +08:00
pub keyword_list: HashSet<StrRef>,
// to prevent duplicate definition
pub defined_names: HashSet<String>,
// get the class def id of a class method
pub method_class: HashMap<DefinitionId, DefinitionId>,
// number of built-in function and classes in the definition list, later skip
pub built_in_num: usize,
}
impl Default for TopLevelComposer {
fn default() -> Self {
Self::new(vec![]).0
}
}
impl TopLevelComposer {
/// 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
2021-09-19 22:54:06 +08:00
pub fn new(
builtins: Vec<(StrRef, FunSignature, Arc<GenCall>)>,
2021-09-22 17:19:27 +08:00
) -> (Self, HashMap<StrRef, DefinitionId>, HashMap<StrRef, Type>) {
let mut primitives = Self::make_primitives();
let int32 = primitives.0.int32;
let int64 = primitives.0.int64;
let float = primitives.0.float;
2021-10-10 15:01:06 +08:00
let boolean = primitives.0.bool;
2021-10-23 23:53:36 +08:00
let range = primitives.0.range;
2021-11-02 23:22:37 +08:00
let string = primitives.0.str;
2021-10-10 15:01:06 +08:00
let num_ty = primitives.1.get_fresh_var_with_range(&[int32, int64, float, boolean]);
let var_map: HashMap<_, _> = vec![(num_ty.1, num_ty.0)].into_iter().collect();
let mut definition_ast_list = {
let top_level_def_list = vec![
2021-09-22 17:19:27 +08:00
Arc::new(RwLock::new(Self::make_top_level_class_def(
0,
None,
"int32".into(),
None,
))),
Arc::new(RwLock::new(Self::make_top_level_class_def(
1,
None,
"int64".into(),
None,
))),
Arc::new(RwLock::new(Self::make_top_level_class_def(
2,
None,
"float".into(),
None,
))),
Arc::new(RwLock::new(Self::make_top_level_class_def(3, None, "bool".into(), None))),
Arc::new(RwLock::new(Self::make_top_level_class_def(4, None, "none".into(), None))),
2021-10-23 23:53:36 +08:00
Arc::new(RwLock::new(Self::make_top_level_class_def(
5,
None,
"range".into(),
None,
))),
2021-11-02 23:22:37 +08:00
Arc::new(RwLock::new(Self::make_top_level_class_def(
6,
None,
"str".into(),
None,
))),
Arc::new(RwLock::new(TopLevelDef::Function {
name: "int32".into(),
simple_name: "int32".into(),
signature: primitives.1.add_ty(TypeEnum::TFunc(RefCell::new(FunSignature {
args: vec![FuncArg { name: "_".into(), ty: num_ty.0, default_value: None }],
ret: int32,
vars: var_map.clone(),
}))),
var_id: Default::default(),
instance_to_symbol: Default::default(),
instance_to_stmt: Default::default(),
resolver: None,
codegen_callback: Some(Arc::new(GenCall::new(Box::new(
|ctx, _, fun, args| {
let int32 = ctx.primitives.int32;
let int64 = ctx.primitives.int64;
let float = ctx.primitives.float;
2021-10-10 15:01:06 +08:00
let boolean = ctx.primitives.bool;
let arg_ty = fun.0.args[0].ty;
let arg = args[0].1;
2021-10-10 15:01:06 +08:00
if ctx.unifier.unioned(arg_ty, boolean) {
Some(
ctx.builder
.build_int_s_extend(
arg.into_int_value(),
ctx.ctx.i32_type(),
"sext",
)
.into(),
)
} else if ctx.unifier.unioned(arg_ty, int32) {
Some(arg)
} else if ctx.unifier.unioned(arg_ty, int64) {
Some(
ctx.builder
.build_int_truncate(
arg.into_int_value(),
ctx.ctx.i32_type(),
"trunc",
)
.into(),
)
} else if ctx.unifier.unioned(arg_ty, float) {
let val = ctx
.builder
.build_float_to_signed_int(
arg.into_float_value(),
ctx.ctx.i32_type(),
"fptosi",
)
.into();
Some(val)
} else {
unreachable!()
}
},
)))),
})),
Arc::new(RwLock::new(TopLevelDef::Function {
name: "int64".into(),
simple_name: "int64".into(),
signature: primitives.1.add_ty(TypeEnum::TFunc(RefCell::new(FunSignature {
args: vec![FuncArg { name: "_".into(), ty: num_ty.0, default_value: None }],
ret: int64,
vars: var_map.clone(),
}))),
var_id: Default::default(),
instance_to_symbol: Default::default(),
instance_to_stmt: Default::default(),
resolver: None,
codegen_callback: Some(Arc::new(GenCall::new(Box::new(
|ctx, _, fun, args| {
let int32 = ctx.primitives.int32;
let int64 = ctx.primitives.int64;
let float = ctx.primitives.float;
2021-10-10 15:01:06 +08:00
let boolean = ctx.primitives.bool;
let arg_ty = fun.0.args[0].ty;
let arg = args[0].1;
2021-10-10 15:01:06 +08:00
if ctx.unifier.unioned(arg_ty, boolean)
|| ctx.unifier.unioned(arg_ty, int32)
{
Some(
ctx.builder
.build_int_s_extend(
arg.into_int_value(),
ctx.ctx.i64_type(),
"sext",
)
.into(),
)
} else if ctx.unifier.unioned(arg_ty, int64) {
Some(arg)
} else if ctx.unifier.unioned(arg_ty, float) {
let val = ctx
.builder
.build_float_to_signed_int(
arg.into_float_value(),
ctx.ctx.i64_type(),
"fptosi",
)
.into();
Some(val)
} else {
unreachable!()
}
},
)))),
})),
Arc::new(RwLock::new(TopLevelDef::Function {
name: "float".into(),
simple_name: "float".into(),
signature: primitives.1.add_ty(TypeEnum::TFunc(RefCell::new(FunSignature {
args: vec![FuncArg { name: "_".into(), ty: num_ty.0, default_value: None }],
ret: float,
vars: var_map,
}))),
var_id: Default::default(),
instance_to_symbol: Default::default(),
instance_to_stmt: Default::default(),
resolver: None,
codegen_callback: Some(Arc::new(GenCall::new(Box::new(
|ctx, _, fun, args| {
let int32 = ctx.primitives.int32;
let int64 = ctx.primitives.int64;
2021-10-10 15:01:06 +08:00
let boolean = ctx.primitives.bool;
let float = ctx.primitives.float;
let arg_ty = fun.0.args[0].ty;
let arg = args[0].1;
2021-10-10 15:01:06 +08:00
if ctx.unifier.unioned(arg_ty, boolean)
|| ctx.unifier.unioned(arg_ty, int32)
|| ctx.unifier.unioned(arg_ty, int64)
{
let arg = args[0].1.into_int_value();
let val = ctx
.builder
.build_signed_int_to_float(arg, ctx.ctx.f64_type(), "sitofp")
.into();
Some(val)
} else if ctx.unifier.unioned(arg_ty, float) {
Some(arg)
} else {
unreachable!()
}
},
)))),
})),
Arc::new(RwLock::new(TopLevelDef::Function {
name: "round".into(),
simple_name: "round".into(),
signature: primitives.1.add_ty(TypeEnum::TFunc(RefCell::new(FunSignature {
args: vec![FuncArg { name: "_".into(), ty: float, default_value: None }],
ret: int32,
vars: Default::default(),
}))),
var_id: Default::default(),
instance_to_symbol: Default::default(),
instance_to_stmt: Default::default(),
resolver: None,
codegen_callback: Some(Arc::new(GenCall::new(Box::new(|ctx, _, _, args| {
let arg = args[0].1;
let round_intrinsic =
ctx.module.get_function("llvm.round.f64").unwrap_or_else(|| {
let float = ctx.ctx.f64_type();
let fn_type = float.fn_type(&[float.into()], false);
ctx.module.add_function("llvm.round.f64", fn_type, None)
});
let val = ctx
.builder
.build_call(round_intrinsic, &[arg], "round")
.try_as_basic_value()
.left()
.unwrap();
Some(
ctx.builder
.build_float_to_signed_int(
val.into_float_value(),
ctx.ctx.i32_type(),
"fptosi",
)
.into(),
)
})))),
})),
Arc::new(RwLock::new(TopLevelDef::Function {
name: "round64".into(),
simple_name: "round64".into(),
signature: primitives.1.add_ty(TypeEnum::TFunc(RefCell::new(FunSignature {
args: vec![FuncArg { name: "_".into(), ty: float, default_value: None }],
ret: int64,
vars: Default::default(),
}))),
var_id: Default::default(),
instance_to_symbol: Default::default(),
instance_to_stmt: Default::default(),
resolver: None,
codegen_callback: Some(Arc::new(GenCall::new(Box::new(|ctx, _, _, args| {
let arg = args[0].1;
let round_intrinsic =
ctx.module.get_function("llvm.round.f64").unwrap_or_else(|| {
let float = ctx.ctx.f64_type();
let fn_type = float.fn_type(&[float.into()], false);
ctx.module.add_function("llvm.round.f64", fn_type, None)
});
let val = ctx
.builder
.build_call(round_intrinsic, &[arg], "round")
.try_as_basic_value()
.left()
.unwrap();
Some(
ctx.builder
.build_float_to_signed_int(
val.into_float_value(),
ctx.ctx.i64_type(),
"fptosi",
)
.into(),
)
})))),
})),
2021-10-23 23:53:36 +08:00
Arc::new(RwLock::new(TopLevelDef::Function {
name: "range".into(),
simple_name: "range".into(),
signature: primitives.1.add_ty(TypeEnum::TFunc(RefCell::new(FunSignature {
args: vec![
FuncArg { name: "start".into(), ty: int32, default_value: None },
FuncArg {
name: "stop".into(),
ty: int32,
// placeholder
default_value: Some(SymbolValue::I32(0)),
},
FuncArg {
name: "step".into(),
ty: int32,
default_value: Some(SymbolValue::I32(1)),
},
],
ret: range,
vars: Default::default(),
}))),
var_id: Default::default(),
instance_to_symbol: Default::default(),
instance_to_stmt: Default::default(),
resolver: None,
codegen_callback: Some(Arc::new(GenCall::new(Box::new(|ctx, _, _, args| {
let mut start = None;
let mut stop = None;
let mut step = None;
let int32 = ctx.ctx.i32_type();
let zero = int32.const_zero();
for (i, arg) in args.iter().enumerate() {
if arg.0 == Some("start".into()) {
start = Some(arg.1);
} else if arg.0 == Some("stop".into()) {
stop = Some(arg.1);
} else if arg.0 == Some("step".into()) {
step = Some(arg.1);
} else if i == 0 {
start = Some(arg.1);
} else if i == 1 {
stop = Some(arg.1);
} else if i == 2 {
step = Some(arg.1);
}
}
// TODO: error when step == 0
let step = step.unwrap_or_else(|| int32.const_int(1, false).into());
let stop = stop.unwrap_or_else(|| {
let v = start.unwrap();
start = None;
v
});
let start = start.unwrap_or_else(|| int32.const_zero().into());
let ty = int32.array_type(3);
let ptr = ctx.builder.build_alloca(ty, "range");
unsafe {
let a = ctx.builder.build_in_bounds_gep(ptr, &[zero, zero], "start");
let b = ctx.builder.build_in_bounds_gep(
ptr,
&[zero, int32.const_int(1, false)],
"end",
);
let c = ctx.builder.build_in_bounds_gep(
ptr,
&[zero, int32.const_int(2, false)],
"step",
);
ctx.builder.build_store(a, start);
ctx.builder.build_store(b, stop);
ctx.builder.build_store(c, step);
}
Some(ptr.into())
})))),
})),
2021-11-02 23:22:37 +08:00
Arc::new(RwLock::new(TopLevelDef::Function {
name: "str".into(),
simple_name: "str".into(),
signature: primitives.1.add_ty(TypeEnum::TFunc(RefCell::new(FunSignature {
args: vec![FuncArg { name: "_".into(), ty: string, default_value: None }],
ret: string,
vars: Default::default(),
}))),
var_id: Default::default(),
instance_to_symbol: Default::default(),
instance_to_stmt: Default::default(),
resolver: None,
codegen_callback: Some(Arc::new(GenCall::new(Box::new(|_, _, _, args|
Some(args[0].1)
)))),
})),
];
let ast_list: Vec<Option<ast::Stmt<()>>> =
(0..top_level_def_list.len()).map(|_| None).collect();
izip!(top_level_def_list, ast_list).collect_vec()
};
let primitives_ty = primitives.0;
let mut unifier = primitives.1;
2021-09-22 17:19:27 +08:00
let mut keyword_list: HashSet<StrRef> = HashSet::from_iter(vec![
"Generic".into(),
"virtual".into(),
"list".into(),
"tuple".into(),
"int32".into(),
"int64".into(),
"float".into(),
"bool".into(),
"none".into(),
"None".into(),
2021-11-02 18:56:14 +08:00
"range".into(),
2021-11-02 23:22:37 +08:00
"str".into(),
"self".into(),
"Kernel".into(),
"KernelImmutable".into(),
]);
let defined_names: HashSet<String> = Default::default();
let method_class: HashMap<DefinitionId, DefinitionId> = Default::default();
2021-09-19 22:54:06 +08:00
2021-09-22 17:19:27 +08:00
let mut built_in_id: HashMap<StrRef, DefinitionId> = Default::default();
let mut built_in_ty: HashMap<StrRef, Type> = Default::default();
2021-09-19 22:54:06 +08:00
2021-10-23 23:53:36 +08:00
for (id, name) in
2021-11-02 23:22:37 +08:00
["int32", "int64", "float", "round", "round64", "range", "str"].iter().rev().enumerate()
2021-10-23 23:53:36 +08:00
{
let name = (**name).into();
let id = definition_ast_list.len() - id - 1;
let def = definition_ast_list[id].0.read();
if let TopLevelDef::Function { simple_name, signature, .. } = &*def {
assert!(name == *simple_name);
built_in_ty.insert(name, *signature);
built_in_id.insert(name, DefinitionId(id));
} else {
unreachable!()
}
}
for (name, sig, codegen_callback) in builtins {
let fun_sig = unifier.add_ty(TypeEnum::TFunc(RefCell::new(sig)));
2021-09-22 17:56:48 +08:00
built_in_ty.insert(name, fun_sig);
built_in_id.insert(name, DefinitionId(definition_ast_list.len()));
definition_ast_list.push((
Arc::new(RwLock::new(TopLevelDef::Function {
2021-09-22 17:19:27 +08:00
name: name.into(),
simple_name: name,
signature: fun_sig,
instance_to_stmt: Default::default(),
instance_to_symbol: Default::default(),
var_id: Default::default(),
resolver: None,
codegen_callback: Some(codegen_callback),
})),
2021-09-19 22:54:06 +08:00
None,
));
keyword_list.insert(name);
}
2021-09-19 22:54:06 +08:00
(
TopLevelComposer {
built_in_num: definition_ast_list.len(),
definition_ast_list,
primitives_ty,
unifier,
keyword_list,
defined_names,
method_class,
},
built_in_id,
built_in_ty,
)
}
pub fn make_top_level_context(&self) -> TopLevelContext {
TopLevelContext {
definitions: RwLock::new(
self.definition_ast_list.iter().map(|(x, ..)| x.clone()).collect_vec(),
)
.into(),
// NOTE: only one for now
unifiers: Arc::new(RwLock::new(vec![(
self.unifier.get_shared_unifier(),
self.primitives_ty,
)])),
personality_symbol: None,
}
}
fn extract_def_list(&self) -> Vec<Arc<RwLock<TopLevelDef>>> {
self.definition_ast_list.iter().map(|(def, ..)| def.clone()).collect_vec()
}
2021-09-27 19:05:13 +08:00
/// register, just remember the names of top level classes/function
/// and check duplicate class/method/function definition
pub fn register_top_level(
&mut self,
ast: ast::Stmt<()>,
2021-10-16 18:08:13 +08:00
resolver: Option<Arc<dyn SymbolResolver + Send + Sync>>,
mod_path: String,
2021-09-22 17:19:27 +08:00
) -> Result<(StrRef, DefinitionId, Option<Type>), String> {
let defined_names = &mut self.defined_names;
match &ast.node {
ast::StmtKind::ClassDef { name: class_name, body, .. } => {
if self.keyword_list.contains(class_name) {
return Err("cannot use keyword as a class name".into());
}
if !defined_names.insert({
let mut n = mod_path.clone();
2021-09-22 17:19:27 +08:00
n.push_str(&class_name.to_string());
n
}) {
return Err("duplicate definition of class".into());
}
2021-09-22 17:56:48 +08:00
let class_name = *class_name;
let class_def_id = self.definition_ast_list.len();
// since later when registering class method, ast will still be used,
2021-09-27 19:05:13 +08:00
// here push None temporarily, later will move the ast inside
2021-09-19 22:54:06 +08:00
let constructor_ty = self.unifier.get_fresh_var().0;
let mut class_def_ast = (
Arc::new(RwLock::new(Self::make_top_level_class_def(
class_def_id,
resolver.clone(),
2021-09-22 17:19:27 +08:00
class_name,
Some(constructor_ty),
))),
None,
);
// parse class def body and register class methods into the def list.
// module's symbol resolver would not know the name of the class methods,
// thus cannot return their definition_id
type MethodInfo = (
// the simple method name without class name
2021-09-22 17:19:27 +08:00
StrRef,
// in this top level def, method name is prefixed with the class name
Arc<RwLock<TopLevelDef>>,
DefinitionId,
Type,
ast::Stmt<()>,
);
let mut class_method_name_def_ids: Vec<MethodInfo> = Vec::new();
// we do not push anything to the def list, so we keep track of the index
// and then push in the correct order after the for loop
let mut class_method_index_offset = 0;
2021-10-09 16:20:49 +08:00
let mut contains_constructor = false;
let init_id = "__init__".into();
for b in body {
if let ast::StmtKind::FunctionDef { name: method_name, .. } = &b.node {
2021-10-09 16:20:49 +08:00
if method_name == &init_id {
contains_constructor = true;
}
if self.keyword_list.contains(method_name) {
return Err("cannot use keyword as a method name".into());
}
let global_class_method_name = {
let mut n = mod_path.clone();
n.push_str(
2021-09-22 17:19:27 +08:00
Self::make_class_method_name(
class_name.into(),
&method_name.to_string(),
)
.as_str(),
);
n
};
if !defined_names.insert(global_class_method_name.clone()) {
return Err("duplicate class method definition".into());
}
let method_def_id = self.definition_ast_list.len() + {
// plus 1 here since we already have the class def
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((
2021-09-22 17:56:48 +08:00
*method_name,
RwLock::new(Self::make_top_level_function_def(
global_class_method_name,
2021-09-22 17:56:48 +08:00
*method_name,
// later unify with parsed type
dummy_method_type.0,
resolver.clone(),
))
.into(),
DefinitionId(method_def_id),
dummy_method_type.0,
b.clone(),
));
} else {
// do nothing
continue;
}
}
// move the ast to the entry of the class in the ast_list
class_def_ast.1 = Some(ast);
// get the methods into the top level class_def
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() {
2021-09-22 17:56:48 +08:00
methods.push((*name, *ty, *id));
self.method_class.insert(*id, DefinitionId(class_def_id));
} 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, _, _, ast) in class_method_name_def_ids {
self.definition_ast_list.push((def, Some(ast)));
}
2021-10-10 15:01:06 +08:00
let result_ty = if contains_constructor { Some(constructor_ty) } else { None };
2021-10-09 16:20:49 +08:00
Ok((class_name, DefinitionId(class_def_id), result_ty))
}
ast::StmtKind::FunctionDef { name, .. } => {
// if self.keyword_list.contains(name) {
// return Err("cannot use keyword as a top level function name".into());
// }
let global_fun_name = {
let mut n = mod_path;
2021-09-22 17:19:27 +08:00
n.push_str(&name.to_string());
n
};
if !defined_names.insert(global_fun_name.clone()) {
return Err("duplicate top level function define".into());
}
2021-09-22 17:19:27 +08:00
let fun_name = *name;
let ty_to_be_unified = self.unifier.get_fresh_var().0;
// add to the definition list
self.definition_ast_list.push((
RwLock::new(Self::make_top_level_function_def(
global_fun_name,
2021-09-22 17:19:27 +08:00
*name,
// dummy here, unify with correct type later
ty_to_be_unified,
resolver,
))
.into(),
Some(ast),
));
// return
Ok((
fun_name,
DefinitionId(self.definition_ast_list.len() - 1),
Some(ty_to_be_unified),
))
}
2021-09-27 19:05:13 +08:00
_ => Err("only registrations of top level classes/functions are supported".into()),
}
}
pub fn start_analysis(&mut self, inference: bool) -> Result<(), String> {
self.analyze_top_level_class_type_var()?;
self.analyze_top_level_class_bases()?;
self.analyze_top_level_class_fields_methods()?;
self.analyze_top_level_function()?;
if inference {
self.analyze_function_instance()?;
}
Ok(())
}
/// step 1, analyze the type vars associated with top level class
fn analyze_top_level_class_type_var(&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;
// skip 5 to skip analyzing the primitives
for (class_def, class_ast) in def_list.iter().skip(self.built_in_num) {
// 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() {
if let Some(ast::Located {
node: ast::StmtKind::ClassDef { bases, .. }, ..
}) = class_ast
{
(bases, type_vars, resolver)
} else {
unreachable!("must be both class")
}
} else {
continue;
}
};
let class_resolver = class_resolver.as_ref().unwrap();
let class_resolver = class_resolver.deref();
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 {
matches!(
&value.node,
2021-09-22 17:19:27 +08:00
ast::ExprKind::Name { id, .. } if id == &"Generic".into()
)
} =>
{
if !is_generic {
is_generic = true;
} else {
return Err("Only single Generic[...] can be in bases".into());
}
let type_var_list: Vec<&ast::Expr<()>>;
// if `class A(Generic[T, V, G])`
if let ast::ExprKind::Tuple { elts, .. } = &slice.node {
type_var_list = elts.iter().collect_vec();
// `class A(Generic[T])`
} else {
type_var_list = vec![slice.deref()];
}
// parse the type vars
let type_vars = type_var_list
.into_iter()
.map(|e| {
class_resolver.parse_type_annotation(
&temp_def_list,
unifier,
primitives_store,
e,
)
})
.collect::<Result<Vec<_>, _>>()?;
// check if all are unique type vars
let all_unique_type_var = {
let mut occured_type_var_id: HashSet<u32> = HashSet::new();
type_vars.iter().all(|x| {
let ty = unifier.get_ty(*x);
if let TypeEnum::TVar { id, .. } = ty.as_ref() {
occured_type_var_id.insert(*id)
} else {
false
}
})
};
if !all_unique_type_var {
return Err("expect unique type variables".into());
}
// add to TopLevelDef
class_def_type_vars.extend(type_vars);
}
// if others, do nothing in this function
_ => continue,
}
}
}
Ok(())
}
/// step 2, base classes.
/// now that the type vars of all classes are done, handle base classes and
/// put Self class into the ancestors list. We only allow single inheritance
fn analyze_top_level_class_bases(&mut self) -> Result<(), String> {
2021-09-25 22:02:19 +08:00
if self.unifier.top_level.is_none() {
let ctx = Arc::new(self.make_top_level_context());
self.unifier.top_level = Some(ctx);
}
let temp_def_list = self.extract_def_list();
let unifier = self.unifier.borrow_mut();
// first, only push direct parent into the list
// skip 5 to skip analyzing the primitives
for (class_def, class_ast) in self.definition_ast_list.iter_mut().skip(self.built_in_num) {
let mut class_def = class_def.write();
let (class_def_id, class_bases, class_ancestors, class_resolver, class_type_vars) = {
if let TopLevelDef::Class { ancestors, resolver, object_id, type_vars, .. } =
class_def.deref_mut()
{
if let Some(ast::Located {
node: ast::StmtKind::ClassDef { bases, .. }, ..
}) = class_ast
{
(object_id, bases, ancestors, resolver, type_vars)
} else {
unreachable!("must be both class")
}
} else {
continue;
}
};
let class_resolver = class_resolver.as_ref().unwrap();
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,
2021-09-22 17:19:27 +08:00
ast::ExprKind::Name { id, .. } if id == &"Generic".into()
)
) {
continue;
}
if has_base {
return Err("a class def can only have at most one base class \
declaration and one generic declaration"
.into());
}
has_base = true;
// the function parse_ast_to make sure that no type var occured in
// bast_ty if it is a CustomClassKind
let base_ty = parse_ast_to_type_annotation_kinds(
2021-10-16 18:08:13 +08:00
class_resolver,
&temp_def_list,
unifier,
&self.primitives_ty,
b,
vec![(*class_def_id, class_type_vars.clone())].into_iter().collect(),
)?;
if let TypeAnnotation::CustomClassKind { .. } = &base_ty {
class_ancestors.push(base_ty);
} else {
return Err("class base declaration can only be custom class".into());
}
}
}
// second, get all ancestors
let mut ancestors_store: HashMap<DefinitionId, Vec<TypeAnnotation>> = Default::default();
// skip 5 to skip analyzing the primitives
for (class_def, _) in self.definition_ast_list.iter().skip(self.built_in_num) {
let class_def = class_def.read();
let (class_ancestors, class_id) = {
if let TopLevelDef::Class { ancestors, object_id, .. } = class_def.deref() {
(ancestors, *object_id)
} else {
continue;
}
};
ancestors_store.insert(
class_id,
// if class has direct parents, get all ancestors of its parents. Else just empty
if class_ancestors.is_empty() {
vec![]
} else {
Self::get_all_ancestors_helper(&class_ancestors[0], temp_def_list.as_slice())?
},
);
}
// insert the ancestors to the def list
// skip 5 to skip analyzing the primitives
for (class_def, _) in self.definition_ast_list.iter_mut().skip(self.built_in_num) {
let mut class_def = class_def.write();
let (class_ancestors, class_id, class_type_vars) = {
if let TopLevelDef::Class { ancestors, object_id, type_vars, .. } =
class_def.deref_mut()
{
(ancestors, *object_id, type_vars)
} else {
continue;
}
};
let ans = ancestors_store.get_mut(&class_id).unwrap();
class_ancestors.append(ans);
// insert self type annotation to the front of the vector to maintain the order
class_ancestors
.insert(0, make_self_type_annotation(class_type_vars.as_slice(), class_id));
}
Ok(())
}
/// step 3, class fields and methods
fn analyze_top_level_class_fields_methods(&mut self) -> Result<(), String> {
let temp_def_list = self.extract_def_list();
let primitives = &self.primitives_ty;
let def_ast_list = &self.definition_ast_list;
let unifier = self.unifier.borrow_mut();
let mut type_var_to_concrete_def: HashMap<Type, TypeAnnotation> = HashMap::new();
// skip 5 to skip analyzing the primitives
for (class_def, class_ast) in def_ast_list.iter().skip(self.built_in_num) {
if matches!(&*class_def.read(), TopLevelDef::Class { .. }) {
Self::analyze_single_class_methods_fields(
class_def.clone(),
&class_ast.as_ref().unwrap().node,
&temp_def_list,
unifier,
primitives,
&mut type_var_to_concrete_def,
&self.keyword_list,
)?
}
}
// println!("type_var_to_concrete_def1: {:?}", type_var_to_concrete_def);
// handle the inheritanced methods and fields
let mut current_ancestor_depth: usize = 2;
loop {
let mut finished = true;
for (class_def, _) in def_ast_list.iter().skip(self.built_in_num) {
let mut class_def = class_def.write();
if let TopLevelDef::Class { ancestors, .. } = class_def.deref() {
// if the length of the ancestor is equal to the current depth
// it means that all the ancestors of the class is handled
if ancestors.len() == current_ancestor_depth {
finished = false;
Self::analyze_single_class_ancestors(
class_def.deref_mut(),
&temp_def_list,
unifier,
primitives,
&mut type_var_to_concrete_def,
)?;
}
}
}
if finished {
break;
} else {
current_ancestor_depth += 1;
}
if current_ancestor_depth > def_ast_list.len() + 1 {
unreachable!("cannot be longer than the whole top level def list")
}
}
// println!("type_var_to_concrete_def3: {:?}\n", type_var_to_concrete_def);
// unification of previously assigned typevar
for (ty, def) in type_var_to_concrete_def {
// println!(
// "{:?}_{} -> {:?}\n",
// ty,
// unifier.stringify(ty,
// &mut |id| format!("class{}", id),
// &mut |id| format!("tvar{}", id)
// ),
// def
// );
let target_ty =
get_type_from_type_annotation_kinds(&temp_def_list, unifier, primitives, &def)?;
unifier.unify(ty, target_ty)?;
}
Ok(())
}
/// step 4, after class methods are done, top level functions have nothing unknown
fn analyze_top_level_function(&mut self) -> Result<(), String> {
let def_list = &self.definition_ast_list;
let keyword_list = &self.keyword_list;
let temp_def_list = self.extract_def_list();
let unifier = self.unifier.borrow_mut();
let primitives_store = &self.primitives_ty;
// skip 5 to skip analyzing the primitives
for (function_def, function_ast) in def_list.iter().skip(self.built_in_num) {
let mut function_def = function_def.write();
let function_def = function_def.deref_mut();
let function_ast = if let Some(x) = function_ast.as_ref() {
x
} else {
// if let TopLevelDef::Function { name, .. } = ``
continue;
};
if let TopLevelDef::Function { signature: dummy_ty, resolver, var_id, .. } =
function_def
{
if matches!(unifier.get_ty(*dummy_ty).as_ref(), TypeEnum::TFunc(_)) {
// already have a function type, is class method, skip
continue;
}
if let ast::StmtKind::FunctionDef { args, returns, .. } = &function_ast.node {
let resolver = resolver.as_ref();
let resolver = resolver.unwrap();
let resolver = resolver.deref();
let mut function_var_map: HashMap<u32, Type> = HashMap::new();
let arg_types = {
// make sure no duplicate parameter
2021-09-22 17:19:27 +08:00
let mut defined_paramter_name: HashSet<_> = HashSet::new();
let have_unique_fuction_parameter_name = args.args.iter().all(|x| {
2021-09-22 17:19:27 +08:00
defined_paramter_name.insert(x.node.arg)
&& !keyword_list.contains(&x.node.arg)
});
if !have_unique_fuction_parameter_name {
return Err("top level function must have unique parameter names \
and names thould not be the same as the keywords"
.into());
}
args.args
.iter()
.map(|x| -> Result<FuncArg, String> {
let annotation = x
.node
.annotation
.as_ref()
2021-10-10 15:01:06 +08:00
.ok_or_else(|| {
format!(
"function parameter `{}` at {} need type annotation",
x.node.arg, x.location
)
})?
.as_ref();
let type_annotation = parse_ast_to_type_annotation_kinds(
2021-10-16 18:08:13 +08:00
resolver,
temp_def_list.as_slice(),
unifier,
primitives_store,
annotation,
// NOTE: since only class need this, for function
// it should be fine to be empty map
HashMap::new(),
)?;
let type_vars_within =
get_type_var_contained_in_type_annotation(&type_annotation)
.into_iter()
.map(|x| -> Result<(u32, Type), String> {
if let TypeAnnotation::TypeVarKind(ty) = x {
Ok((Self::get_var_id(ty, unifier)?, ty))
} else {
unreachable!("must be type var annotation kind")
}
})
.collect::<Result<Vec<_>, _>>()?;
for (id, ty) in type_vars_within {
if let Some(prev_ty) = function_var_map.insert(id, ty) {
// if already have the type inserted, make sure they are the same thing
assert_eq!(prev_ty, ty);
}
}
let ty = get_type_from_type_annotation_kinds(
temp_def_list.as_ref(),
unifier,
primitives_store,
&type_annotation,
)?;
Ok(FuncArg {
2021-09-22 17:56:48 +08:00
name: x.node.arg,
ty,
default_value: Default::default(),
})
})
.collect::<Result<Vec<_>, _>>()?
};
let return_ty = {
if let Some(returns) = returns {
let return_ty_annotation = {
let return_annotation = returns.as_ref();
parse_ast_to_type_annotation_kinds(
2021-10-16 18:08:13 +08:00
resolver,
&temp_def_list,
unifier,
primitives_store,
return_annotation,
// NOTE: since only class need this, for function
// it should be fine to be empty map
HashMap::new(),
)?
};
let type_vars_within =
get_type_var_contained_in_type_annotation(&return_ty_annotation)
.into_iter()
.map(|x| -> Result<(u32, Type), String> {
if let TypeAnnotation::TypeVarKind(ty) = x {
Ok((Self::get_var_id(ty, unifier)?, ty))
} else {
unreachable!("must be type var here")
}
})
.collect::<Result<Vec<_>, _>>()?;
for (id, ty) in type_vars_within {
if let Some(prev_ty) = function_var_map.insert(id, ty) {
// if already have the type inserted, make sure they are the same thing
assert_eq!(prev_ty, ty);
}
}
get_type_from_type_annotation_kinds(
&temp_def_list,
unifier,
primitives_store,
&return_ty_annotation,
)?
} else {
primitives_store.none
}
};
var_id.extend_from_slice(
function_var_map.keys().into_iter().copied().collect_vec().as_slice(),
);
let function_ty = unifier.add_ty(TypeEnum::TFunc(
FunSignature { args: arg_types, ret: return_ty, vars: function_var_map }
.into(),
));
2021-09-25 22:02:19 +08:00
unifier
.unify(*dummy_ty, function_ty)
.map_err(|old| format!("{} at {}", old, function_ast.location))?;
} else {
unreachable!("must be both function");
}
} else {
// not top level function def, skip
continue;
}
}
Ok(())
}
fn analyze_single_class_methods_fields(
class_def: Arc<RwLock<TopLevelDef>>,
class_ast: &ast::StmtKind<()>,
temp_def_list: &[Arc<RwLock<TopLevelDef>>],
unifier: &mut Unifier,
primitives: &PrimitiveStore,
type_var_to_concrete_def: &mut HashMap<Type, TypeAnnotation>,
2021-09-22 17:19:27 +08:00
keyword_list: &HashSet<StrRef>,
) -> Result<(), String> {
let mut class_def = class_def.write();
let (
class_id,
_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,
..
2021-09-19 22:54:06 +08:00
} = &mut *class_def
{
if let ast::StmtKind::ClassDef { name, bases, body, .. } = &class_ast {
(*object_id, *name, bases, body, ancestors, fields, methods, type_vars, resolver)
} else {
unreachable!("here must be class def ast");
}
} else {
unreachable!("here must be toplevel class def");
};
let class_resolver = class_resolver.as_ref().unwrap();
let class_resolver = class_resolver.as_ref();
2021-09-22 17:19:27 +08:00
let mut defined_fields: HashSet<_> = HashSet::new();
for b in class_body_ast {
match &b.node {
ast::StmtKind::FunctionDef { args, returns, name, .. } => {
let (method_dummy_ty, method_id) =
2021-09-22 17:19:27 +08:00
Self::get_class_method_def_info(class_methods_def, *name)?;
// the method var map can surely include the class's generic parameters
let mut method_var_map: HashMap<u32, Type> = class_type_vars_def
.iter()
.map(|ty| {
if let TypeEnum::TVar { id, .. } = unifier.get_ty(*ty).as_ref() {
(*id, *ty)
} else {
unreachable!("must be type var here")
}
})
.collect();
let arg_types: Vec<FuncArg> = {
// check method parameters cannot have same name
2021-09-22 17:19:27 +08:00
let mut defined_paramter_name: HashSet<_> = HashSet::new();
let zelf: StrRef = "self".into();
let have_unique_fuction_parameter_name = args.args.iter().all(|x| {
2021-09-22 17:56:48 +08:00
defined_paramter_name.insert(x.node.arg)
2021-09-22 17:19:27 +08:00
&& (!keyword_list.contains(&x.node.arg) || x.node.arg == zelf)
});
if !have_unique_fuction_parameter_name {
return Err("class method must have unique parameter names \
and names thould not be the same as the keywords"
.into());
}
2021-09-22 17:19:27 +08:00
if name == &"__init__".into() && !defined_paramter_name.contains(&zelf) {
return Err("__init__ function must have a `self` parameter".into());
}
2021-09-22 17:19:27 +08:00
if !defined_paramter_name.contains(&zelf) {
return Err("currently does not support static method".into());
}
let mut result = Vec::new();
for x in &args.args {
2021-09-22 17:19:27 +08:00
let name = x.node.arg;
if name != zelf {
let type_ann = {
let annotation_expr = x
.node
.annotation
.as_ref()
2021-10-10 15:01:06 +08:00
.ok_or_else(|| {
format!(
"type annotation for `{}` at {} needed",
x.node.arg, x.location
)
})?
.as_ref();
parse_ast_to_type_annotation_kinds(
2021-10-16 18:08:13 +08:00
class_resolver,
temp_def_list,
unifier,
primitives,
annotation_expr,
vec![(class_id, class_type_vars_def.clone())]
.into_iter()
.collect(),
)?
};
// find type vars within this method parameter type annotation
let type_vars_within =
get_type_var_contained_in_type_annotation(&type_ann);
// handle the class type var and the method type var
for type_var_within in type_vars_within {
if let TypeAnnotation::TypeVarKind(ty) = type_var_within {
let id = Self::get_var_id(ty, unifier)?;
if let Some(prev_ty) = method_var_map.insert(id, ty) {
// if already in the list, make sure they are the same?
assert_eq!(prev_ty, ty);
}
} else {
unreachable!("must be type var annotation");
}
}
// finish handling type vars
let dummy_func_arg = FuncArg {
name,
ty: unifier.get_fresh_var().0,
// TODO: 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)
}
}
result
};
let ret_type = {
if let Some(result) = returns {
let result = result.as_ref();
let annotation = parse_ast_to_type_annotation_kinds(
2021-10-16 18:08:13 +08:00
class_resolver,
temp_def_list,
unifier,
primitives,
result,
vec![(class_id, class_type_vars_def.clone())].into_iter().collect(),
)?;
// find type vars within this return type annotation
let type_vars_within =
get_type_var_contained_in_type_annotation(&annotation);
// handle the class type var and the method type var
for type_var_within in type_vars_within {
if let TypeAnnotation::TypeVarKind(ty) = type_var_within {
let id = Self::get_var_id(ty, unifier)?;
if let Some(prev_ty) = method_var_map.insert(id, ty) {
// if already in the list, make sure they are the same?
assert_eq!(prev_ty, ty);
}
} else {
unreachable!("must be type var annotation");
}
}
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 do not have return annotation, return none
// for uniform handling, still use type annoatation
let dummy_return_type = unifier.get_fresh_var().0;
type_var_to_concrete_def.insert(
dummy_return_type,
TypeAnnotation::PrimitiveKind(primitives.none),
);
dummy_return_type
}
};
if let TopLevelDef::Function { var_id, .. } =
temp_def_list.get(method_id.0).unwrap().write().deref_mut()
{
var_id.extend_from_slice(
method_var_map.keys().into_iter().copied().collect_vec().as_slice(),
);
}
let method_type = unifier.add_ty(TypeEnum::TFunc(
FunSignature { args: arg_types, ret: ret_type, vars: method_var_map }
.into(),
));
// unify now since function type is not in type annotation define
// which should be fine since type within method_type will be subst later
unifier.unify(method_dummy_ty, method_type)?;
}
ast::StmtKind::AnnAssign { target, annotation, value: None, .. } => {
if let ast::ExprKind::Name { id: attr, .. } = &target.node {
if defined_fields.insert(attr.to_string()) {
let dummy_field_type = unifier.get_fresh_var().0;
2021-09-22 17:19:27 +08:00
class_fields_def.push((*attr, dummy_field_type));
// handle Kernel[T], KernelImmutable[T]
let annotation = {
match &annotation.as_ref().node {
ast::ExprKind::Subscript { value, slice, .. }
if {
2021-09-22 17:19:27 +08:00
matches!(&value.node, ast::ExprKind::Name { id, .. }
if id == &"Kernel".into() || id == &"KernelImmutable".into())
} =>
{
slice
}
_ => annotation,
}
};
let annotation = parse_ast_to_type_annotation_kinds(
2021-10-16 18:08:13 +08:00
class_resolver,
temp_def_list,
unifier,
primitives,
annotation.as_ref(),
vec![(class_id, class_type_vars_def.clone())].into_iter().collect(),
)?;
// find type vars within this return type annotation
let type_vars_within =
get_type_var_contained_in_type_annotation(&annotation);
// handle the class type var and the method type var
for type_var_within in type_vars_within {
if let TypeAnnotation::TypeVarKind(t) = type_var_within {
if !class_type_vars_def.contains(&t) {
return Err("class fields can only use type \
vars declared as class generic type vars"
.into());
}
} else {
unreachable!("must be type var annotation");
}
}
type_var_to_concrete_def.insert(dummy_field_type, annotation);
} else {
return Err("same class fields defined twice".into());
}
} else {
return Err("unsupported statement type in class definition body".into());
}
}
ast::StmtKind::Pass => {}
ast::StmtKind::Expr { value: _ } => {} // typically a docstring; ignoring all expressions matches CPython behavior
_ => return Err("unsupported statement type in class definition body".into()),
}
}
Ok(())
}
fn analyze_single_class_ancestors(
class_def: &mut TopLevelDef,
temp_def_list: &[Arc<RwLock<TopLevelDef>>],
unifier: &mut Unifier,
_primitives: &PrimitiveStore,
type_var_to_concrete_def: &mut HashMap<Type, TypeAnnotation>,
) -> Result<(), String> {
let (
_class_id,
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,
..
} = class_def
{
(*object_id, ancestors, fields, methods, type_vars, resolver)
} else {
unreachable!("here must be class def ast");
};
// since when this function is called, the ancestors of the direct parent
// are supposed to be already handled, so we only need to deal with the direct parent
let base = class_ancestor_def.get(1).unwrap();
if let TypeAnnotation::CustomClassKind { id, params: _ } = base {
let base = temp_def_list.get(id.0).unwrap();
let base = base.read();
if let TopLevelDef::Class { methods, fields, .. } = &*base {
// handle methods override
// since we need to maintain the order, create a new list
2021-09-22 17:19:27 +08:00
let mut new_child_methods: Vec<(StrRef, Type, DefinitionId)> = Vec::new();
let mut is_override: HashSet<StrRef> = HashSet::new();
for (anc_method_name, anc_method_ty, anc_method_def_id) in methods {
// find if there is a method with same name in the child class
let mut to_be_added = (*anc_method_name, *anc_method_ty, *anc_method_def_id);
for (class_method_name, class_method_ty, class_method_defid) in
class_methods_def.iter()
{
if class_method_name == anc_method_name {
// ignore and handle self
// if is __init__ method, no need to check return type
2021-09-22 17:19:27 +08:00
let ok = class_method_name == &"__init__".into()
|| Self::check_overload_function_type(
*class_method_ty,
*anc_method_ty,
unifier,
type_var_to_concrete_def,
);
if !ok {
return Err("method has same name as ancestors' method, but incompatible type".into());
}
// mark it as added
2021-09-22 17:19:27 +08:00
is_override.insert(*class_method_name);
to_be_added =
(*class_method_name, *class_method_ty, *class_method_defid);
break;
}
}
new_child_methods.push(to_be_added);
}
// add those that are not overriding method to the new_child_methods
for (class_method_name, class_method_ty, class_method_defid) in
class_methods_def.iter()
{
if !is_override.contains(class_method_name) {
new_child_methods.push((
2021-09-22 17:19:27 +08:00
*class_method_name,
*class_method_ty,
*class_method_defid,
));
}
}
// use the new_child_methods to replace all the elements in `class_methods_def`
class_methods_def.drain(..);
class_methods_def.extend(new_child_methods);
// handle class fields
2021-09-22 17:19:27 +08:00
let mut new_child_fields: Vec<(StrRef, Type)> = Vec::new();
// let mut is_override: HashSet<_> = HashSet::new();
for (anc_field_name, anc_field_ty) in fields {
2021-09-22 17:19:27 +08:00
let to_be_added = (*anc_field_name, *anc_field_ty);
// find if there is a fields with the same name in the child class
for (class_field_name, ..) in class_fields_def.iter() {
if class_field_name == anc_field_name {
// let ok = Self::check_overload_field_type(
// *class_field_ty,
// *anc_field_ty,
// unifier,
// type_var_to_concrete_def,
// );
// if !ok {
// return Err("fields has same name as ancestors' field, but incompatible type".into());
// }
// // mark it as added
// is_override.insert(class_field_name.to_string());
// to_be_added = (class_field_name.to_string(), *class_field_ty);
// break;
return Err(format!(
"field `{}` has already declared in the ancestor classes",
class_field_name
));
}
}
new_child_fields.push(to_be_added);
}
for (class_field_name, class_field_ty) in class_fields_def.iter() {
if !is_override.contains(class_field_name) {
2021-09-22 17:19:27 +08:00
new_child_fields.push((*class_field_name, *class_field_ty));
}
}
class_fields_def.drain(..);
class_fields_def.extend(new_child_fields);
} else {
unreachable!("must be top level class def")
}
} else {
unreachable!("must be class type annotation")
}
Ok(())
}
/// step 5, analyze and call type inferecer to fill the `instance_to_stmt` of topleveldef::function
fn analyze_function_instance(&mut self) -> Result<(), String> {
// first get the class contructor type correct for the following type check in function body
// also do class field instantiation check
2021-09-25 22:02:19 +08:00
for (def, ast) in self.definition_ast_list.iter().skip(self.built_in_num) {
let class_def = def.read();
if let TopLevelDef::Class {
constructor,
methods,
fields,
type_vars,
name: class_name,
object_id,
resolver: _,
..
} = &*class_def
{
let mut init_id: Option<DefinitionId> = None;
// get the class contructor type correct
let (contor_args, contor_type_vars) = {
let mut constructor_args: Vec<FuncArg> = Vec::new();
let mut type_vars: HashMap<u32, Type> = HashMap::new();
for (name, func_sig, id) in methods {
2021-09-22 17:19:27 +08:00
if name == &"__init__".into() {
init_id = Some(*id);
if let TypeEnum::TFunc(sig) = self.unifier.get_ty(*func_sig).as_ref() {
let FunSignature { args, vars, .. } = &*sig.borrow();
constructor_args.extend_from_slice(args);
type_vars.extend(vars);
} else {
unreachable!("must be typeenum::tfunc")
}
}
}
(constructor_args, type_vars)
};
let self_type = get_type_from_type_annotation_kinds(
self.extract_def_list().as_slice(),
&mut self.unifier,
&self.primitives_ty,
&make_self_type_annotation(type_vars, *object_id),
)?;
let contor_type = self.unifier.add_ty(TypeEnum::TFunc(
FunSignature { args: contor_args, ret: self_type, vars: contor_type_vars }
.into(),
));
2021-09-25 22:02:19 +08:00
self.unifier
.unify(constructor.unwrap(), contor_type)
.map_err(|old| format!("{} at {}", old, ast.as_ref().unwrap().location))?;
// class field instantiation check
if let (Some(init_id), false) = (init_id, fields.is_empty()) {
let init_ast =
self.definition_ast_list.get(init_id.0).unwrap().1.as_ref().unwrap();
if let ast::StmtKind::FunctionDef { name, body, .. } = &init_ast.node {
2021-09-22 17:19:27 +08:00
if name != &"__init__".into() {
unreachable!("must be init function here")
}
let all_inited = Self::get_all_assigned_field(body.as_slice())?;
if fields.iter().any(|(x, _)| !all_inited.contains(x)) {
return Err(format!(
"fields of class {} not fully initialized",
class_name
));
}
}
}
}
}
2021-09-25 22:02:19 +08:00
let ctx = Arc::new(self.make_top_level_context());
// type inference inside function body
2021-09-19 22:54:06 +08:00
for (id, (def, ast)) in self.definition_ast_list.iter().enumerate().skip(self.built_in_num)
{
let mut function_def = def.write();
if let TopLevelDef::Function {
instance_to_stmt,
instance_to_symbol,
name,
simple_name,
signature,
resolver,
..
} = &mut *function_def
{
if let TypeEnum::TFunc(func_sig) = self.unifier.get_ty(*signature).as_ref() {
let FunSignature { args, ret, vars } = &*func_sig.borrow();
// None if is not class method
let self_type = {
if let Some(class_id) = self.method_class.get(&DefinitionId(id)) {
let class_def = self.definition_ast_list.get(class_id.0).unwrap();
let class_def = class_def.0.read();
if let TopLevelDef::Class { type_vars, .. } = &*class_def {
let ty_ann = make_self_type_annotation(type_vars, *class_id);
2021-09-19 22:54:06 +08:00
let self_ty = get_type_from_type_annotation_kinds(
self.extract_def_list().as_slice(),
&mut self.unifier,
&self.primitives_ty,
&ty_ann,
2021-09-19 22:54:06 +08:00
)?;
Some(self_ty)
} else {
unreachable!("must be class def")
}
} else {
None
}
};
let (type_var_subst_comb, no_range_vars) = {
let unifier = &mut self.unifier;
let mut no_ranges: Vec<Type> = Vec::new();
let var_ids = vars.iter().map(|(id, ty)| {
if matches!(unifier.get_ty(*ty).as_ref(), TypeEnum::TVar { range, .. } if range.borrow().is_empty()) {
no_ranges.push(*ty);
}
*id
})
.collect_vec();
let var_combs = vars
.iter()
.map(|(_, ty)| {
unifier.get_instantiations(*ty).unwrap_or_else(|| vec![*ty])
})
.multi_cartesian_product()
.collect_vec();
let mut result: Vec<HashMap<u32, Type>> = Default::default();
for comb in var_combs {
result.push(var_ids.clone().into_iter().zip(comb).collect());
}
// NOTE: if is empty, means no type var, append a empty subst, ok to do this?
if result.is_empty() {
result.push(HashMap::new())
}
(result, no_ranges)
};
for subst in type_var_subst_comb {
// for each instance
let inst_ret = self.unifier.subst(*ret, &subst).unwrap_or(*ret);
let inst_args = {
let unifier = &mut self.unifier;
2021-09-19 22:54:06 +08:00
args.iter()
.map(|a| FuncArg {
2021-09-22 17:56:48 +08:00
name: a.name,
ty: unifier.subst(a.ty, &subst).unwrap_or(a.ty),
default_value: a.default_value.clone(),
})
.collect_vec()
};
let self_type = {
let unifier = &mut self.unifier;
self_type.map(|x| unifier.subst(x, &subst).unwrap_or(x))
};
let mut identifiers = {
// NOTE: none and function args?
2021-09-22 17:19:27 +08:00
let mut result: HashSet<_> = HashSet::new();
result.insert("None".into());
if self_type.is_some() {
result.insert("self".into());
}
2021-09-22 17:56:48 +08:00
result.extend(inst_args.iter().map(|x| x.name));
result
};
let mut calls: HashMap<CodeLocation, CallId> = HashMap::new();
let mut inferencer = Inferencer {
2021-09-25 22:02:19 +08:00
top_level: ctx.as_ref(),
defined_identifiers: identifiers.clone(),
function_data: &mut FunctionData {
resolver: resolver.as_ref().unwrap().clone(),
return_type: if self
.unifier
.unioned(inst_ret, self.primitives_ty.none)
{
None
} else {
Some(inst_ret)
},
// NOTE: allowed type vars
bound_variables: no_range_vars.clone(),
},
unifier: &mut self.unifier,
variable_mapping: {
// NOTE: none and function args?
2021-09-22 17:19:27 +08:00
let mut result: HashMap<StrRef, Type> = HashMap::new();
result.insert("None".into(), self.primitives_ty.none);
if let Some(self_ty) = self_type {
result.insert("self".into(), self_ty);
}
2021-09-22 17:56:48 +08:00
result.extend(inst_args.iter().map(|x| (x.name, x.ty)));
result
},
primitives: &self.primitives_ty,
virtual_checks: &mut Vec::new(),
calls: &mut calls,
};
2021-09-22 16:04:25 +08:00
let fun_body =
if let ast::StmtKind::FunctionDef { body, decorator_list, .. } =
ast.clone().unwrap().node
{
if !decorator_list.is_empty()
&& matches!(&decorator_list[0].node,
2021-09-27 10:13:03 +08:00
ast::ExprKind::Name{ id, .. } if id == &"extern".into())
2021-09-22 16:04:25 +08:00
{
2021-09-22 17:19:27 +08:00
instance_to_symbol.insert("".into(), simple_name.to_string());
2021-09-22 16:04:25 +08:00
continue;
}
body
} else {
unreachable!("must be function def ast")
}
2021-09-22 16:04:25 +08:00
.into_iter()
.map(|b| inferencer.fold_stmt(b))
.collect::<Result<Vec<_>, _>>()?;
let returned =
inferencer.check_block(fun_body.as_slice(), &mut identifiers)?;
if !self.unifier.unioned(inst_ret, self.primitives_ty.none) && !returned {
let def_ast_list = &self.definition_ast_list;
let ret_str = self.unifier.stringify(
inst_ret,
&mut |id| {
2021-10-10 15:01:06 +08:00
if let TopLevelDef::Class { name, .. } =
&*def_ast_list[id].0.read()
{
name.to_string()
} else {
unreachable!("must be class id here")
}
},
&mut |id| format!("tvar{}", id),
);
return Err(format!(
"expected return type of `{}` in function `{}` at {}",
2021-10-10 15:01:06 +08:00
ret_str,
name,
ast.as_ref().unwrap().location
));
}
instance_to_stmt.insert(
// NOTE: refer to codegen/expr/get_subst_key function
{
let unifier = &mut self.unifier;
subst
.keys()
.sorted()
.map(|id| {
let ty = subst.get(id).unwrap();
2021-09-19 22:54:06 +08:00
unifier.stringify(
*ty,
&mut |id| id.to_string(),
&mut |id| id.to_string(),
)
})
.join(", ")
},
2021-09-22 16:04:25 +08:00
FunInstance {
body: Arc::new(fun_body),
unifier_id: 0,
calls: Arc::new(calls),
subst,
},
);
}
} else {
unreachable!("must be typeenum::tfunc")
}
} else {
continue;
}
}
Ok(())
}
}