lyken
/
nac3
forked from M-Labs/nac3
1
0
Fork 0
Code Pull Requests Activity

split top level handling in several functions

This commit is contained in:
ychenfo 2021-08-16 09:46:55 +08:00
parent d3ad894521
commit d8c3c063ec
1 changed files with 251 additions and 323 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

568
nac3core/src/top_level.rs
View File

@ -60,8 +60,6 @@ pub struct TopLevelContext {
pub struct TopLevelComposer { pub struct TopLevelComposer {
// list of top level definitions, same as top level context // list of top level definitions, same as top level context
pub definition_list: Arc<RwLock<Vec<RwLock<TopLevelDef>>>>, pub definition_list: Arc<RwLock<Vec<RwLock<TopLevelDef>>>>,
// list of top level Type, the index is same as the field `definition_list`
pub ty_list: RwLock<Vec<Type>>,
// list of top level ast, the index is same as the field `definition_list` and `ty_list` // list of top level ast, the index is same as the field `definition_list` and `ty_list`
pub ast_list: RwLock<Vec<Option<ast::Stmt<()>>>>, pub ast_list: RwLock<Vec<Option<ast::Stmt<()>>>>,
// start as a primitive unifier, will add more top_level defs inside // start as a primitive unifier, will add more top_level defs inside
@ -70,6 +68,8 @@ pub struct TopLevelComposer {
pub primitives: PrimitiveStore, pub primitives: PrimitiveStore,
// mangled class method name to def_id // mangled class method name to def_id
pub class_method_to_def_id: RwLock<HashMap<String, DefinitionId>>, pub class_method_to_def_id: RwLock<HashMap<String, DefinitionId>>,
// record the def id of the classes whoses fields and methods are to be analyzed
pub to_be_analyzed_class: RwLock<Vec<DefinitionId>>,
} }
impl TopLevelComposer { impl TopLevelComposer {
@ -133,21 +133,13 @@ impl TopLevelComposer {
let ast_list: Vec<Option<ast::Stmt<()>>> = vec![None, None, None, None, None]; let ast_list: Vec<Option<ast::Stmt<()>>> = vec![None, None, None, None, None];
let ty_list: Vec<Type> = vec![
primitives.0.int32,
primitives.0.int64,
primitives.0.float,
primitives.0.bool,
primitives.0.none,
];
let composer = TopLevelComposer { let composer = TopLevelComposer {
definition_list: RwLock::new(top_level_def_list).into(), definition_list: RwLock::new(top_level_def_list).into(),
ty_list: RwLock::new(ty_list),
ast_list: RwLock::new(ast_list), ast_list: RwLock::new(ast_list),
primitives: primitives.0, primitives: primitives.0,
unifier: primitives.1.into(), unifier: primitives.1.into(),
class_method_to_def_id: Default::default(), class_method_to_def_id: Default::default(),
to_be_analyzed_class: Default::default(),
}; };
( (
vec![ vec![
@ -190,17 +182,20 @@ impl TopLevelComposer {
} }
} }
/// step 0, register, just remeber the names of top level classes/function
pub fn register_top_level( pub fn register_top_level(
&mut self, &mut self,
ast: ast::Stmt<()>, ast: ast::Stmt<()>,
resolver: Option<Arc<Mutex<dyn SymbolResolver + Send + Sync>>>, resolver: Option<Arc<Mutex<dyn SymbolResolver + Send + Sync>>>,
) -> Result<(String, DefinitionId, Type), String> { ) -> Result<(String, DefinitionId), String> {
// get write access to the lists let (
let (mut def_list, mut ty_list, mut ast_list) = mut def_list,
(self.definition_list.write(), self.ty_list.write(), self.ast_list.write()); mut ast_list
) = (
self.definition_list.write(),
self.ast_list.write()
);
// will be deleted after tested
assert_eq!(ty_list.len(), def_list.len());
assert_eq!(def_list.len(), ast_list.len()); assert_eq!(def_list.len(), ast_list.len());
match &ast.node { match &ast.node {
@ -208,25 +203,17 @@ impl TopLevelComposer {
let class_name = name.to_string(); let class_name = name.to_string();
let class_def_id = def_list.len(); let class_def_id = def_list.len();
// add the class to the unifier
let ty = self.unifier.write().add_ty(TypeEnum::TObj {
obj_id: DefinitionId(class_def_id),
fields: Default::default(),
params: Default::default(),
});
// add the class to the definition lists // add the class to the definition lists
def_list def_list
.push(Self::make_top_level_class_def(class_def_id, resolver.clone()).into()); .push(Self::make_top_level_class_def(class_def_id, resolver.clone()).into());
ty_list.push(ty);
// since later when registering class method, ast will still be used, // since later when registering class method, ast will still be used,
// here push None temporarly, later will push the ast // here push None temporarly, later will move the ast inside
ast_list.push(None); ast_list.push(None);
// parse class def body and register class methods into the def list. // 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, // 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 // thus cannot return their definition_id? so we have to manage it ourselves
// by using the field `class_method_to_def_id` // by using `class_method_to_def_id`
for b in body { for b in body {
if let ast::StmtKind::FunctionDef { name, .. } = &b.node { if let ast::StmtKind::FunctionDef { name, .. } = &b.node {
let fun_name = Self::name_mangling(class_name.clone(), name); let fun_name = Self::name_mangling(class_name.clone(), name);
@ -248,356 +235,297 @@ impl TopLevelComposer {
) )
.into(), .into(),
); );
ty_list.push(ty);
// the ast of class method is in the class, push None in to the list here // the ast of class method is in the class, push None in to the list here
ast_list.push(None); ast_list.push(None);
// class method, do not let the symbol manager manage it, use our own map // class method, do not let the symbol manager manage it, use our own map
self.class_method_to_def_id.write().insert(fun_name, DefinitionId(def_id)); self.class_method_to_def_id.write().insert(fun_name, DefinitionId(def_id));
// 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__" {
// NOTE: how can this later be fetched?
def_list.push(
TopLevelDef::Initializer { class_id: DefinitionId(class_def_id) }
.into(),
);
// arbitarily push one to make sure the index is correct
ty_list.push(self.primitives.none);
ast_list.push(None);
}
} }
} }
// move the ast to the entry of the class in the ast_list // move the ast to the entry of the class in the ast_list
ast_list[class_def_id] = Some(ast); ast_list[class_def_id] = Some(ast);
// return // put the constructor into the def_list
Ok((class_name, DefinitionId(class_def_id), ty)) def_list.push(
TopLevelDef::Initializer { class_id: DefinitionId(class_def_id) }
.into(),
);
ast_list.push(None);
// class, put its def_id into the to be analyzed set
let mut to_be_analyzed = self.to_be_analyzed_class.write();
to_be_analyzed.push(DefinitionId(class_def_id));
Ok((class_name, DefinitionId(class_def_id)))
} }
ast::StmtKind::FunctionDef { name, .. } => { ast::StmtKind::FunctionDef { name, .. } => {
let fun_name = name.to_string(); let fun_name = name.to_string();
// add to the unifier
let ty = self.unifier.write().add_ty(TypeEnum::TFunc(FunSignature {
args: Default::default(),
ret: self.primitives.none,
vars: Default::default(),
}));
// add to the definition list // add to the definition list
def_list.push( def_list.push(
Self::make_top_level_function_def(name.into(), self.primitives.none, resolver) Self::make_top_level_function_def(name.into(), self.primitives.none, resolver)
.into(), .into(),
); );
ty_list.push(ty);
ast_list.push(Some(ast)); ast_list.push(Some(ast));
// return // return
Ok((fun_name, DefinitionId(def_list.len() - 1), ty)) Ok((fun_name, DefinitionId(def_list.len() - 1)))
} }
_ => Err("only registrations of top level classes/functions are supprted".into()), _ => Err("only registrations of top level classes/functions are supprted".into()),
} }
} }
pub fn analyze_top_level_class_type_var(&mut self) -> Result<(), String> { /// step 1, analyze the type vars associated with top level class
fn analyze_top_level_class_type_var(&mut self) -> Result<(), String> {
let mut def_list = self.definition_list.write(); let mut def_list = self.definition_list.write();
let ty_list = self.ty_list.read();
let ast_list = self.ast_list.read(); let ast_list = self.ast_list.read();
let mut unifier = self.unifier.write(); let mut unifier = self.unifier.write();
for (def, ty, ast) in def_list for (class_def, class_ast) in def_list
.iter_mut() .iter_mut()
.zip(ty_list.iter())
.zip(ast_list.iter()) .zip(ast_list.iter())
.map(|((x, y), z)| (x, y, z)) .collect::<Vec<(&mut RwLock<TopLevelDef>, &Option<ast::Stmt<()>>)>>() {
.collect::<Vec<(&mut RwLock<TopLevelDef>, &Type, &Option<ast::Stmt<()>>)>>() // only deal with class def here
{ let (
unimplemented!() class_bases,
} class_def_type_vars,
unimplemented!() class_resolver
} ) = {
if let TopLevelDef::Class {
/// this should be called after all top level classes are registered, and type_vars,
/// will actually fill in those fields of the previous dummy one resolver,
pub fn analyze_top_level(&mut self) -> Result<(), String> {
let mut def_list = self.definition_list.write();
let ty_list = self.ty_list.read();
let ast_list = self.ast_list.read();
let mut unifier = self.unifier.write();
for (def, ty, ast) in def_list
.iter_mut()
.zip(ty_list.iter())
.zip(ast_list.iter())
.map(|((x, y), z)| (x, y, z))
.collect::<Vec<(&mut RwLock<TopLevelDef>, &Type, &Option<ast::Stmt<()>>)>>()
{
// only analyze those entries with ast, and class_method(whose ast in class def)
match ast {
Some(ast::Located{node: ast::StmtKind::ClassDef {
bases,
body,
name: class_name,
.. ..
}, .. }) => { } = class_def.get_mut() {
// get the mutable reference of the entry in the if let Some(ast::Located {node: ast::StmtKind::ClassDef {
// definition list, get the `TopLevelDef` bases,
let (
def_ancestors,
def_fields,
def_methods,
def_type_vars,
resolver,
) = if let TopLevelDef::Class {
object_id: _,
ancestors,
fields,
methods,
type_vars,
resolver: Some(resolver)
} = def.get_mut() {
(ancestors, fields, methods, type_vars, resolver.lock())
} else { unreachable!() };
// try to get mutable reference of the entry in the
// unification table, get the `TypeEnum`
let type_enum = unifier.get_ty(*ty);
let (
enum_params,
enum_fields
) = if let TypeEnum::TObj {
params,
fields,
.. ..
} = type_enum.borrow() { }, .. }) = class_ast {
(params, fields) (bases, type_vars, resolver)
} else { unreachable!() }; } else { unreachable!("must be both class") }
} else { continue }
};
// ancestors and typevars associate with the class are analyzed by looking let mut generic_occured = false;
// into the `bases` ast node for b in class_bases {
// `Generic` should only occur once, use this flag match &b.node {
let mut generic_occured = false; // analyze typevars bounded to the class,
// TODO: haven't check this yet // only support things like `class A(Generic[T, V])`,
let mut occured_type_var: HashSet<Type> = Default::default(); // things like `class A(Generic[T, V, ImportedModule.T])` is not supported
// TODO: haven't check this yet // i.e. only simple names are allowed in the subscript
let mut occured_base: HashSet<DefinitionId> = Default::default(); // should update the TopLevelDef::Class.typevars and the TypeEnum::TObj.params
for b in bases { ast::ExprKind::Subscript {value, slice, ..} if {
match &b.node { // can only be `Generic[...]` and this can only appear once
// analyze typevars bounded to the class, if let ast::ExprKind::Name { id, .. } = &value.node {
// only support things like `class A(Generic[T, V])`, if id == "Generic" {
// things like `class A(Generic[T, V, ImportedModule.T])` is not supported if !generic_occured {
// i.e. only simple names are allowed in the subscript generic_occured = true;
// should update the TopLevelDef::Class.typevars and the TypeEnum::TObj.params true
ast::ExprKind::Subscript {value, slice, ..} if { } else {
// can only be `Generic[...]` and this can only appear once return Err("Only single Generic[...] can be in bases".into())
if let ast::ExprKind::Name { id, .. } = &value.node { }
if id == "Generic" { } else { false }
if !generic_occured { } else { false }
generic_occured = true; } => {
true // if `class A(Generic[T, V, G])`
} else { if let ast::ExprKind::Tuple { elts, .. } = &slice.node {
return Err("Only single Generic[...] or Protocol[...] can be in bases".into()) // parse the type vars
} let type_vars = elts
} else { false }
} else { false }
} => {
match &slice.node {
// `class Foo(Generic[T, V, P]):` multiple element inside the subscript
ast::ExprKind::Tuple {elts, ..} => {
let tys = elts
.iter()
// here parse_type_annotation should be fine,
// since we only expect type vars, which is not relevant
// to the top-level parsing
.map(|x| resolver.parse_type_annotation(
&self.to_top_level_context(),
unifier.borrow_mut(),
&self.primitives,
x))
.collect::<Result<Vec<_>, _>>()?;
let ty_var_ids = tys
.iter()
.map(|t| {
let tmp = unifier.get_ty(*t);
// make sure it is type var
if let TypeEnum::TVar {id, ..} = tmp.as_ref() {
Ok(*id)
} else {
Err("Expect type variabls here".to_string())
}
})
.collect::<Result<Vec<_>, _>>()?;
// write to TypeEnum
for (id, ty) in ty_var_ids.iter().zip(tys.iter()) {
enum_params.borrow_mut().insert(*id, *ty);
}
// write to TopLevelDef
for ty in tys{
def_type_vars.push(ty)
}
},
// `class Foo(Generic[T]):`, only single element
_ => {
let ty = resolver.parse_type_annotation(
&self.to_top_level_context(),
unifier.borrow_mut(),
&self.primitives,
&slice
)?;
let ty_var_id = if let TypeEnum::TVar { id, .. } = unifier
.get_ty(ty)
.as_ref() { *id } else {
return Err("Expect type variabls here".to_string())
};
// write to TypeEnum
enum_params.borrow_mut().insert(ty_var_id, ty);
// write to TopLevelDef
def_type_vars.push(ty);
},
};
}
// analyze base classes, which is possible in
// other cases, we parse for the base class
// FIXME: calling parse_type_annotation here might cause some problem
// when the base class is parametrized `BaseClass[int, bool]`, since the
// analysis of type var of some class is not done yet.
// we can first only look at the name, and later check the
// parameter when others are done
// Or
// first get all the class' type var analyzed, and then
// analyze the base class
_ => {
let ty = resolver.parse_type_annotation(
&self.to_top_level_context(),
unifier.borrow_mut(),
&self.primitives,
b
)?;
let obj_def_id = if let TypeEnum::TObj { obj_id, .. } = unifier
.get_ty(ty)
.as_ref() {
*obj_id
} else {
return Err("Expect concrete classes/types here".into())
};
// write to TopLevelDef
def_ancestors.push(obj_def_id);
}
}
}
// class method and field are analyzed by
// looking into the class body ast node
// NOTE: should consider parents' method and fields(check re-def and add),
// but we do it later we go over these again after we finish analyze the
// fields/methods as declared in the ast
// method with same name should not occur twice, so use this
let defined_method: HashSet<String> = Default::default();
for stmt in body {
if let ast::StmtKind::FunctionDef {
name: func_name,
args,
body,
returns,
..
} = &stmt.node {
// build type enum, need FunSignature {args, vars, ret}
// args. Now only args with no default TODO: other kinds of args
let func_args = args.args
.iter() .iter()
.map(|x| -> Result<FuncArg, String> { .map(|e|
Ok(FuncArg { class_resolver
name: x.node.arg.clone(), .as_ref()
ty: resolver.parse_type_annotation( .unwrap()
.lock()
.parse_type_annotation(
&self.to_top_level_context(), &self.to_top_level_context(),
unifier.borrow_mut(), unifier.borrow_mut(),
&self.primitives, &self.primitives,
x e)
.node )
.annotation .collect::<Result<Vec<_>, _>>()?;
.as_ref()
.ok_or_else(|| "type annotations required for function parameters".to_string())? // check if all are unique type vars
)?, let mut occured_type_var_id: HashSet<u32> = HashSet::new();
default_value: None let all_unique_type_var = type_vars
})
})
.collect::<Result<Vec<FuncArg>, _>>()?;
// vars. find TypeVars used in the argument type annotation
let func_vars = func_args
.iter() .iter()
.filter_map(|FuncArg { ty, .. } | { .all(|x| {
if let TypeEnum::TVar { id, .. } = unifier.get_ty(*ty).as_ref() { let ty = unifier.get_ty(*x);
Some((*id, *ty)) if let TypeEnum::TVar {id, ..} = ty.as_ref() {
} else { None } occured_type_var_id.insert(*id)
}) } else { false }
.collect::<HashMap<u32, Type>>(); });
// return type
let func_ret = resolver if !all_unique_type_var { return Err("expect unique type variables".into()) }
.parse_type_annotation(
// add to TopLevelDef
class_def_type_vars.extend(type_vars);
// `class A(Generic[T])`
} else {
let ty =
class_resolver
.as_ref()
.unwrap()
.lock()
.parse_type_annotation(
&self.to_top_level_context(), &self.to_top_level_context(),
unifier.borrow_mut(), unifier.borrow_mut(),
&self.primitives, &self.primitives,
returns &slice
.as_ref()
.ok_or_else(|| "return type annotations required here".to_string())?
.as_ref(),
)?; )?;
// build the TypeEnum // check if it is type var
let func_type_sig = FunSignature { let is_type_var = matches!(
args: func_args, unifier.get_ty(ty).as_ref(),
vars: func_vars, &TypeEnum::TVar { .. }
ret: func_ret );
}; if !is_type_var { return Err("expect type variable here".into()) }
// write to the TypeEnum and Def_list (by replacing the ty with the new Type created above) // add to TopLevelDef
let func_name_mangled = Self::name_mangling(class_name.clone(), func_name); class_def_type_vars.push(ty);
let def_id = self.class_method_to_def_id.read()[&func_name_mangled];
unimplemented!();
if func_name == "__init__" {
// special for constructor, need to look into the fields
// TODO: look into the function body and see
}
} else {
// do nothing. we do not care about things like this?
// class A:
// a = 3
// b = [2, 3]
} }
} }
},
// top level function definition // if others, do nothing in this function
Some(ast::Located{node: ast::StmtKind::FunctionDef { _ => continue
name, }
args, }
body,
returns, };
Ok(())
}
/// step 2, base classes. Need to separate step1 and step2 for this reason:
/// `class B(Generic[T, V]);
/// class A(B[int, bool])`
/// if the type var associated with class `B` has not been handled properly,
/// the parse of type annotation of `B[int, bool]` will fail
fn analyze_top_level_class_bases(&mut self) -> Result<(), String> {
let mut def_list = self.definition_list.write();
let ast_list = self.ast_list.read();
let mut unifier = self.unifier.write();
for (class_def, class_ast) in def_list
.iter_mut()
.zip(ast_list.iter())
.collect::<Vec<(&mut RwLock<TopLevelDef>, &Option<ast::Stmt<()>>)>>() {
let (
class_bases,
class_ancestors,
class_resolver
) = {
if let TopLevelDef::Class {
ancestors,
resolver,
.. ..
}, .. }) => { } = class_def.get_mut() {
// TODO: if let Some(ast::Located {node: ast::StmtKind::ClassDef {
unimplemented!() bases,
..
}, .. }) = class_ast {
(bases, ancestors, resolver)
} else { unreachable!("must be both class") }
} else { continue }
};
for b in class_bases {
// type vars have already been handled, so skip on `Generic[...]`
if let ast::ExprKind::Subscript {value, ..} = &b.node {
if let ast::ExprKind::Name {id, ..} = &value.node {
if id == "Generic" { continue }
}
}
// get the def id of the base class
let base_ty = class_resolver.as_ref().unwrap().lock().parse_type_annotation(
&self.to_top_level_context(),
unifier.borrow_mut(),
&self.primitives,
b
)?;
let base_id =
if let TypeEnum::TObj {obj_id, ..} = unifier.get_ty(base_ty).as_ref() {
*obj_id
} else { return Err("expect concrete class/type to be base class".into()) };
// write to the class ancestors
class_ancestors.push(base_id);
}
};
Ok(())
}
/// step 3, class_fields
fn analyze_top_level_class_fields_methods(&mut self) -> Result<(), String> {
let mut def_list = self.definition_list.write();
let ast_list = self.ast_list.read();
let mut unifier = self.unifier.write();
let class_method_to_def_id = self.class_method_to_def_id.read();
let mut to_be_analyzed_class = self.to_be_analyzed_class.write();
while !to_be_analyzed_class.is_empty() {
let ind = to_be_analyzed_class.remove(0).0;
let (class_def, class_ast) = (
&mut def_list[ind], &ast_list[ind]
);
let (
class_name,
class_fields,
class_methods,
class_resolver,
class_body
) = {
if let TopLevelDef::Class {
resolver,
fields,
methods,
..
} = class_def.get_mut() {
if let Some(ast::Located {node: ast::StmtKind::ClassDef {
name,
body,
..
}, .. }) = class_ast {
(name, fields, methods, resolver, body)
} else { unreachable!("must be both class") }
} else { continue }
};
for b in class_body {
if let ast::StmtKind::FunctionDef {
args: func_args,
body: func_body,
name: func_name,
returns: func_returns,
..
} = &b.node {
// unwrap should not fail
let method_def_id =
class_method_to_def_id
.get(&Self::name_mangling(
class_name.into(),
func_name)
).unwrap();
let a = &def_list[method_def_id.0];
} else {
// what should we do with `class A: a = 3`?
continue
} }
// only expect class def and function def ast
_ => return Err("only expect function and class definitions to be submitted here to be analyzed".into())
} }
} }
Ok(()) Ok(())
}
fn analyze_top_level_inheritance(&mut self) -> Result<(), String> {
unimplemented!()
}
fn analyze_top_level_field_instantiation(&mut self) -> Result<(), String> {
unimplemented!()
} }
} }