1
0
Fork 0
mirror of https://github.com/ton-blockchain/ton synced 2025-03-09 15:40:10 +00:00

[Tolk] Nullable types T? and null safety

This commit introduces nullable types `T?` that are
distinct from non-nullable `T`.
Example: `int?` (int or null) and `int` are different now.
Previously, `null` could be assigned to any primitive type.
Now, it can be assigned only to `T?`.

A non-null assertion operator `!` was also introduced,
similar to `!` in TypeScript and `!!` in Kotlin.

If `int?` still occupies 1 stack slot, `(int,int)?` and
other nullable tensors occupy N+1 slots, the last for
"null precedence". `v == null` actually compares that slot.
Assigning `(int,int)` to `(int,int)?` implicitly creates
a null presence slot. Assigning `null` to `(int,int)?` widens
this null value to 3 slots. This is called "type transitioning".

All stdlib functions prototypes have been updated to reflect
whether they return/accept a nullable or a strict value.

This commit also contains refactoring from `const FunctionData*`
to `FunctionPtr` and similar.
This commit is contained in:
tolk-vm 2025-02-24 20:13:36 +03:00
parent 1389ff6789
commit f3e620f48c
No known key found for this signature in database
GPG key ID: 7905DD7FE0324B12
62 changed files with 2031 additions and 702 deletions

View file

@ -402,7 +402,7 @@ void CodeBlob::print(std::ostream& os, int flags) const {
std::vector<var_idx_t> CodeBlob::create_var(TypePtr var_type, SrcLocation loc, std::string name) {
std::vector<var_idx_t> ir_idx;
int stack_w = var_type->calc_width_on_stack();
int stack_w = var_type->get_width_on_stack();
ir_idx.reserve(stack_w);
if (const TypeDataTensor* t_tensor = var_type->try_as<TypeDataTensor>()) {
for (int i = 0; i < t_tensor->size(); ++i) {
@ -410,6 +410,10 @@ std::vector<var_idx_t> CodeBlob::create_var(TypePtr var_type, SrcLocation loc, s
std::vector<var_idx_t> nested = create_var(t_tensor->items[i], loc, std::move(sub_name));
ir_idx.insert(ir_idx.end(), nested.begin(), nested.end());
}
} else if (const TypeDataNullable* t_nullable = var_type->try_as<TypeDataNullable>(); t_nullable && stack_w != 1) {
std::string null_flag_name = name.empty() ? name : name + ".NNFlag";
ir_idx = create_var(t_nullable->inner, loc, std::move(name));
ir_idx.emplace_back(create_var(TypeDataBool::create(), loc, std::move(null_flag_name))[0]);
} else if (var_type != TypeDataVoid::create()) {
#ifdef TOLK_DEBUG
tolk_assert(stack_w == 1);

View file

@ -111,23 +111,16 @@ static void diagnose_addition_in_bitshift(SrcLocation loc, std::string_view bits
}
}
// replace (a == null) and similar to isNull(a) (call of a built-in function)
static AnyExprV maybe_replace_eq_null_with_isNull_call(V<ast_binary_operator> v) {
// replace (a == null) and similar to ast_is_null_check(a) (special AST vertex)
static AnyExprV maybe_replace_eq_null_with_isNull_check(V<ast_binary_operator> v) {
bool has_null = v->get_lhs()->type == ast_null_keyword || v->get_rhs()->type == ast_null_keyword;
bool replace = has_null && (v->tok == tok_eq || v->tok == tok_neq);
if (!replace) {
return v;
}
auto v_ident = createV<ast_identifier>(v->loc, "__isNull"); // built-in function
auto v_ref = createV<ast_reference>(v->loc, v_ident, nullptr);
AnyExprV v_null = v->get_lhs()->type == ast_null_keyword ? v->get_rhs() : v->get_lhs();
AnyExprV v_arg = createV<ast_argument>(v->loc, v_null, false);
AnyExprV v_isNull = createV<ast_function_call>(v->loc, v_ref, createV<ast_argument_list>(v->loc, {v_arg}));
if (v->tok == tok_neq) {
v_isNull = createV<ast_unary_operator>(v->loc, "!", tok_logical_not, v_isNull);
}
return v_isNull;
AnyExprV v_nullable = v->get_lhs()->type == ast_null_keyword ? v->get_rhs() : v->get_lhs();
return createV<ast_is_null_check>(v->loc, v_nullable, v->tok == tok_neq);
}
@ -372,16 +365,31 @@ static AnyExprV parse_expr100(Lexer& lex) {
}
}
// parse E(...) (left-to-right)
// parse E(...) and E! having parsed E already (left-to-right)
static AnyExprV parse_fun_call_postfix(Lexer& lex, AnyExprV lhs) {
while (true) {
if (lex.tok() == tok_oppar) {
lhs = createV<ast_function_call>(lhs->loc, lhs, parse_argument_list(lex));
} else if (lex.tok() == tok_logical_not) {
lex.next();
lhs = createV<ast_not_null_operator>(lhs->loc, lhs);
} else {
break;
}
}
return lhs;
}
// parse E(...) and E! (left-to-right)
static AnyExprV parse_expr90(Lexer& lex) {
AnyExprV res = parse_expr100(lex);
while (lex.tok() == tok_oppar) {
res = createV<ast_function_call>(res->loc, res, parse_argument_list(lex));
if (lex.tok() == tok_oppar || lex.tok() == tok_logical_not) {
res = parse_fun_call_postfix(lex, res);
}
return res;
}
// parse E.field and E.method(...) (left-to-right)
// parse E.field and E.method(...) and E.field! (left-to-right)
static AnyExprV parse_expr80(Lexer& lex) {
AnyExprV lhs = parse_expr90(lex);
while (lex.tok() == tok_dot) {
@ -402,8 +410,8 @@ static AnyExprV parse_expr80(Lexer& lex) {
lex.unexpected("method name");
}
lhs = createV<ast_dot_access>(loc, lhs, v_ident, v_instantiationTs);
while (lex.tok() == tok_oppar) {
lhs = createV<ast_function_call>(lex.cur_location(), lhs, parse_argument_list(lex));
if (lex.tok() == tok_oppar || lex.tok() == tok_logical_not) {
lhs = parse_fun_call_postfix(lex, lhs);
}
}
return lhs;
@ -491,7 +499,7 @@ static AnyExprV parse_expr15(Lexer& lex) {
AnyExprV rhs = parse_expr17(lex);
lhs = createV<ast_binary_operator>(loc, operator_name, t, lhs, rhs);
if (t == tok_eq || t == tok_neq) {
lhs = maybe_replace_eq_null_with_isNull_call(lhs->as<ast_binary_operator>());
lhs = maybe_replace_eq_null_with_isNull_check(lhs->as<ast_binary_operator>());
}
}
return lhs;

View file

@ -108,6 +108,8 @@ protected:
virtual AnyExprV replace(V<ast_binary_operator> v) { return replace_children(v); }
virtual AnyExprV replace(V<ast_ternary_operator> v) { return replace_children(v); }
virtual AnyExprV replace(V<ast_cast_as_operator> v) { return replace_children(v); }
virtual AnyExprV replace(V<ast_not_null_operator> v) { return replace_children(v); }
virtual AnyExprV replace(V<ast_is_null_check> v) { return replace_children(v); }
// statements
virtual AnyV replace(V<ast_empty_statement> v) { return replace_children(v); }
virtual AnyV replace(V<ast_sequence> v) { return replace_children(v); }
@ -144,6 +146,8 @@ protected:
case ast_binary_operator: return replace(v->as<ast_binary_operator>());
case ast_ternary_operator: return replace(v->as<ast_ternary_operator>());
case ast_cast_as_operator: return replace(v->as<ast_cast_as_operator>());
case ast_not_null_operator: return replace(v->as<ast_not_null_operator>());
case ast_is_null_check: return replace(v->as<ast_is_null_check>());
default:
throw UnexpectedASTNodeType(v, "ASTReplacerInFunctionBody::replace");
}
@ -174,20 +178,20 @@ protected:
}
public:
virtual bool should_visit_function(const FunctionData* fun_ref) = 0;
virtual bool should_visit_function(FunctionPtr fun_ref) = 0;
void start_replacing_in_function(const FunctionData* fun_ref, V<ast_function_declaration> v_function) {
void start_replacing_in_function(FunctionPtr fun_ref, V<ast_function_declaration> v_function) {
replace(v_function->get_body());
}
};
const std::vector<const FunctionData*>& get_all_not_builtin_functions();
const std::vector<FunctionPtr>& get_all_not_builtin_functions();
template<class BodyReplacerT>
void replace_ast_of_all_functions() {
BodyReplacerT visitor;
for (const FunctionData* fun_ref : get_all_not_builtin_functions()) {
for (FunctionPtr fun_ref : get_all_not_builtin_functions()) {
if (visitor.should_visit_function(fun_ref)) {
visitor.start_replacing_in_function(fun_ref, fun_ref->ast_root->as<ast_function_declaration>());
}

View file

@ -121,6 +121,12 @@ protected:
virtual V<ast_cast_as_operator> clone(V<ast_cast_as_operator> v) {
return createV<ast_cast_as_operator>(v->loc, clone(v->get_expr()), clone(v->cast_to_type));
}
virtual V<ast_not_null_operator> clone(V<ast_not_null_operator> v) {
return createV<ast_not_null_operator>(v->loc, clone(v->get_expr()));
}
virtual V<ast_is_null_check> clone(V<ast_is_null_check> v) {
return createV<ast_is_null_check>(v->loc, clone(v->get_expr()), v->is_negated);
}
// statements
@ -200,6 +206,8 @@ protected:
case ast_binary_operator: return clone(v->as<ast_binary_operator>());
case ast_ternary_operator: return clone(v->as<ast_ternary_operator>());
case ast_cast_as_operator: return clone(v->as<ast_cast_as_operator>());
case ast_not_null_operator: return clone(v->as<ast_not_null_operator>());
case ast_is_null_check: return clone(v->as<ast_is_null_check>());
default:
throw UnexpectedASTNodeType(v, "ASTReplicatorFunction::clone");
}

View file

@ -56,6 +56,8 @@ class ASTStringifier final : public ASTVisitor {
{ast_binary_operator, "ast_binary_operator"},
{ast_ternary_operator, "ast_ternary_operator"},
{ast_cast_as_operator, "ast_cast_as_operator"},
{ast_not_null_operator, "ast_not_null_operator"},
{ast_is_null_check, "ast_is_null_check"},
// statements
{ast_empty_statement, "ast_empty_statement"},
{ast_sequence, "ast_sequence"},
@ -268,6 +270,8 @@ public:
case ast_binary_operator: return handle_vertex(v->as<ast_binary_operator>());
case ast_ternary_operator: return handle_vertex(v->as<ast_ternary_operator>());
case ast_cast_as_operator: return handle_vertex(v->as<ast_cast_as_operator>());
case ast_not_null_operator: return handle_vertex(v->as<ast_not_null_operator>());
case ast_is_null_check: return handle_vertex(v->as<ast_is_null_check>());
// statements
case ast_empty_statement: return handle_vertex(v->as<ast_empty_statement>());
case ast_sequence: return handle_vertex(v->as<ast_sequence>());

View file

@ -109,6 +109,8 @@ protected:
virtual void visit(V<ast_binary_operator> v) { return visit_children(v); }
virtual void visit(V<ast_ternary_operator> v) { return visit_children(v); }
virtual void visit(V<ast_cast_as_operator> v) { return visit_children(v); }
virtual void visit(V<ast_not_null_operator> v) { return visit_children(v); }
virtual void visit(V<ast_is_null_check> v) { return visit_children(v); }
// statements
virtual void visit(V<ast_empty_statement> v) { return visit_children(v); }
virtual void visit(V<ast_sequence> v) { return visit_children(v); }
@ -146,6 +148,8 @@ protected:
case ast_binary_operator: return visit(v->as<ast_binary_operator>());
case ast_ternary_operator: return visit(v->as<ast_ternary_operator>());
case ast_cast_as_operator: return visit(v->as<ast_cast_as_operator>());
case ast_not_null_operator: return visit(v->as<ast_not_null_operator>());
case ast_is_null_check: return visit(v->as<ast_is_null_check>());
// statements
case ast_empty_statement: return visit(v->as<ast_empty_statement>());
case ast_sequence: return visit(v->as<ast_sequence>());
@ -167,20 +171,20 @@ protected:
}
public:
virtual bool should_visit_function(const FunctionData* fun_ref) = 0;
virtual bool should_visit_function(FunctionPtr fun_ref) = 0;
virtual void start_visiting_function(const FunctionData* fun_ref, V<ast_function_declaration> v_function) {
virtual void start_visiting_function(FunctionPtr fun_ref, V<ast_function_declaration> v_function) {
visit(v_function->get_body());
}
};
const std::vector<const FunctionData*>& get_all_not_builtin_functions();
const std::vector<FunctionPtr>& get_all_not_builtin_functions();
template<class BodyVisitorT>
void visit_ast_of_all_functions() {
BodyVisitorT visitor;
for (const FunctionData* fun_ref : get_all_not_builtin_functions()) {
for (FunctionPtr fun_ref : get_all_not_builtin_functions()) {
if (visitor.should_visit_function(fun_ref)) {
visitor.start_visiting_function(fun_ref, fun_ref->ast_root->as<ast_function_declaration>());
}

View file

@ -121,7 +121,7 @@ void Vertex<ast_reference>::assign_sym(const Symbol* sym) {
this->sym = sym;
}
void Vertex<ast_function_call>::assign_fun_ref(const FunctionData* fun_ref) {
void Vertex<ast_function_call>::assign_fun_ref(FunctionPtr fun_ref) {
this->fun_maybe = fun_ref;
}
@ -129,7 +129,7 @@ void Vertex<ast_cast_as_operator>::assign_resolved_type(TypePtr cast_to_type) {
this->cast_to_type = cast_to_type;
}
void Vertex<ast_global_var_declaration>::assign_var_ref(const GlobalVarData* var_ref) {
void Vertex<ast_global_var_declaration>::assign_var_ref(GlobalVarPtr var_ref) {
this->var_ref = var_ref;
}
@ -137,7 +137,7 @@ void Vertex<ast_global_var_declaration>::assign_resolved_type(TypePtr declared_t
this->declared_type = declared_type;
}
void Vertex<ast_constant_declaration>::assign_const_ref(const GlobalConstData* const_ref) {
void Vertex<ast_constant_declaration>::assign_const_ref(GlobalConstPtr const_ref) {
this->const_ref = const_ref;
}
@ -149,7 +149,7 @@ void Vertex<ast_instantiationT_item>::assign_resolved_type(TypePtr substituted_t
this->substituted_type = substituted_type;
}
void Vertex<ast_parameter>::assign_param_ref(const LocalVarData* param_ref) {
void Vertex<ast_parameter>::assign_param_ref(LocalVarPtr param_ref) {
this->param_ref = param_ref;
}
@ -157,23 +157,27 @@ void Vertex<ast_parameter>::assign_resolved_type(TypePtr declared_type) {
this->declared_type = declared_type;
}
void Vertex<ast_set_assign>::assign_fun_ref(const FunctionData* fun_ref) {
void Vertex<ast_set_assign>::assign_fun_ref(FunctionPtr fun_ref) {
this->fun_ref = fun_ref;
}
void Vertex<ast_unary_operator>::assign_fun_ref(const FunctionData* fun_ref) {
void Vertex<ast_unary_operator>::assign_fun_ref(FunctionPtr fun_ref) {
this->fun_ref = fun_ref;
}
void Vertex<ast_binary_operator>::assign_fun_ref(const FunctionData* fun_ref) {
void Vertex<ast_binary_operator>::assign_fun_ref(FunctionPtr fun_ref) {
this->fun_ref = fun_ref;
}
void Vertex<ast_is_null_check>::assign_is_negated(bool is_negated) {
this->is_negated = is_negated;
}
void Vertex<ast_dot_access>::assign_target(const DotTarget& target) {
this->target = target;
}
void Vertex<ast_function_declaration>::assign_fun_ref(const FunctionData* fun_ref) {
void Vertex<ast_function_declaration>::assign_fun_ref(FunctionPtr fun_ref) {
this->fun_ref = fun_ref;
}
@ -181,7 +185,7 @@ void Vertex<ast_function_declaration>::assign_resolved_type(TypePtr declared_ret
this->declared_return_type = declared_return_type;
}
void Vertex<ast_local_var_lhs>::assign_var_ref(const LocalVarData* var_ref) {
void Vertex<ast_local_var_lhs>::assign_var_ref(LocalVarPtr var_ref) {
this->var_ref = var_ref;
}

View file

@ -88,6 +88,8 @@ enum ASTNodeType {
ast_binary_operator,
ast_ternary_operator,
ast_cast_as_operator,
ast_not_null_operator,
ast_is_null_check,
// statements
ast_empty_statement,
ast_sequence,
@ -408,7 +410,7 @@ private:
V<ast_identifier> identifier;
public:
const LocalVarData* var_ref = nullptr; // filled on resolve identifiers; for `redef` points to declared above; for underscore, name is empty
LocalVarPtr var_ref = nullptr; // filled on resolve identifiers; for `redef` points to declared above; for underscore, name is empty
TypePtr declared_type; // not null for `var x: int = rhs`, otherwise nullptr
bool is_immutable; // declared via 'val', not 'var'
bool marked_as_redef; // var (existing_var redef, new_var: int) = ...
@ -417,7 +419,7 @@ public:
std::string_view get_name() const { return identifier->name; } // empty for underscore
Vertex* mutate() const { return const_cast<Vertex*>(this); }
void assign_var_ref(const LocalVarData* var_ref);
void assign_var_ref(LocalVarPtr var_ref);
void assign_resolved_type(TypePtr declared_type);
Vertex(SrcLocation loc, V<ast_identifier> identifier, TypePtr declared_type, bool is_immutable, bool marked_as_redef)
@ -530,12 +532,12 @@ private:
public:
typedef std::variant<
const FunctionData*, // for `t.tupleAt` target is `tupleAt` global function
FunctionPtr, // for `t.tupleAt` target is `tupleAt` global function
int // for `t.0` target is "indexed access" 0
> DotTarget;
DotTarget target = static_cast<FunctionData*>(nullptr); // filled at type inferring
bool is_target_fun_ref() const { return std::holds_alternative<const FunctionData*>(target); }
bool is_target_fun_ref() const { return std::holds_alternative<FunctionPtr>(target); }
bool is_target_indexed_access() const { return std::holds_alternative<int>(target); }
AnyExprV get_obj() const { return child; }
@ -560,7 +562,7 @@ template<>
// example: `getF()()` then callee is another func call (which type is TypeDataFunCallable)
// example: `obj.method()` then callee is dot access (resolved while type inferring)
struct Vertex<ast_function_call> final : ASTExprBinary {
const FunctionData* fun_maybe = nullptr; // filled while type inferring for `globalF()` / `obj.f()`; remains nullptr for `local_var()` / `getF()()`
FunctionPtr fun_maybe = nullptr; // filled while type inferring for `globalF()` / `obj.f()`; remains nullptr for `local_var()` / `getF()()`
AnyExprV get_callee() const { return lhs; }
bool is_dot_call() const { return lhs->type == ast_dot_access; }
@ -570,7 +572,7 @@ struct Vertex<ast_function_call> final : ASTExprBinary {
auto get_arg(int i) const { return rhs->as<ast_argument_list>()->get_arg(i); }
Vertex* mutate() const { return const_cast<Vertex*>(this); }
void assign_fun_ref(const FunctionData* fun_ref);
void assign_fun_ref(FunctionPtr fun_ref);
Vertex(SrcLocation loc, AnyExprV lhs_f, V<ast_argument_list> arguments)
: ASTExprBinary(ast_function_call, loc, lhs_f, arguments) {}
@ -603,7 +605,7 @@ template<>
// ast_set_assign represents assignment-and-set operation "lhs <op>= rhs"
// examples: `a += 4` / `b <<= c`
struct Vertex<ast_set_assign> final : ASTExprBinary {
const FunctionData* fun_ref = nullptr; // filled at type inferring, points to `_+_` built-in for +=
FunctionPtr fun_ref = nullptr; // filled at type inferring, points to `_+_` built-in for +=
std::string_view operator_name; // without equal sign, "+" for operator +=
TokenType tok; // tok_set_*
@ -611,7 +613,7 @@ struct Vertex<ast_set_assign> final : ASTExprBinary {
AnyExprV get_rhs() const { return rhs; }
Vertex* mutate() const { return const_cast<Vertex*>(this); }
void assign_fun_ref(const FunctionData* fun_ref);
void assign_fun_ref(FunctionPtr fun_ref);
Vertex(SrcLocation loc, std::string_view operator_name, TokenType tok, AnyExprV lhs, AnyExprV rhs)
: ASTExprBinary(ast_set_assign, loc, lhs, rhs)
@ -622,14 +624,14 @@ template<>
// ast_unary_operator is "some operator over one expression"
// examples: `-1` / `~found`
struct Vertex<ast_unary_operator> final : ASTExprUnary {
const FunctionData* fun_ref = nullptr; // filled at type inferring, points to some built-in function
FunctionPtr fun_ref = nullptr; // filled at type inferring, points to some built-in function
std::string_view operator_name;
TokenType tok;
AnyExprV get_rhs() const { return child; }
Vertex* mutate() const { return const_cast<Vertex*>(this); }
void assign_fun_ref(const FunctionData* fun_ref);
void assign_fun_ref(FunctionPtr fun_ref);
Vertex(SrcLocation loc, std::string_view operator_name, TokenType tok, AnyExprV rhs)
: ASTExprUnary(ast_unary_operator, loc, rhs)
@ -641,7 +643,7 @@ template<>
// examples: `a + b` / `x & true` / `(a, b) << g()`
// note, that `a = b` is NOT a binary operator, it's ast_assign, also `a += b`, it's ast_set_assign
struct Vertex<ast_binary_operator> final : ASTExprBinary {
const FunctionData* fun_ref = nullptr; // filled at type inferring, points to some built-in function
FunctionPtr fun_ref = nullptr; // filled at type inferring, points to some built-in function
std::string_view operator_name;
TokenType tok;
@ -649,7 +651,7 @@ struct Vertex<ast_binary_operator> final : ASTExprBinary {
AnyExprV get_rhs() const { return rhs; }
Vertex* mutate() const { return const_cast<Vertex*>(this); }
void assign_fun_ref(const FunctionData* fun_ref);
void assign_fun_ref(FunctionPtr fun_ref);
Vertex(SrcLocation loc, std::string_view operator_name, TokenType tok, AnyExprV lhs, AnyExprV rhs)
: ASTExprBinary(ast_binary_operator, loc, lhs, rhs)
@ -684,6 +686,32 @@ struct Vertex<ast_cast_as_operator> final : ASTExprUnary {
, cast_to_type(cast_to_type) {}
};
template<>
// ast_not_null_operator is non-null assertion: like TypeScript ! or Kotlin !!
// examples: `nullableInt!` / `getNullableBuilder()!`
struct Vertex<ast_not_null_operator> final : ASTExprUnary {
AnyExprV get_expr() const { return child; }
Vertex(SrcLocation loc, AnyExprV expr)
: ASTExprUnary(ast_not_null_operator, loc, expr) {}
};
template<>
// ast_is_null_check is an artificial vertex for "expr == null" / "expr != null" / same but null on the left
// it's created instead of a general binary expression to emphasize its purpose
struct Vertex<ast_is_null_check> final : ASTExprUnary {
bool is_negated;
AnyExprV get_expr() const { return child; }
Vertex* mutate() const { return const_cast<Vertex*>(this); }
void assign_is_negated(bool is_negated);
Vertex(SrcLocation loc, AnyExprV expr, bool is_negated)
: ASTExprUnary(ast_is_null_check, loc, expr)
, is_negated(is_negated) {}
};
//
// ---------------------------------------------------------
@ -892,7 +920,7 @@ template<>
// ast_parameter is a parameter of a function in its declaration
// example: `fun f(a: int, mutate b: slice)` has 2 parameters
struct Vertex<ast_parameter> final : ASTOtherLeaf {
const LocalVarData* param_ref = nullptr; // filled on resolve identifiers
LocalVarPtr param_ref = nullptr; // filled on resolve identifiers
std::string_view param_name;
TypePtr declared_type;
bool declared_as_mutate; // declared as `mutate param_name`
@ -900,7 +928,7 @@ struct Vertex<ast_parameter> final : ASTOtherLeaf {
bool is_underscore() const { return param_name.empty(); }
Vertex* mutate() const { return const_cast<Vertex*>(this); }
void assign_param_ref(const LocalVarData* param_ref);
void assign_param_ref(LocalVarPtr param_ref);
void assign_resolved_type(TypePtr declared_type);
Vertex(SrcLocation loc, std::string_view param_name, TypePtr declared_type, bool declared_as_mutate)
@ -951,7 +979,7 @@ struct Vertex<ast_function_declaration> final : ASTOtherVararg {
auto get_param(int i) const { return children.at(1)->as<ast_parameter_list>()->get_param(i); }
AnyV get_body() const { return children.at(2); } // ast_sequence / ast_asm_body
const FunctionData* fun_ref = nullptr; // filled after register
FunctionPtr fun_ref = nullptr; // filled after register
TypePtr declared_return_type; // filled at ast parsing; if unspecified (nullptr), means "auto infer"
V<ast_genericsT_list> genericsT_list; // for non-generics it's nullptr
td::RefInt256 method_id; // specified via @method_id annotation
@ -962,7 +990,7 @@ struct Vertex<ast_function_declaration> final : ASTOtherVararg {
bool is_builtin_function() const { return children.at(2)->type == ast_empty_statement; }
Vertex* mutate() const { return const_cast<Vertex*>(this); }
void assign_fun_ref(const FunctionData* fun_ref);
void assign_fun_ref(FunctionPtr fun_ref);
void assign_resolved_type(TypePtr declared_return_type);
Vertex(SrcLocation loc, V<ast_identifier> name_identifier, V<ast_parameter_list> parameters, AnyV body, TypePtr declared_return_type, V<ast_genericsT_list> genericsT_list, td::RefInt256 method_id, int flags)
@ -975,13 +1003,13 @@ template<>
// example: `global g: int;`
// note, that globals don't have default values, since there is no single "entrypoint" for a contract
struct Vertex<ast_global_var_declaration> final : ASTOtherVararg {
const GlobalVarData* var_ref = nullptr; // filled after register
GlobalVarPtr var_ref = nullptr; // filled after register
TypePtr declared_type; // filled always, typing globals is mandatory
auto get_identifier() const { return children.at(0)->as<ast_identifier>(); }
Vertex* mutate() const { return const_cast<Vertex*>(this); }
void assign_var_ref(const GlobalVarData* var_ref);
void assign_var_ref(GlobalVarPtr var_ref);
void assign_resolved_type(TypePtr declared_type);
Vertex(SrcLocation loc, V<ast_identifier> name_identifier, TypePtr declared_type)
@ -993,14 +1021,14 @@ template<>
// ast_constant_declaration is declaring a global constant, outside a function
// example: `const op = 0x123;`
struct Vertex<ast_constant_declaration> final : ASTOtherVararg {
const GlobalConstData* const_ref = nullptr; // filled after register
GlobalConstPtr const_ref = nullptr; // filled after register
TypePtr declared_type; // not null for `const op: int = ...`
auto get_identifier() const { return children.at(0)->as<ast_identifier>(); }
AnyExprV get_init_value() const { return child_as_expr(1); }
Vertex* mutate() const { return const_cast<Vertex*>(this); }
void assign_const_ref(const GlobalConstData* const_ref);
void assign_const_ref(GlobalConstPtr const_ref);
void assign_resolved_type(TypePtr declared_type);
Vertex(SrcLocation loc, V<ast_identifier> name_identifier, TypePtr declared_type, AnyExprV init_value)

View file

@ -348,9 +348,9 @@ bool Op::generate_code_step(Stack& stack) {
std::vector<VarDescr> args0, res;
int w_arg = 0;
for (const LocalVarData& param : f_sym->parameters) {
w_arg += param.declared_type->calc_width_on_stack();
w_arg += param.declared_type->get_width_on_stack();
}
int w_ret = f_sym->inferred_return_type->calc_width_on_stack();
int w_ret = f_sym->inferred_return_type->get_width_on_stack();
tolk_assert(w_ret >= 0 && w_arg >= 0);
for (int i = 0; i < w_ret; i++) {
res.emplace_back(0);

View file

@ -66,7 +66,7 @@ void CompilerSettings::parse_experimental_options_cmd_arg(const std::string& cmd
}
}
const std::vector<const FunctionData*>& get_all_not_builtin_functions() {
const std::vector<FunctionPtr>& get_all_not_builtin_functions() {
return G.all_functions;
}

View file

@ -95,10 +95,10 @@ struct CompilerState {
GlobalSymbolTable symtable;
PersistentHeapAllocator persistent_mem;
std::vector<const FunctionData*> all_functions; // all user-defined (not built-in) functions, with generic instantiations
std::vector<const FunctionData*> all_get_methods;
std::vector<const GlobalVarData*> all_global_vars;
std::vector<const GlobalConstData*> all_constants;
std::vector<FunctionPtr> all_functions; // all user-defined (not built-in) functions, with generic instantiations
std::vector<FunctionPtr> all_get_methods;
std::vector<GlobalVarPtr> all_global_vars;
std::vector<GlobalConstPtr> all_constants;
AllRegisteredSrcFiles all_src_files;
bool is_verbosity(int gt_eq) const { return settings.verbosity >= gt_eq; }

View file

@ -255,7 +255,7 @@ struct ConstantEvaluator {
if (!sym) {
v->error("undefined symbol `" + static_cast<std::string>(name) + "`");
}
const GlobalConstData* const_ref = sym->try_as<GlobalConstData>();
GlobalConstPtr const_ref = sym->try_as<GlobalConstPtr>();
if (!const_ref) {
v->error("symbol `" + static_cast<std::string>(name) + "` is not a constant");
}

View file

@ -32,6 +32,11 @@ struct FunctionData;
struct GlobalVarData;
struct GlobalConstData;
using LocalVarPtr = const LocalVarData*;
using FunctionPtr = const FunctionData*;
using GlobalVarPtr = const GlobalVarData*;
using GlobalConstPtr = const GlobalConstData*;
class TypeData;
using TypePtr = const TypeData*;

View file

@ -37,12 +37,38 @@ static TypePtr replace_genericT_with_deduced(TypePtr orig, const GenericsDeclara
if (idx == -1) {
throw Fatal("can not replace generic " + asT->nameT);
}
if (substitutionTs[idx] == nullptr) {
throw GenericDeduceError("can not deduce " + asT->nameT);
}
return substitutionTs[idx];
}
return child;
});
}
GenericSubstitutionsDeduceForCall::GenericSubstitutionsDeduceForCall(FunctionPtr fun_ref)
: fun_ref(fun_ref) {
substitutionTs.resize(fun_ref->genericTs->size()); // filled with nullptr (nothing deduced)
}
void GenericSubstitutionsDeduceForCall::provide_deducedT(const std::string& nameT, TypePtr deduced) {
if (deduced == TypeDataNullLiteral::create() || deduced->has_unknown_inside()) {
return; // just 'null' doesn't give sensible info
}
int idx = fun_ref->genericTs->find_nameT(nameT);
if (substitutionTs[idx] == nullptr) {
substitutionTs[idx] = deduced;
} else if (substitutionTs[idx] != deduced) {
throw GenericDeduceError(nameT + " is both " + substitutionTs[idx]->as_human_readable() + " and " + deduced->as_human_readable());
}
}
void GenericSubstitutionsDeduceForCall::provide_manually_specified(std::vector<TypePtr>&& substitutionTs) {
this->substitutionTs = std::move(substitutionTs);
this->manually_specified = true;
}
// purpose: having `f<T>(value: T)` and call `f(5)`, deduce T = int
// generally, there may be many generic Ts for declaration, and many arguments
// for every argument, `consider_next_condition()` is called
@ -51,71 +77,67 @@ static TypePtr replace_genericT_with_deduced(TypePtr orig, const GenericsDeclara
// - next condition: param_type = `T1`, arg_type = `int`, deduce T1 = int
// - next condition: param_type = `(T1, T2)`, arg_type = `(int, slice)`, deduce T1 = int, T2 = slice
// for call `f(6, cs, (8, cs))` T1 will be both `slice` and `int`, fired an error
class GenericSubstitutionsDeduceForFunctionCall final {
const FunctionData* fun_ref;
std::vector<TypePtr> substitutions;
void provideDeducedT(const std::string& nameT, TypePtr deduced) {
if (deduced == TypeDataNullLiteral::create() || deduced->has_unknown_inside()) {
return; // just 'null' doesn't give sensible info
void GenericSubstitutionsDeduceForCall::consider_next_condition(TypePtr param_type, TypePtr arg_type) {
if (const auto* asT = param_type->try_as<TypeDataGenericT>()) {
// `(arg: T)` called as `f([1, 2])` => T is [int, int]
provide_deducedT(asT->nameT, arg_type);
} else if (const auto* p_nullable = param_type->try_as<TypeDataNullable>()) {
// `arg: T?` called as `f(nullableInt)` => T is int
if (const auto* a_nullable = arg_type->try_as<TypeDataNullable>()) {
consider_next_condition(p_nullable->inner, a_nullable->inner);
}
int idx = fun_ref->genericTs->find_nameT(nameT);
if (substitutions[idx] == nullptr) {
substitutions[idx] = deduced;
} else if (substitutions[idx] != deduced) {
throw std::runtime_error(nameT + " is both " + substitutions[idx]->as_human_readable() + " and " + deduced->as_human_readable());
// `arg: T?` called as `f(int)` => T is int
else {
consider_next_condition(p_nullable->inner, arg_type);
}
}
public:
explicit GenericSubstitutionsDeduceForFunctionCall(const FunctionData* fun_ref)
: fun_ref(fun_ref) {
substitutions.resize(fun_ref->genericTs->size()); // filled with nullptr (nothing deduced)
}
void consider_next_condition(TypePtr param_type, TypePtr arg_type) {
if (const auto* asT = param_type->try_as<TypeDataGenericT>()) {
// `(arg: T)` called as `f([1, 2])` => T is [int, int]
provideDeducedT(asT->nameT, arg_type);
} else if (const auto* p_tensor = param_type->try_as<TypeDataTensor>()) {
// `arg: (int, T)` called as `f((5, cs))` => T is slice
if (const auto* a_tensor = arg_type->try_as<TypeDataTensor>(); a_tensor && a_tensor->size() == p_tensor->size()) {
for (int i = 0; i < a_tensor->size(); ++i) {
consider_next_condition(p_tensor->items[i], a_tensor->items[i]);
}
}
} else if (const auto* p_tuple = param_type->try_as<TypeDataTypedTuple>()) {
// `arg: [int, T]` called as `f([5, cs])` => T is slice
if (const auto* a_tuple = arg_type->try_as<TypeDataTypedTuple>(); a_tuple && a_tuple->size() == p_tuple->size()) {
for (int i = 0; i < a_tuple->size(); ++i) {
consider_next_condition(p_tuple->items[i], a_tuple->items[i]);
}
}
} else if (const auto* p_callable = param_type->try_as<TypeDataFunCallable>()) {
// `arg: fun(TArg) -> TResult` called as `f(calcTupleLen)` => TArg is tuple, TResult is int
if (const auto* a_callable = arg_type->try_as<TypeDataFunCallable>(); a_callable && a_callable->params_size() == p_callable->params_size()) {
for (int i = 0; i < a_callable->params_size(); ++i) {
consider_next_condition(p_callable->params_types[i], a_callable->params_types[i]);
}
consider_next_condition(p_callable->return_type, a_callable->return_type);
} else if (const auto* p_tensor = param_type->try_as<TypeDataTensor>()) {
// `arg: (int, T)` called as `f((5, cs))` => T is slice
if (const auto* a_tensor = arg_type->try_as<TypeDataTensor>(); a_tensor && a_tensor->size() == p_tensor->size()) {
for (int i = 0; i < a_tensor->size(); ++i) {
consider_next_condition(p_tensor->items[i], a_tensor->items[i]);
}
}
}
int get_first_not_deduced_idx() const {
for (int i = 0; i < static_cast<int>(substitutions.size()); ++i) {
if (substitutions[i] == nullptr) {
return i;
} else if (const auto* p_tuple = param_type->try_as<TypeDataTypedTuple>()) {
// `arg: [int, T]` called as `f([5, cs])` => T is slice
if (const auto* a_tuple = arg_type->try_as<TypeDataTypedTuple>(); a_tuple && a_tuple->size() == p_tuple->size()) {
for (int i = 0; i < a_tuple->size(); ++i) {
consider_next_condition(p_tuple->items[i], a_tuple->items[i]);
}
}
return -1;
} else if (const auto* p_callable = param_type->try_as<TypeDataFunCallable>()) {
// `arg: fun(TArg) -> TResult` called as `f(calcTupleLen)` => TArg is tuple, TResult is int
if (const auto* a_callable = arg_type->try_as<TypeDataFunCallable>(); a_callable && a_callable->params_size() == p_callable->params_size()) {
for (int i = 0; i < a_callable->params_size(); ++i) {
consider_next_condition(p_callable->params_types[i], a_callable->params_types[i]);
}
consider_next_condition(p_callable->return_type, a_callable->return_type);
}
}
}
std::vector<TypePtr> flush() {
return {std::move(substitutions)};
TypePtr GenericSubstitutionsDeduceForCall::replace_by_manually_specified(TypePtr param_type) const {
return replace_genericT_with_deduced(param_type, fun_ref->genericTs, substitutionTs);
}
TypePtr GenericSubstitutionsDeduceForCall::auto_deduce_from_argument(SrcLocation loc, TypePtr param_type, TypePtr arg_type) {
try {
if (!manually_specified) {
consider_next_condition(param_type, arg_type);
}
return replace_genericT_with_deduced(param_type, fun_ref->genericTs, substitutionTs);
} catch (const GenericDeduceError& ex) {
throw ParseError(loc, ex.message + " for generic function `" + fun_ref->as_human_readable() + "`; instantiate it manually with `" + fun_ref->name + "<...>()`");
}
};
}
int GenericSubstitutionsDeduceForCall::get_first_not_deduced_idx() const {
for (int i = 0; i < static_cast<int>(substitutionTs.size()); ++i) {
if (substitutionTs[i] == nullptr) {
return i;
}
}
return -1;
}
// clone the body of `f<T>` replacing T everywhere with a substitution
// before: `fun f<T>(v: T) { var cp: [T] = [v]; }`
@ -175,7 +197,7 @@ int GenericsDeclaration::find_nameT(std::string_view nameT) const {
// after creating a deep copy of `f<T>` like `f<int>`, its new and fresh body needs the previous pipeline to run
// for example, all local vars need to be registered as symbols, etc.
static void run_pipeline_for_instantiated_function(const FunctionData* inst_fun_ref) {
static void run_pipeline_for_instantiated_function(FunctionPtr inst_fun_ref) {
// these pipes are exactly the same as in tolk.cpp — all preceding (and including) type inferring
pipeline_resolve_identifiers_and_assign_symbols(inst_fun_ref);
pipeline_calculate_rvalue_lvalue(inst_fun_ref);
@ -198,34 +220,12 @@ std::string generate_instantiated_name(const std::string& orig_name, const std::
return name;
}
td::Result<std::vector<TypePtr>> deduce_substitutionTs_on_generic_func_call(const FunctionData* called_fun, std::vector<TypePtr>&& arg_types, TypePtr return_hint) {
try {
GenericSubstitutionsDeduceForFunctionCall deducing(called_fun);
for (const LocalVarData& param : called_fun->parameters) {
if (param.declared_type->has_genericT_inside() && param.param_idx < static_cast<int>(arg_types.size())) {
deducing.consider_next_condition(param.declared_type, arg_types[param.param_idx]);
}
}
int idx = deducing.get_first_not_deduced_idx();
if (idx != -1 && return_hint && called_fun->declared_return_type->has_genericT_inside()) {
deducing.consider_next_condition(called_fun->declared_return_type, return_hint);
idx = deducing.get_first_not_deduced_idx();
}
if (idx != -1) {
return td::Status::Error(td::Slice{"can not deduce " + called_fun->genericTs->get_nameT(idx)});
}
return deducing.flush();
} catch (const std::runtime_error& ex) {
return td::Status::Error(td::Slice{ex.what()});
}
}
const FunctionData* instantiate_generic_function(SrcLocation loc, const FunctionData* fun_ref, const std::string& inst_name, std::vector<TypePtr>&& substitutionTs) {
FunctionPtr instantiate_generic_function(SrcLocation loc, FunctionPtr fun_ref, const std::string& inst_name, std::vector<TypePtr>&& substitutionTs) {
tolk_assert(fun_ref->genericTs);
// if `f<int>` was earlier instantiated, return it
if (const auto* existing = lookup_global_symbol(inst_name)) {
const FunctionData* inst_ref = existing->try_as<FunctionData>();
FunctionPtr inst_ref = existing->try_as<FunctionPtr>();
tolk_assert(inst_ref);
return inst_ref;
}

View file

@ -57,8 +57,46 @@ struct GenericsInstantiation {
}
};
// this class helps to deduce Ts on the fly
// purpose: having `f<T>(value: T)` and call `f(5)`, deduce T = int
// while analyzing a call, arguments are handled one by one, by `auto_deduce_from_argument()`
// this class also handles manually specified substitutions like `f<int>(5)`
class GenericSubstitutionsDeduceForCall {
FunctionPtr fun_ref;
std::vector<TypePtr> substitutionTs;
bool manually_specified = false;
void provide_deducedT(const std::string& nameT, TypePtr deduced);
void consider_next_condition(TypePtr param_type, TypePtr arg_type);
public:
explicit GenericSubstitutionsDeduceForCall(FunctionPtr fun_ref);
bool is_manually_specified() const {
return manually_specified;
}
void provide_manually_specified(std::vector<TypePtr>&& substitutionTs);
TypePtr replace_by_manually_specified(TypePtr param_type) const;
TypePtr auto_deduce_from_argument(SrcLocation loc, TypePtr param_type, TypePtr arg_type);
int get_first_not_deduced_idx() const;
std::vector<TypePtr>&& flush() {
return std::move(substitutionTs);
}
};
struct GenericDeduceError final : std::exception {
std::string message;
explicit GenericDeduceError(std::string message)
: message(std::move(message)) { }
const char* what() const noexcept override {
return message.c_str();
}
};
std::string generate_instantiated_name(const std::string& orig_name, const std::vector<TypePtr>& substitutions);
td::Result<std::vector<TypePtr>> deduce_substitutionTs_on_generic_func_call(const FunctionData* called_fun, std::vector<TypePtr>&& arg_types, TypePtr return_hint);
const FunctionData* instantiate_generic_function(SrcLocation loc, const FunctionData* fun_ref, const std::string& inst_name, std::vector<TypePtr>&& substitutionTs);
FunctionPtr instantiate_generic_function(SrcLocation loc, FunctionPtr fun_ref, const std::string& inst_name, std::vector<TypePtr>&& substitutionTs);
} // namespace tolk

View file

@ -32,12 +32,22 @@
* So, if execution reaches this pass, the input is (almost) correct, and code generation should succeed.
* (previously, there was a check for one variable modified twice like `(t.0, t.0) = rhs`, but after changing
* execution order of assignment to "first lhs, then lhs", it was removed for several reasons)
*
* A noticeable property for IR generation is "target_type" used to extend/shrink stack.
* Example: `var a: (int,int)? = null`. This `null` has inferred_type "null literal", but target_type "nullable tensor",
* and when it's assigned, it's "expanded" from 1 stack slot to 3 (int + int + null flag).
* Example: `fun analyze(t: (int,int)?)` and a call `analyze((1,2))`. `(1,2)` is `(int,int)` (2 stack slots),
* and when passed to target (3 slots, one for null flag), this null flag is implicitly added (zero value).
* Example: `nullableInt!`; for `nullableInt` inferred_type is `int?`, and target_type is `int`
* (this doesn't lead to stack reorganization, but in case `nullableTensor!` does)
* (inferred_type of `nullableInt!` is `int`, and its target_type depends on its usage).
* The same mechanism will work for union types in the future.
*/
namespace tolk {
class LValContext;
std::vector<var_idx_t> pre_compile_expr(AnyExprV v, CodeBlob& code, LValContext* lval_ctx = nullptr);
std::vector<var_idx_t> pre_compile_expr(AnyExprV v, CodeBlob& code, TypePtr target_type = nullptr, LValContext* lval_ctx = nullptr);
std::vector<var_idx_t> pre_compile_symbol(SrcLocation loc, const Symbol* sym, CodeBlob& code, LValContext* lval_ctx);
void process_any_statement(AnyV v, CodeBlob& code);
@ -101,8 +111,8 @@ class LValContext {
// every global variable used as lvalue is registered here
// example: `globalInt = 9`, implicit var is created `$tmp = 9`, and `SetGlob "globalInt" $tmp` is done after
struct ModifiedGlobal {
const GlobalVarData* glob_ref;
std::vector<var_idx_t> lval_ir_idx; // typically 1, generally calc_width_on_stack() of global var (tensors)
GlobalVarPtr glob_ref;
std::vector<var_idx_t> lval_ir_idx; // typically 1, generally get_width_on_stack() of global var (tensors)
// for 1-slot globals int/cell/slice, assigning to them is just SETGLOB
// same for tensors, if they are fully rewritten in an expression: `gTensor = (5,6)`
@ -139,13 +149,13 @@ class LValContext {
void apply(CodeBlob& code, SrcLocation loc) const {
LValContext local_lval;
local_lval.enter_rval_inside_lval();
std::vector<var_idx_t> obj_ir_idx = pre_compile_expr(tensor_obj, code, &local_lval);
std::vector<var_idx_t> obj_ir_idx = pre_compile_expr(tensor_obj, code, nullptr, &local_lval);
const TypeDataTensor* t_tensor = tensor_obj->inferred_type->try_as<TypeDataTensor>();
tolk_assert(t_tensor);
int stack_width = t_tensor->items[index_at]->calc_width_on_stack();
int stack_width = t_tensor->items[index_at]->get_width_on_stack();
int stack_offset = 0;
for (int i = 0; i < index_at; ++i) {
stack_offset += t_tensor->items[i]->calc_width_on_stack();
stack_offset += t_tensor->items[i]->get_width_on_stack();
}
std::vector<var_idx_t> field_ir_idx = {obj_ir_idx.begin() + stack_offset, obj_ir_idx.begin() + stack_offset + stack_width};
tolk_assert(field_ir_idx.size() == lval_ir_idx.size());
@ -167,12 +177,12 @@ class LValContext {
void apply(CodeBlob& code, SrcLocation loc) const {
LValContext local_lval;
local_lval.enter_rval_inside_lval();
std::vector<var_idx_t> tuple_ir_idx = pre_compile_expr(tuple_obj, code, &local_lval);
std::vector<var_idx_t> tuple_ir_idx = pre_compile_expr(tuple_obj, code, nullptr, &local_lval);
std::vector<var_idx_t> index_ir_idx = code.create_tmp_var(TypeDataInt::create(), loc, "(tuple-idx)");
code.emplace_back(loc, Op::_IntConst, index_ir_idx, td::make_refint(index_at));
vars_modification_watcher.trigger_callbacks(tuple_ir_idx, loc);
const FunctionData* builtin_sym = lookup_global_symbol("tupleSetAt")->as<FunctionData>();
FunctionPtr builtin_sym = lookup_global_symbol("tupleSetAt")->try_as<FunctionPtr>();
code.emplace_back(loc, Op::_Call, std::vector{tuple_ir_idx}, std::vector{tuple_ir_idx[0], lval_ir_idx[0], index_ir_idx[0]}, builtin_sym);
local_lval.after_let(std::move(tuple_ir_idx), code, loc);
}
@ -195,7 +205,7 @@ public:
void exit_rval_inside_lval() { level_rval_inside_lval--; }
bool is_rval_inside_lval() const { return level_rval_inside_lval > 0; }
void capture_global_modification(const GlobalVarData* glob_ref, std::vector<var_idx_t> lval_ir_idx) {
void capture_global_modification(GlobalVarPtr glob_ref, std::vector<var_idx_t> lval_ir_idx) {
modifications.emplace_back(ModifiedGlobal{glob_ref, std::move(lval_ir_idx)});
}
@ -245,29 +255,39 @@ public:
}
};
// given `{some_expr}!`, return some_expr
static AnyExprV unwrap_not_null_operator(AnyExprV v) {
while (auto v_notnull = v->try_as<ast_not_null_operator>()) {
v = v_notnull->get_expr();
}
return v;
}
// given `{some_expr}.{i}`, check it for pattern `some_var.0` / `some_var.0.1` / etc.
// return some_var if satisfies (it may be a local or a global var, a tensor or a tuple)
// return nullptr otherwise: `f().0` / `(v = rhs).0` / `some_var.method().0` / etc.
static V<ast_reference> calc_sink_leftmost_obj(V<ast_dot_access> v) {
AnyExprV leftmost_obj = v->get_obj();
AnyExprV leftmost_obj = unwrap_not_null_operator(v->get_obj());
while (auto v_dot = leftmost_obj->try_as<ast_dot_access>()) {
if (!v_dot->is_target_indexed_access()) {
break;
}
leftmost_obj = v_dot->get_obj();
leftmost_obj = unwrap_not_null_operator(v_dot->get_obj());
}
return leftmost_obj->type == ast_reference ? leftmost_obj->as<ast_reference>() : nullptr;
}
static std::vector<std::vector<var_idx_t>> pre_compile_tensor_inner(CodeBlob& code, const std::vector<AnyExprV>& args,
LValContext* lval_ctx) {
const TypeDataTensor* tensor_target_type, LValContext* lval_ctx) {
const int n = static_cast<int>(args.size());
if (n == 0) { // just `()`
return {};
}
tolk_assert(!tensor_target_type || tensor_target_type->size() == n);
if (n == 1) { // just `(x)`: even if x is modified (e.g. `f(x=x+2)`), there are no next arguments
return {pre_compile_expr(args[0], code, lval_ctx)};
TypePtr child_target_type = tensor_target_type ? tensor_target_type->items[0] : nullptr;
return {pre_compile_expr(args[0], code, child_target_type, lval_ctx)};
}
// the purpose is to handle such cases: `return (x, x += y, x)`
@ -321,7 +341,8 @@ static std::vector<std::vector<var_idx_t>> pre_compile_tensor_inner(CodeBlob& co
WatchingVarList watched_vars(n);
for (int arg_idx = 0; arg_idx < n; ++arg_idx) {
std::vector<var_idx_t> vars_of_ith_arg = pre_compile_expr(args[arg_idx], code, lval_ctx);
TypePtr child_target_type = tensor_target_type ? tensor_target_type->items[arg_idx] : nullptr;
std::vector<var_idx_t> vars_of_ith_arg = pre_compile_expr(args[arg_idx], code, child_target_type, lval_ctx);
watched_vars.add_and_watch_modifications(std::move(vars_of_ith_arg), code);
}
return watched_vars.clear_and_stop_watching();
@ -329,7 +350,13 @@ static std::vector<std::vector<var_idx_t>> pre_compile_tensor_inner(CodeBlob& co
static std::vector<var_idx_t> pre_compile_tensor(CodeBlob& code, const std::vector<AnyExprV>& args,
LValContext* lval_ctx = nullptr) {
std::vector<std::vector<var_idx_t>> res_lists = pre_compile_tensor_inner(code, args, lval_ctx);
std::vector<TypePtr> types_list;
types_list.reserve(args.size());
for (AnyExprV item : args) {
types_list.push_back(item->inferred_type);
}
const TypeDataTensor* tensor_target_type = TypeDataTensor::create(std::move(types_list))->try_as<TypeDataTensor>();
std::vector<std::vector<var_idx_t>> res_lists = pre_compile_tensor_inner(code, args, tensor_target_type, lval_ctx);
std::vector<var_idx_t> res;
for (const std::vector<var_idx_t>& list : res_lists) {
res.insert(res.end(), list.cbegin(), list.cend());
@ -340,6 +367,7 @@ static std::vector<var_idx_t> pre_compile_tensor(CodeBlob& code, const std::vect
static std::vector<var_idx_t> pre_compile_let(CodeBlob& code, AnyExprV lhs, AnyExprV rhs, SrcLocation loc) {
// [lhs] = [rhs]; since type checking is ok, it's the same as "lhs = rhs"
if (lhs->type == ast_typed_tuple && rhs->type == ast_typed_tuple) {
// note: there are no type transitions (adding nullability flag, etc.), since only 1-slot elements allowed in tuples
LValContext local_lval;
std::vector<var_idx_t> left = pre_compile_tensor(code, lhs->as<ast_typed_tuple>()->get_items(), &local_lval);
vars_modification_watcher.trigger_callbacks(left, loc);
@ -355,7 +383,7 @@ static std::vector<var_idx_t> pre_compile_let(CodeBlob& code, AnyExprV lhs, AnyE
LValContext local_lval;
std::vector<var_idx_t> left = pre_compile_tensor(code, lhs->as<ast_typed_tuple>()->get_items(), &local_lval);
vars_modification_watcher.trigger_callbacks(left, loc);
std::vector<var_idx_t> right = pre_compile_expr(rhs, code);
std::vector<var_idx_t> right = pre_compile_expr(rhs, code, nullptr);
const TypeDataTypedTuple* inferred_tuple = rhs->inferred_type->try_as<TypeDataTypedTuple>();
std::vector<TypePtr> types_list = inferred_tuple->items;
std::vector<var_idx_t> rvect = code.create_tmp_var(TypeDataTensor::create(std::move(types_list)), rhs->loc, "(unpack-tuple)");
@ -365,25 +393,25 @@ static std::vector<var_idx_t> pre_compile_let(CodeBlob& code, AnyExprV lhs, AnyE
return right;
}
// small optimization: `var x = rhs` or `local_var = rhs` (90% cases), LValContext not needed actually
if (lhs->type == ast_local_var_lhs || (lhs->type == ast_reference && lhs->as<ast_reference>()->sym->try_as<LocalVarData>())) {
std::vector<var_idx_t> left = pre_compile_expr(lhs, code); // effectively, local_var->ir_idx
if (lhs->type == ast_local_var_lhs || (lhs->type == ast_reference && lhs->as<ast_reference>()->sym->try_as<LocalVarPtr>())) {
std::vector<var_idx_t> left = pre_compile_expr(lhs, code, nullptr); // effectively, local_var->ir_idx
vars_modification_watcher.trigger_callbacks(left, loc);
std::vector<var_idx_t> right = pre_compile_expr(rhs, code);
std::vector<var_idx_t> right = pre_compile_expr(rhs, code, lhs->inferred_type);
code.emplace_back(loc, Op::_Let, std::move(left), right);
return right;
}
// lhs = rhs
LValContext local_lval;
std::vector<var_idx_t> left = pre_compile_expr(lhs, code, &local_lval);
std::vector<var_idx_t> left = pre_compile_expr(lhs, code, nullptr, &local_lval);
vars_modification_watcher.trigger_callbacks(left, loc);
std::vector<var_idx_t> right = pre_compile_expr(rhs, code);
std::vector<var_idx_t> right = pre_compile_expr(rhs, code, lhs->inferred_type);
code.emplace_back(loc, Op::_Let, left, right);
local_lval.after_let(std::move(left), code, loc);
return right;
}
static std::vector<var_idx_t> gen_op_call(CodeBlob& code, TypePtr ret_type, SrcLocation loc,
std::vector<var_idx_t>&& args_vars, const FunctionData* fun_ref, const char* debug_desc) {
std::vector<var_idx_t>&& args_vars, FunctionPtr fun_ref, const char* debug_desc) {
std::vector<var_idx_t> rvect = code.create_tmp_var(ret_type, loc, debug_desc);
Op& op = code.emplace_back(loc, Op::_Call, rvect, std::move(args_vars), fun_ref);
if (!fun_ref->is_marked_as_pure()) {
@ -392,9 +420,161 @@ static std::vector<var_idx_t> gen_op_call(CodeBlob& code, TypePtr ret_type, SrcL
return rvect;
}
// "Transition to target (runtime) type" is the following process.
// Imagine `fun analyze(t: (int,int)?)` and a call `analyze((1,2))`.
// `(1,2)` (inferred_type) is 2 stack slots, but `t` (target_type) is 3 (one for null-flag).
// So, this null flag should be implicitly added (non-zero, since a variable is not null).
// Another example: `var t: (int, int)? = null`.
// `null` (inferred_type) is 1 stack slots, but target_type is 3, we should add 2 nulls.
// Another example: `var t1 = (1, null); var t2: (int, (int,int)?) = t1;`.
// Then t1's rvect is 2 vars (1 and null), but t1's `null` should be converted to 3 stack slots (resulting in 4 total).
// The same mechanism will work for union types in the future.
// Here rvect is a list of IR vars for inferred_type, probably patched due to target_type.
GNU_ATTRIBUTE_NOINLINE
static std::vector<var_idx_t> transition_expr_to_runtime_type_impl(std::vector<var_idx_t>&& rvect, CodeBlob& code, TypePtr original_type, TypePtr target_type, SrcLocation loc) {
// pass `T` to `T`
// could occur for passing tensor `(..., T, ...)` to `(..., T, ...)` while traversing tensor's components
if (target_type == original_type) {
return rvect;
}
int target_w = target_type->get_width_on_stack();
const TypeDataNullable* t_nullable = target_type->try_as<TypeDataNullable>();
const TypeDataNullable* o_nullable = original_type->try_as<TypeDataNullable>();
// pass `null` to `T?`
// for primitives like `int?`, no changes in rvect, null occupies the same TVM slot
// for tensors like `(int,int)?`, `null` is represented as N nulls + 1 null flag, insert N nulls
if (t_nullable && original_type == TypeDataNullLiteral::create()) {
tolk_assert(rvect.size() == 1);
if (target_w == 1 && !t_nullable->is_primitive_nullable()) { // `null` to `()?`
rvect = code.create_tmp_var(TypeDataInt::create(), loc, "(NNFlag)");
code.emplace_back(loc, Op::_IntConst, rvect, td::make_refint(0));
}
if (target_w > 1) {
FunctionPtr builtin_sym = lookup_global_symbol("__null")->try_as<FunctionPtr>();
rvect.reserve(target_w + 1);
for (int i = 1; i < target_w - 1; ++i) {
std::vector<var_idx_t> ith_null = code.create_tmp_var(TypeDataNullLiteral::create(), loc, "(null-literal)");
code.emplace_back(loc, Op::_Call, ith_null, std::vector<var_idx_t>{}, builtin_sym);
rvect.push_back(ith_null[0]);
}
std::vector<var_idx_t> null_flag_ir = code.create_tmp_var(TypeDataInt::create(), loc, "(NNFlag)");
var_idx_t null_flag_ir_idx = null_flag_ir[0];
code.emplace_back(loc, Op::_IntConst, std::move(null_flag_ir), td::make_refint(0));
rvect.push_back(null_flag_ir_idx);
}
return rvect;
}
// pass `T` to `T?`
// for primitives like `int?`, no changes in rvect: `int` and `int?` occupy the same TVM slot (null is represented as NULL TVM value)
// for passing `(int, int)` to `(int, int)?` / `(int, null)` to `(int, (int,int)?)?`, add a null flag equals to 0
if (t_nullable && !o_nullable) {
if (!t_nullable->is_primitive_nullable()) {
rvect = transition_expr_to_runtime_type_impl(std::move(rvect), code, original_type, t_nullable->inner, loc);
tolk_assert(target_w == static_cast<int>(rvect.size() + 1));
std::vector<var_idx_t> null_flag_ir = code.create_tmp_var(TypeDataInt::create(), loc, "(NNFlag)");
var_idx_t null_flag_ir_idx = null_flag_ir[0];
code.emplace_back(loc, Op::_IntConst, std::move(null_flag_ir), td::make_refint(-1));
rvect.push_back(null_flag_ir_idx);
}
return rvect;
}
// pass `T1?` to `T2?`
// for example, `int8?` to `int16?`
// transition inner types, leaving nullable flag unchanged for tensors
if (t_nullable && o_nullable) {
if (target_w > 1) {
var_idx_t null_flag_ir_idx = rvect.back();
rvect.pop_back();
rvect = transition_expr_to_runtime_type_impl(std::move(rvect), code, o_nullable->inner, t_nullable->inner, loc);
rvect.push_back(null_flag_ir_idx);
}
return rvect;
}
// pass `T?` to `null`
if (target_type == TypeDataNullLiteral::create() && original_type->can_rhs_be_assigned(target_type)) {
tolk_assert(o_nullable || original_type == TypeDataUnknown::create());
if (o_nullable && !o_nullable->is_primitive_nullable()) {
FunctionPtr builtin_sym = lookup_global_symbol("__null")->try_as<FunctionPtr>();
rvect = code.create_tmp_var(TypeDataNullLiteral::create(), loc, "(null-literal)");
code.emplace_back(loc, Op::_Call, rvect, std::vector<var_idx_t>{}, builtin_sym);
}
return rvect;
}
// pass `T?` to `T`
// it may occur due to operator `!` or smart cast
// for primitives like `int?`, no changes in rvect
// for passing `(int, int)?` to `(int, int)`, drop the null flag from the tail
if (!t_nullable && o_nullable) {
if (!o_nullable->is_primitive_nullable()) {
rvect.pop_back();
rvect = transition_expr_to_runtime_type_impl(std::move(rvect), code, original_type->try_as<TypeDataNullable>()->inner, target_type, loc);
}
return rvect;
}
// pass `bool` to `int`
// in code, it's done via `as` operator, like `boolVar as int`
// no changes in rvect, boolVar is guaranteed to be -1 or 0 at TVM level
if (target_type == TypeDataInt::create() && original_type == TypeDataBool::create()) {
return rvect;
}
// pass something to `unknown`
// probably, it comes from `_ = rhs`, type of `_` is unknown, it's target_type of rhs
// no changes in rvect
if (target_type == TypeDataUnknown::create()) {
return rvect;
}
// pass `unknown` to something
// probably, it comes from `arg` in exception, it's inferred as `unknown` and could be cast to any value
if (original_type == TypeDataUnknown::create()) {
tolk_assert(rvect.size() == 1);
return rvect;
}
// pass tensor to tensor, e.g. `(1, null)` to `(int, slice?)` / `(1, null)` to `(int, (int,int)?)`
// every element of rhs tensor should be transitioned
if (target_type->try_as<TypeDataTensor>() && original_type->try_as<TypeDataTensor>()) {
const TypeDataTensor* target_tensor = target_type->try_as<TypeDataTensor>();
const TypeDataTensor* inferred_tensor = original_type->try_as<TypeDataTensor>();
tolk_assert(target_tensor->size() == inferred_tensor->size());
tolk_assert(inferred_tensor->get_width_on_stack() == static_cast<int>(rvect.size()));
std::vector<var_idx_t> result_rvect;
result_rvect.reserve(target_w);
int stack_offset = 0;
for (int i = 0; i < inferred_tensor->size(); ++i) {
int ith_w = inferred_tensor->items[i]->get_width_on_stack();
std::vector<var_idx_t> rvect_i{rvect.begin() + stack_offset, rvect.begin() + stack_offset + ith_w};
std::vector<var_idx_t> result_i = transition_expr_to_runtime_type_impl(std::move(rvect_i), code, inferred_tensor->items[i], target_tensor->items[i], loc);
result_rvect.insert(result_rvect.end(), result_i.begin(), result_i.end());
stack_offset += ith_w;
}
return result_rvect;
}
// pass tuple to tuple, e.g. `[1, null]` to `[int, int?]` / `[1, null]` to `[int, [int?,int?]?]`
// to changes to rvect, since tuples contain only 1-slot elements
if (target_type->try_as<TypeDataTypedTuple>() && original_type->try_as<TypeDataTypedTuple>()) {
tolk_assert(target_type->get_width_on_stack() == original_type->get_width_on_stack());
return rvect;
}
throw Fatal("unhandled transition_expr_to_runtime_type_impl() combination");
}
// invoke the function above only if potentially needed to
// (if an expression is targeted to another type)
#ifndef TOLK_DEBUG
GNU_ATTRIBUTE_ALWAYS_INLINE
#endif
static std::vector<var_idx_t> transition_to_target_type(std::vector<var_idx_t>&& rvect, CodeBlob& code, TypePtr target_type, AnyExprV v) {
if (target_type != nullptr && target_type != v->inferred_type) {
rvect = transition_expr_to_runtime_type_impl(std::move(rvect), code, v->inferred_type, target_type, v->loc);
}
return rvect;
}
std::vector<var_idx_t> pre_compile_symbol(SrcLocation loc, const Symbol* sym, CodeBlob& code, LValContext* lval_ctx) {
if (const auto* glob_ref = sym->try_as<GlobalVarData>()) {
if (GlobalVarPtr glob_ref = sym->try_as<GlobalVarPtr>()) {
// handle `globalVar = rhs` / `mutate globalVar`
if (lval_ctx && !lval_ctx->is_rval_inside_lval()) {
std::vector<var_idx_t> lval_ir_idx = code.create_tmp_var(glob_ref->declared_type, loc, "(lval-glob)");
@ -410,7 +590,7 @@ std::vector<var_idx_t> pre_compile_symbol(SrcLocation loc, const Symbol* sym, Co
}
return local_ir_idx;
}
if (const auto* const_ref = sym->try_as<GlobalConstData>()) {
if (GlobalConstPtr const_ref = sym->try_as<GlobalConstPtr>()) {
if (const_ref->is_int_const()) {
std::vector<var_idx_t> rvect = code.create_tmp_var(TypeDataInt::create(), loc, "(glob-const)");
code.emplace_back(loc, Op::_IntConst, rvect, const_ref->as_int_const());
@ -421,44 +601,59 @@ std::vector<var_idx_t> pre_compile_symbol(SrcLocation loc, const Symbol* sym, Co
return rvect;
}
}
if (const auto* fun_ref = sym->try_as<FunctionData>()) {
if (FunctionPtr fun_ref = sym->try_as<FunctionPtr>()) {
std::vector<var_idx_t> rvect = code.create_tmp_var(fun_ref->inferred_full_type, loc, "(glob-var-fun)");
code.emplace_back(loc, Op::_GlobVar, rvect, std::vector<var_idx_t>{}, fun_ref);
return rvect;
}
if (const auto* var_ref = sym->try_as<LocalVarData>()) {
if (LocalVarPtr var_ref = sym->try_as<LocalVarPtr>()) {
#ifdef TOLK_DEBUG
tolk_assert(static_cast<int>(var_ref->ir_idx.size()) == var_ref->declared_type->calc_width_on_stack());
tolk_assert(static_cast<int>(var_ref->ir_idx.size()) == var_ref->declared_type->get_width_on_stack());
#endif
return var_ref->ir_idx;
}
throw Fatal("pre_compile_symbol");
}
static std::vector<var_idx_t> process_assignment(V<ast_assign> v, CodeBlob& code) {
static std::vector<var_idx_t> process_reference(V<ast_reference> v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) {
std::vector<var_idx_t> rvect = pre_compile_symbol(v->loc, v->sym, code, lval_ctx);
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_assignment(V<ast_assign> v, CodeBlob& code, TypePtr target_type) {
if (auto lhs_decl = v->get_lhs()->try_as<ast_local_vars_declaration>()) {
return pre_compile_let(code, lhs_decl->get_expr(), v->get_rhs(), v->loc);
std::vector<var_idx_t> rvect = pre_compile_let(code, lhs_decl->get_expr(), v->get_rhs(), v->loc);
return transition_to_target_type(std::move(rvect), code, target_type, v);
} else {
return pre_compile_let(code, v->get_lhs(), v->get_rhs(), v->loc);
std::vector<var_idx_t> rvect = pre_compile_let(code, v->get_lhs(), v->get_rhs(), v->loc);
// now rvect contains rhs IR vars constructed to fit lhs (for correct assignment, lhs type was target_type for rhs)
// but the type of `lhs = rhs` is RHS (see type inferring), so rvect now should fit rhs->inferred_type (= v->inferred_type)
// example: `t1 = t2 = null`, we're at `t2 = null`, earlier declared t1: `int?`, t2: `(int,int)?`
// currently "null" matches t2 (3 null slots), but type of this assignment is "plain null" (1 slot) assigned later to t1
rvect = transition_expr_to_runtime_type_impl(std::move(rvect), code, v->get_lhs()->inferred_type, v->inferred_type, v->loc);
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
}
static std::vector<var_idx_t> process_set_assign(V<ast_set_assign> v, CodeBlob& code) {
static std::vector<var_idx_t> process_set_assign(V<ast_set_assign> v, CodeBlob& code, TypePtr target_type) {
// for "a += b", emulate "a = a + b"
// seems not beautiful, but it works; probably, this transformation should be done at AST level in advance
std::string_view calc_operator = v->operator_name; // "+" for operator +=
auto v_apply = createV<ast_binary_operator>(v->loc, calc_operator, static_cast<TokenType>(v->tok - 1), v->get_lhs(), v->get_rhs());
v_apply->assign_inferred_type(v->inferred_type);
v_apply->assign_fun_ref(v->fun_ref);
return pre_compile_let(code, v->get_lhs(), v_apply, v->loc);
std::vector<var_idx_t> rvect = pre_compile_let(code, v->get_lhs(), v_apply, v->loc);
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_binary_operator(V<ast_binary_operator> v, CodeBlob& code) {
static std::vector<var_idx_t> process_binary_operator(V<ast_binary_operator> v, CodeBlob& code, TypePtr target_type) {
TokenType t = v->tok;
if (v->fun_ref) { // almost all operators, fun_ref was assigned at type inferring
std::vector<var_idx_t> args_vars = pre_compile_tensor(code, {v->get_lhs(), v->get_rhs()});
return gen_op_call(code, v->inferred_type, v->loc, std::move(args_vars), v->fun_ref, "(binary-op)");
std::vector<var_idx_t> rvect = gen_op_call(code, v->inferred_type, v->loc, std::move(args_vars), v->fun_ref, "(binary-op)");
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
if (t == tok_logical_and || t == tok_logical_or) {
// do the following transformations:
@ -470,43 +665,86 @@ static std::vector<var_idx_t> process_binary_operator(V<ast_binary_operator> v,
v_1->mutate()->assign_inferred_type(TypeDataInt::create());
auto v_b_ne_0 = createV<ast_binary_operator>(v->loc, "!=", tok_neq, v->get_rhs(), v_0);
v_b_ne_0->mutate()->assign_inferred_type(TypeDataInt::create());
v_b_ne_0->mutate()->assign_fun_ref(lookup_global_symbol("_!=_")->as<FunctionData>());
std::vector<var_idx_t> cond = pre_compile_expr(v->get_lhs(), code);
v_b_ne_0->mutate()->assign_fun_ref(lookup_global_symbol("_!=_")->try_as<FunctionPtr>());
std::vector<var_idx_t> cond = pre_compile_expr(v->get_lhs(), code, nullptr);
tolk_assert(cond.size() == 1);
std::vector<var_idx_t> rvect = code.create_tmp_var(v->inferred_type, v->loc, "(cond)");
std::vector<var_idx_t> rvect = code.create_tmp_var(v->inferred_type, v->loc, "(ternary)");
Op& if_op = code.emplace_back(v->loc, Op::_If, cond);
code.push_set_cur(if_op.block0);
code.emplace_back(v->loc, Op::_Let, rvect, pre_compile_expr(t == tok_logical_and ? v_b_ne_0 : v_1, code));
code.emplace_back(v->loc, Op::_Let, rvect, pre_compile_expr(t == tok_logical_and ? v_b_ne_0 : v_1, code, nullptr));
code.close_pop_cur(v->loc);
code.push_set_cur(if_op.block1);
code.emplace_back(v->loc, Op::_Let, rvect, pre_compile_expr(t == tok_logical_and ? v_0 : v_b_ne_0, code));
code.emplace_back(v->loc, Op::_Let, rvect, pre_compile_expr(t == tok_logical_and ? v_0 : v_b_ne_0, code, nullptr));
code.close_pop_cur(v->loc);
return rvect;
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
throw UnexpectedASTNodeType(v, "process_binary_operator");
}
static std::vector<var_idx_t> process_unary_operator(V<ast_unary_operator> v, CodeBlob& code) {
std::vector<var_idx_t> args_vars = pre_compile_tensor(code, {v->get_rhs()});
return gen_op_call(code, v->inferred_type, v->loc, std::move(args_vars), v->fun_ref, "(unary-op)");
static std::vector<var_idx_t> process_unary_operator(V<ast_unary_operator> v, CodeBlob& code, TypePtr target_type) {
std::vector<var_idx_t> rhs_vars = pre_compile_expr(v->get_rhs(), code, nullptr);
std::vector<var_idx_t> rvect = gen_op_call(code, v->inferred_type, v->loc, std::move(rhs_vars), v->fun_ref, "(unary-op)");
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_ternary_operator(V<ast_ternary_operator> v, CodeBlob& code) {
std::vector<var_idx_t> cond = pre_compile_expr(v->get_cond(), code);
static std::vector<var_idx_t> process_ternary_operator(V<ast_ternary_operator> v, CodeBlob& code, TypePtr target_type) {
std::vector<var_idx_t> cond = pre_compile_expr(v->get_cond(), code, nullptr);
tolk_assert(cond.size() == 1);
std::vector<var_idx_t> rvect = code.create_tmp_var(v->inferred_type, v->loc, "(cond)");
Op& if_op = code.emplace_back(v->loc, Op::_If, cond);
code.push_set_cur(if_op.block0);
code.emplace_back(v->get_when_true()->loc, Op::_Let, rvect, pre_compile_expr(v->get_when_true(), code));
code.emplace_back(v->get_when_true()->loc, Op::_Let, rvect, pre_compile_expr(v->get_when_true(), code, v->inferred_type));
code.close_pop_cur(v->get_when_true()->loc);
code.push_set_cur(if_op.block1);
code.emplace_back(v->get_when_false()->loc, Op::_Let, rvect, pre_compile_expr(v->get_when_false(), code));
code.emplace_back(v->get_when_false()->loc, Op::_Let, rvect, pre_compile_expr(v->get_when_false(), code, v->inferred_type));
code.close_pop_cur(v->get_when_false()->loc);
return rvect;
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_dot_access(V<ast_dot_access> v, CodeBlob& code, LValContext* lval_ctx) {
static std::vector<var_idx_t> process_cast_as_operator(V<ast_cast_as_operator> v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) {
TypePtr child_target_type = v->cast_to_type;
std::vector<var_idx_t> rvect = pre_compile_expr(v->get_expr(), code, child_target_type, lval_ctx);
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_not_null_operator(V<ast_not_null_operator> v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) {
TypePtr child_target_type = v->get_expr()->inferred_type;
if (const auto* as_nullable = child_target_type->try_as<TypeDataNullable>()) {
child_target_type = as_nullable->inner;
}
std::vector<var_idx_t> rvect = pre_compile_expr(v->get_expr(), code, child_target_type, lval_ctx);
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_is_null_check(V<ast_is_null_check> v, CodeBlob& code, TypePtr target_type) {
std::vector<var_idx_t> expr_ir_idx = pre_compile_expr(v->get_expr(), code, nullptr);
std::vector<var_idx_t> isnull_ir_idx = code.create_tmp_var(TypeDataBool::create(), v->loc, "(is-null)");
TypePtr expr_type = v->get_expr()->inferred_type;
if (const TypeDataNullable* t_nullable = expr_type->try_as<TypeDataNullable>()) {
if (!t_nullable->is_primitive_nullable()) {
std::vector<var_idx_t> zero_ir_idx = code.create_tmp_var(TypeDataInt::create(), v->loc, "(zero)");
code.emplace_back(v->loc, Op::_IntConst, zero_ir_idx, td::make_refint(0));
FunctionPtr eq_sym = lookup_global_symbol("_==_")->try_as<FunctionPtr>();
code.emplace_back(v->loc, Op::_Call, isnull_ir_idx, std::vector{expr_ir_idx.back(), zero_ir_idx[0]}, eq_sym);
} else {
FunctionPtr builtin_sym = lookup_global_symbol("__isNull")->try_as<FunctionPtr>();
code.emplace_back(v->loc, Op::_Call, isnull_ir_idx, expr_ir_idx, builtin_sym);
}
} else {
bool always_null = expr_type == TypeDataNullLiteral::create();
code.emplace_back(v->loc, Op::_IntConst, isnull_ir_idx, td::make_refint(always_null ? -1 : 0));
}
if (v->is_negated) {
FunctionPtr not_sym = lookup_global_symbol("!b_")->try_as<FunctionPtr>();
code.emplace_back(v->loc, Op::_Call, isnull_ir_idx, std::vector{isnull_ir_idx}, not_sym);
}
return transition_to_target_type(std::move(isnull_ir_idx), code, target_type, v);
}
static std::vector<var_idx_t> process_dot_access(V<ast_dot_access> v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) {
// it's NOT a method call `t.tupleSize()` (since such cases are handled by process_function_call)
// it's `t.0`, `getUser().id`, and `t.tupleSize` (as a reference, not as a call)
if (!v->is_target_fun_ref()) {
@ -516,20 +754,21 @@ static std::vector<var_idx_t> process_dot_access(V<ast_dot_access> v, CodeBlob&
if (const auto* t_tensor = obj_type->try_as<TypeDataTensor>()) {
// handle `tensorVar.0 = rhs` if tensors is a global, special case, then the global will be read on demand
if (lval_ctx && !lval_ctx->is_rval_inside_lval()) {
if (auto sink = calc_sink_leftmost_obj(v); sink && sink->sym->try_as<GlobalVarData>()) {
if (auto sink = calc_sink_leftmost_obj(v); sink && sink->sym->try_as<GlobalVarPtr>()) {
std::vector<var_idx_t> lval_ir_idx = code.create_tmp_var(v->inferred_type, v->loc, "(lval-global-tensor)");
lval_ctx->capture_field_of_global_modification(v->get_obj(), index_at, lval_ir_idx);
return lval_ir_idx;
}
}
// since a tensor of N elems are N vars on a stack actually, calculate offset
std::vector<var_idx_t> lhs_vars = pre_compile_expr(v->get_obj(), code, lval_ctx);
int stack_width = t_tensor->items[index_at]->calc_width_on_stack();
std::vector<var_idx_t> lhs_vars = pre_compile_expr(v->get_obj(), code, nullptr, lval_ctx);
int stack_width = t_tensor->items[index_at]->get_width_on_stack();
int stack_offset = 0;
for (int i = 0; i < index_at; ++i) {
stack_offset += t_tensor->items[i]->calc_width_on_stack();
stack_offset += t_tensor->items[i]->get_width_on_stack();
}
return {lhs_vars.begin() + stack_offset, lhs_vars.begin() + stack_offset + stack_width};
std::vector<var_idx_t> rvect{lhs_vars.begin() + stack_offset, lhs_vars.begin() + stack_offset + stack_width};
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
// `tupleVar.0`
if (obj_type->try_as<TypeDataTypedTuple>() || obj_type->try_as<TypeDataTuple>()) {
@ -545,40 +784,52 @@ static std::vector<var_idx_t> process_dot_access(V<ast_dot_access> v, CodeBlob&
code.emplace_back(v->loc, Op::_IntConst, index_ir_idx, td::make_refint(index_at));
std::vector<var_idx_t> field_ir_idx = code.create_tmp_var(v->inferred_type, v->loc, "(tuple-field)");
tolk_assert(tuple_ir_idx.size() == 1 && field_ir_idx.size() == 1); // tuples contain only 1-slot values
const FunctionData* builtin_sym = lookup_global_symbol("tupleAt")->as<FunctionData>();
FunctionPtr builtin_sym = lookup_global_symbol("tupleAt")->try_as<FunctionPtr>();
code.emplace_back(v->loc, Op::_Call, field_ir_idx, std::vector{tuple_ir_idx[0], index_ir_idx[0]}, builtin_sym);
if (lval_ctx && calc_sink_leftmost_obj(v)) { // `tupleVar.0.1 = rhs`, then `tupleVar.0` is rval inside lval
lval_ctx->capture_tuple_index_modification(v->get_obj(), index_at, field_ir_idx);
}
return field_ir_idx;
// like tensor index, `tupleVar.1` also might be smart cast, for example we're in `if (tupleVar.1 != null)`
// but since tuple's elements are only 1-slot width (no tensors and unions), no stack transformations required
return transition_to_target_type(std::move(field_ir_idx), code, target_type, v);
}
tolk_assert(false);
}
// okay, v->target refs a function, like `obj.method`, filled at type inferring
// (currently, nothing except a global function can be referenced, no object-scope methods exist)
const FunctionData* fun_ref = std::get<const FunctionData*>(v->target);
FunctionPtr fun_ref = std::get<FunctionPtr>(v->target);
tolk_assert(fun_ref);
return pre_compile_symbol(v->loc, fun_ref, code, lval_ctx);
std::vector<var_idx_t> rvect = pre_compile_symbol(v->loc, fun_ref, code, lval_ctx);
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_function_call(V<ast_function_call> v, CodeBlob& code) {
static std::vector<var_idx_t> process_function_call(V<ast_function_call> v, CodeBlob& code, TypePtr target_type) {
// v is `globalF(args)` / `globalF<int>(args)` / `obj.method(args)` / `local_var(args)` / `getF()(args)`
const FunctionData* fun_ref = v->fun_maybe;
FunctionPtr fun_ref = v->fun_maybe;
if (!fun_ref) {
// it's `local_var(args)`, treat args like a tensor:
// 1) when variables are modified like `local_var(x, x += 2, x)`, regular mechanism of watching automatically works
// 2) when `null` is passed to `(int, int)?`, or any other type transitions, it automatically works
std::vector<AnyExprV> args;
args.reserve(v->get_num_args());
for (int i = 0; i < v->get_num_args(); ++i) {
args.push_back(v->get_arg(i)->get_expr());
}
std::vector<var_idx_t> args_vars = pre_compile_tensor(code, args);
std::vector<var_idx_t> tfunc = pre_compile_expr(v->get_callee(), code);
std::vector<TypePtr> params_types = v->get_callee()->inferred_type->try_as<TypeDataFunCallable>()->params_types;
const TypeDataTensor* tensor_tt = TypeDataTensor::create(std::move(params_types))->try_as<TypeDataTensor>();
std::vector<std::vector<var_idx_t>> vars_per_arg = pre_compile_tensor_inner(code, args, tensor_tt, nullptr);
std::vector<var_idx_t> args_vars;
for (const std::vector<var_idx_t>& list : vars_per_arg) {
args_vars.insert(args_vars.end(), list.cbegin(), list.cend());
}
std::vector<var_idx_t> tfunc = pre_compile_expr(v->get_callee(), code, nullptr);
tolk_assert(tfunc.size() == 1);
args_vars.push_back(tfunc[0]);
std::vector<var_idx_t> rvect = code.create_tmp_var(v->inferred_type, v->loc, "(call-ind)");
Op& op = code.emplace_back(v->loc, Op::_CallInd, rvect, std::move(args_vars));
op.set_impure_flag();
return rvect;
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
int delta_self = v->is_dot_call();
@ -595,7 +846,11 @@ static std::vector<var_idx_t> process_function_call(V<ast_function_call> v, Code
for (int i = 0; i < v->get_num_args(); ++i) {
args.push_back(v->get_arg(i)->get_expr());
}
std::vector<std::vector<var_idx_t>> vars_per_arg = pre_compile_tensor_inner(code, args, nullptr);
// the purpose of tensor_tt ("tensor target type") is to transition `null` to `(int, int)?` and so on
// the purpose of calling `pre_compile_tensor_inner` is to have 0-th IR vars to handle return self
std::vector<TypePtr> params_types = fun_ref->inferred_full_type->try_as<TypeDataFunCallable>()->params_types;
const TypeDataTensor* tensor_tt = TypeDataTensor::create(std::move(params_types))->try_as<TypeDataTensor>();
std::vector<std::vector<var_idx_t>> vars_per_arg = pre_compile_tensor_inner(code, args, tensor_tt, nullptr);
TypePtr op_call_type = v->inferred_type;
TypePtr real_ret_type = v->inferred_type;
@ -609,7 +864,7 @@ static std::vector<var_idx_t> process_function_call(V<ast_function_call> v, Code
std::vector<TypePtr> types_list;
for (int i = 0; i < delta_self + v->get_num_args(); ++i) {
if (fun_ref->parameters[i].is_mutate_parameter()) {
types_list.push_back(args[i]->inferred_type);
types_list.push_back(fun_ref->parameters[i].declared_type);
}
}
types_list.push_back(real_ret_type);
@ -630,7 +885,7 @@ static std::vector<var_idx_t> process_function_call(V<ast_function_call> v, Code
AnyExprV arg_i = obj_leftmost && i == 0 ? obj_leftmost : args[i];
tolk_assert(arg_i->is_lvalue || i == 0);
if (arg_i->is_lvalue) {
std::vector<var_idx_t> ith_var_idx = pre_compile_expr(arg_i, code, &local_lval);
std::vector<var_idx_t> ith_var_idx = pre_compile_expr(arg_i, code, nullptr, &local_lval);
left.insert(left.end(), ith_var_idx.begin(), ith_var_idx.end());
} else {
left.insert(left.end(), vars_per_arg[0].begin(), vars_per_arg[0].end());
@ -647,36 +902,39 @@ static std::vector<var_idx_t> process_function_call(V<ast_function_call> v, Code
if (obj_leftmost && fun_ref->does_return_self()) {
if (obj_leftmost->is_lvalue) { // to handle if obj is global var, potentially re-assigned inside a chain
rvect_apply = pre_compile_expr(obj_leftmost, code);
rvect_apply = pre_compile_expr(obj_leftmost, code, nullptr);
} else { // temporary object, not lvalue, pre_compile_expr
rvect_apply = vars_per_arg[0];
}
}
return rvect_apply;
return transition_to_target_type(std::move(rvect_apply), code, target_type, v);
}
static std::vector<var_idx_t> process_tensor(V<ast_tensor> v, CodeBlob& code, LValContext* lval_ctx) {
return pre_compile_tensor(code, v->get_items(), lval_ctx);
static std::vector<var_idx_t> process_tensor(V<ast_tensor> v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) {
// tensor is compiled "as is", for example `(1, null)` occupies 2 slots
// and if assigned/passed to something other, like `(int, (int,int)?)`, a whole tensor is transitioned, it works
std::vector<var_idx_t> rvect = pre_compile_tensor(code, v->get_items(), lval_ctx);
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_typed_tuple(V<ast_typed_tuple> v, CodeBlob& code, LValContext* lval_ctx) {
static std::vector<var_idx_t> process_typed_tuple(V<ast_typed_tuple> v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) {
if (lval_ctx) { // todo some time, make "var (a, [b,c]) = (1, [2,3])" work
v->error("[...] can not be used as lvalue here");
}
std::vector<var_idx_t> left = code.create_tmp_var(v->inferred_type, v->loc, "(pack-tuple)");
std::vector<var_idx_t> right = pre_compile_tensor(code, v->get_items(), lval_ctx);
code.emplace_back(v->loc, Op::_Tuple, left, std::move(right));
return left;
return transition_to_target_type(std::move(left), code, target_type, v);
}
static std::vector<var_idx_t> process_int_const(V<ast_int_const> v, CodeBlob& code) {
static std::vector<var_idx_t> process_int_const(V<ast_int_const> v, CodeBlob& code, TypePtr target_type) {
std::vector<var_idx_t> rvect = code.create_tmp_var(v->inferred_type, v->loc, "(int-const)");
code.emplace_back(v->loc, Op::_IntConst, rvect, v->intval);
return rvect;
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_string_const(V<ast_string_const> v, CodeBlob& code) {
static std::vector<var_idx_t> process_string_const(V<ast_string_const> v, CodeBlob& code, TypePtr target_type) {
ConstantValue value = eval_const_init_value(v);
std::vector<var_idx_t> rvect = code.create_tmp_var(v->inferred_type, v->loc, "(str-const)");
if (value.is_int()) {
@ -684,27 +942,31 @@ static std::vector<var_idx_t> process_string_const(V<ast_string_const> v, CodeBl
} else {
code.emplace_back(v->loc, Op::_SliceConst, rvect, value.as_slice());
}
return rvect;
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_bool_const(V<ast_bool_const> v, CodeBlob& code) {
const FunctionData* builtin_sym = lookup_global_symbol(v->bool_val ? "__true" : "__false")->as<FunctionData>();
return gen_op_call(code, v->inferred_type, v->loc, {}, builtin_sym, "(bool-const)");
static std::vector<var_idx_t> process_bool_const(V<ast_bool_const> v, CodeBlob& code, TypePtr target_type) {
FunctionPtr builtin_sym = lookup_global_symbol(v->bool_val ? "__true" : "__false")->try_as<FunctionPtr>();
std::vector<var_idx_t> rvect = gen_op_call(code, v->inferred_type, v->loc, {}, builtin_sym, "(bool-const)");
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_null_keyword(V<ast_null_keyword> v, CodeBlob& code) {
const FunctionData* builtin_sym = lookup_global_symbol("__null")->as<FunctionData>();
return gen_op_call(code, v->inferred_type, v->loc, {}, builtin_sym, "(null-literal)");
static std::vector<var_idx_t> process_null_keyword(V<ast_null_keyword> v, CodeBlob& code, TypePtr target_type) {
FunctionPtr builtin_sym = lookup_global_symbol("__null")->try_as<FunctionPtr>();
std::vector<var_idx_t> rvect = gen_op_call(code, v->inferred_type, v->loc, {}, builtin_sym, "(null-literal)");
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_local_var(V<ast_local_var_lhs> v, CodeBlob& code) {
static std::vector<var_idx_t> process_local_var(V<ast_local_var_lhs> v, CodeBlob& code, TypePtr target_type) {
if (v->marked_as_redef) {
return pre_compile_symbol(v->loc, v->var_ref, code, nullptr);
std::vector<var_idx_t> rvect = pre_compile_symbol(v->loc, v->var_ref, code, nullptr);
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
tolk_assert(v->var_ref->ir_idx.empty());
v->var_ref->mutate()->assign_ir_idx(code.create_var(v->inferred_type, v->loc, v->var_ref->name));
return v->var_ref->ir_idx;
std::vector<var_idx_t> rvect = v->var_ref->ir_idx;
return transition_to_target_type(std::move(rvect), code, target_type, v);
}
static std::vector<var_idx_t> process_local_vars_declaration(V<ast_local_vars_declaration>, CodeBlob&) {
@ -718,42 +980,46 @@ static std::vector<var_idx_t> process_underscore(V<ast_underscore> v, CodeBlob&
return code.create_tmp_var(v->inferred_type, v->loc, "(underscore)");
}
std::vector<var_idx_t> pre_compile_expr(AnyExprV v, CodeBlob& code, LValContext* lval_ctx) {
std::vector<var_idx_t> pre_compile_expr(AnyExprV v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) {
switch (v->type) {
case ast_reference:
return pre_compile_symbol(v->loc, v->as<ast_reference>()->sym, code, lval_ctx);
return process_reference(v->as<ast_reference>(), code, target_type, lval_ctx);
case ast_assign:
return process_assignment(v->as<ast_assign>(), code);
return process_assignment(v->as<ast_assign>(), code, target_type);
case ast_set_assign:
return process_set_assign(v->as<ast_set_assign>(), code);
return process_set_assign(v->as<ast_set_assign>(), code, target_type);
case ast_binary_operator:
return process_binary_operator(v->as<ast_binary_operator>(), code);
return process_binary_operator(v->as<ast_binary_operator>(), code, target_type);
case ast_unary_operator:
return process_unary_operator(v->as<ast_unary_operator>(), code);
return process_unary_operator(v->as<ast_unary_operator>(), code, target_type);
case ast_ternary_operator:
return process_ternary_operator(v->as<ast_ternary_operator>(), code);
return process_ternary_operator(v->as<ast_ternary_operator>(), code, target_type);
case ast_cast_as_operator:
return pre_compile_expr(v->as<ast_cast_as_operator>()->get_expr(), code, lval_ctx);
return process_cast_as_operator(v->as<ast_cast_as_operator>(), code, target_type, lval_ctx);
case ast_not_null_operator:
return process_not_null_operator(v->as<ast_not_null_operator>(), code, target_type, lval_ctx);
case ast_is_null_check:
return process_is_null_check(v->as<ast_is_null_check>(), code, target_type);
case ast_dot_access:
return process_dot_access(v->as<ast_dot_access>(), code, lval_ctx);
return process_dot_access(v->as<ast_dot_access>(), code, target_type, lval_ctx);
case ast_function_call:
return process_function_call(v->as<ast_function_call>(), code);
return process_function_call(v->as<ast_function_call>(), code, target_type);
case ast_parenthesized_expression:
return pre_compile_expr(v->as<ast_parenthesized_expression>()->get_expr(), code, lval_ctx);
return pre_compile_expr(v->as<ast_parenthesized_expression>()->get_expr(), code, target_type, lval_ctx);
case ast_tensor:
return process_tensor(v->as<ast_tensor>(), code, lval_ctx);
return process_tensor(v->as<ast_tensor>(), code, target_type, lval_ctx);
case ast_typed_tuple:
return process_typed_tuple(v->as<ast_typed_tuple>(), code, lval_ctx);
return process_typed_tuple(v->as<ast_typed_tuple>(), code, target_type, lval_ctx);
case ast_int_const:
return process_int_const(v->as<ast_int_const>(), code);
return process_int_const(v->as<ast_int_const>(), code, target_type);
case ast_string_const:
return process_string_const(v->as<ast_string_const>(), code);
return process_string_const(v->as<ast_string_const>(), code, target_type);
case ast_bool_const:
return process_bool_const(v->as<ast_bool_const>(), code);
return process_bool_const(v->as<ast_bool_const>(), code, target_type);
case ast_null_keyword:
return process_null_keyword(v->as<ast_null_keyword>(), code);
return process_null_keyword(v->as<ast_null_keyword>(), code, target_type);
case ast_local_var_lhs:
return process_local_var(v->as<ast_local_var_lhs>(), code);
return process_local_var(v->as<ast_local_var_lhs>(), code, target_type);
case ast_local_vars_declaration:
return process_local_vars_declaration(v->as<ast_local_vars_declaration>(), code);
case ast_underscore:
@ -784,14 +1050,14 @@ static void process_assert_statement(V<ast_assert_statement> v, CodeBlob& code)
args[2]->mutate()->assign_inferred_type(TypeDataInt::create());
}
const FunctionData* builtin_sym = lookup_global_symbol("__throw_if_unless")->as<FunctionData>();
FunctionPtr builtin_sym = lookup_global_symbol("__throw_if_unless")->try_as<FunctionPtr>();
std::vector<var_idx_t> args_vars = pre_compile_tensor(code, args);
gen_op_call(code, TypeDataVoid::create(), v->loc, std::move(args_vars), builtin_sym, "(throw-call)");
}
static void process_catch_variable(AnyExprV v_catch_var, CodeBlob& code) {
if (auto v_ref = v_catch_var->try_as<ast_reference>(); v_ref && v_ref->sym) { // not underscore
const LocalVarData* var_ref = v_ref->sym->as<LocalVarData>();
LocalVarPtr var_ref = v_ref->sym->try_as<LocalVarPtr>();
tolk_assert(var_ref->ir_idx.empty());
var_ref->mutate()->assign_ir_idx(code.create_var(v_catch_var->inferred_type, v_catch_var->loc, var_ref->name));
}
@ -816,7 +1082,7 @@ static void process_try_catch_statement(V<ast_try_catch_statement> v, CodeBlob&
}
static void process_repeat_statement(V<ast_repeat_statement> v, CodeBlob& code) {
std::vector<var_idx_t> tmp_vars = pre_compile_expr(v->get_cond(), code);
std::vector<var_idx_t> tmp_vars = pre_compile_expr(v->get_cond(), code, nullptr);
Op& repeat_op = code.emplace_back(v->loc, Op::_Repeat, tmp_vars);
code.push_set_cur(repeat_op.block0);
process_any_statement(v->get_body(), code);
@ -824,7 +1090,7 @@ static void process_repeat_statement(V<ast_repeat_statement> v, CodeBlob& code)
}
static void process_if_statement(V<ast_if_statement> v, CodeBlob& code) {
std::vector<var_idx_t> tmp_vars = pre_compile_expr(v->get_cond(), code);
std::vector<var_idx_t> tmp_vars = pre_compile_expr(v->get_cond(), code, nullptr);
Op& if_op = code.emplace_back(v->loc, Op::_If, std::move(tmp_vars));
code.push_set_cur(if_op.block0);
process_any_statement(v->get_if_body(), code);
@ -869,19 +1135,21 @@ static void process_do_while_statement(V<ast_do_while_statement> v, CodeBlob& co
}
until_cond->mutate()->assign_inferred_type(TypeDataInt::create());
if (auto v_bin = until_cond->try_as<ast_binary_operator>(); v_bin && !v_bin->fun_ref) {
v_bin->mutate()->assign_fun_ref(lookup_global_symbol("_" + static_cast<std::string>(v_bin->operator_name) + "_")->as<FunctionData>());
v_bin->mutate()->assign_fun_ref(lookup_global_symbol("_" + static_cast<std::string>(v_bin->operator_name) + "_")->try_as<FunctionPtr>());
} else if (auto v_un = until_cond->try_as<ast_unary_operator>(); v_un && !v_un->fun_ref) {
v_un->mutate()->assign_fun_ref(lookup_global_symbol(static_cast<std::string>(v_un->operator_name) + "_")->as<FunctionData>());
v_un->mutate()->assign_fun_ref(lookup_global_symbol(static_cast<std::string>(v_un->operator_name) + "_")->try_as<FunctionPtr>());
}
until_op.left = pre_compile_expr(until_cond, code);
until_op.left = pre_compile_expr(until_cond, code, nullptr);
tolk_assert(until_op.left.size() == 1);
code.close_pop_cur(v->get_body()->loc_end);
}
static void process_while_statement(V<ast_while_statement> v, CodeBlob& code) {
Op& while_op = code.emplace_back(v->loc, Op::_While);
code.push_set_cur(while_op.block0);
while_op.left = pre_compile_expr(v->get_cond(), code);
while_op.left = pre_compile_expr(v->get_cond(), code, nullptr);
tolk_assert(while_op.left.size() == 1);
code.close_pop_cur(v->get_body()->loc);
code.push_set_cur(while_op.block1);
process_any_statement(v->get_body(), code);
@ -890,18 +1158,25 @@ static void process_while_statement(V<ast_while_statement> v, CodeBlob& code) {
static void process_throw_statement(V<ast_throw_statement> v, CodeBlob& code) {
if (v->has_thrown_arg()) {
const FunctionData* builtin_sym = lookup_global_symbol("__throw_arg")->as<FunctionData>();
FunctionPtr builtin_sym = lookup_global_symbol("__throw_arg")->try_as<FunctionPtr>();
std::vector<var_idx_t> args_vars = pre_compile_tensor(code, {v->get_thrown_arg(), v->get_thrown_code()});
gen_op_call(code, TypeDataVoid::create(), v->loc, std::move(args_vars), builtin_sym, "(throw-call)");
} else {
const FunctionData* builtin_sym = lookup_global_symbol("__throw")->as<FunctionData>();
FunctionPtr builtin_sym = lookup_global_symbol("__throw")->try_as<FunctionPtr>();
std::vector<var_idx_t> args_vars = pre_compile_tensor(code, {v->get_thrown_code()});
gen_op_call(code, TypeDataVoid::create(), v->loc, std::move(args_vars), builtin_sym, "(throw-call)");
}
}
static void process_return_statement(V<ast_return_statement> v, CodeBlob& code) {
std::vector<var_idx_t> return_vars = v->has_return_value() ? pre_compile_expr(v->get_return_value(), code) : std::vector<var_idx_t>{};
std::vector<var_idx_t> return_vars;
if (v->has_return_value()) {
TypePtr child_target_type = code.fun_ref->inferred_return_type;
if (code.fun_ref->does_return_self()) {
child_target_type = code.fun_ref->parameters[0].declared_type;
}
return_vars = pre_compile_expr(v->get_return_value(), code, child_target_type);
}
if (code.fun_ref->does_return_self()) {
return_vars = {};
}
@ -953,18 +1228,18 @@ void process_any_statement(AnyV v, CodeBlob& code) {
case ast_empty_statement:
return;
default:
pre_compile_expr(reinterpret_cast<AnyExprV>(v), code);
pre_compile_expr(reinterpret_cast<AnyExprV>(v), code, nullptr);
}
}
static void convert_function_body_to_CodeBlob(const FunctionData* fun_ref, FunctionBodyCode* code_body) {
static void convert_function_body_to_CodeBlob(FunctionPtr fun_ref, FunctionBodyCode* code_body) {
auto v_body = fun_ref->ast_root->as<ast_function_declaration>()->get_body()->as<ast_sequence>();
CodeBlob* blob = new CodeBlob{fun_ref->name, fun_ref->loc, fun_ref};
std::vector<var_idx_t> rvect_import;
int total_arg_width = 0;
for (int i = 0; i < fun_ref->get_num_params(); ++i) {
total_arg_width += fun_ref->parameters[i].declared_type->calc_width_on_stack();
total_arg_width += fun_ref->parameters[i].declared_type->get_width_on_stack();
}
rvect_import.reserve(total_arg_width);
@ -990,9 +1265,9 @@ static void convert_function_body_to_CodeBlob(const FunctionData* fun_ref, Funct
tolk_assert(vars_modification_watcher.empty());
}
static void convert_asm_body_to_AsmOp(const FunctionData* fun_ref, FunctionBodyAsm* asm_body) {
static void convert_asm_body_to_AsmOp(FunctionPtr fun_ref, FunctionBodyAsm* asm_body) {
int cnt = fun_ref->get_num_params();
int width = fun_ref->inferred_return_type->calc_width_on_stack();
int width = fun_ref->inferred_return_type->get_width_on_stack();
std::vector<AsmOp> asm_ops;
for (AnyV v_child : fun_ref->ast_root->as<ast_function_declaration>()->get_body()->as<ast_asm_body>()->get_asm_commands()) {
std::string_view ops = v_child->as<ast_string_const>()->str_val; // <op>\n<op>\n...
@ -1023,15 +1298,15 @@ static void convert_asm_body_to_AsmOp(const FunctionData* fun_ref, FunctionBodyA
class UpdateArgRetOrderConsideringStackWidth final {
public:
static bool should_visit_function(const FunctionData* fun_ref) {
static bool should_visit_function(FunctionPtr fun_ref) {
return !fun_ref->is_generic_function() && (!fun_ref->ret_order.empty() || !fun_ref->arg_order.empty());
}
static void start_visiting_function(const FunctionData* fun_ref, V<ast_function_declaration> v_function) {
static void start_visiting_function(FunctionPtr fun_ref, V<ast_function_declaration> v_function) {
int total_arg_mutate_width = 0;
bool has_arg_width_not_1 = false;
for (const LocalVarData& param : fun_ref->parameters) {
int arg_width = param.declared_type->calc_width_on_stack();
int arg_width = param.declared_type->get_width_on_stack();
has_arg_width_not_1 |= arg_width != 1;
total_arg_mutate_width += param.is_mutate_parameter() * arg_width;
}
@ -1045,7 +1320,7 @@ public:
cum_arg_width.reserve(1 + fun_ref->get_num_params());
cum_arg_width.push_back(0);
for (const LocalVarData& param : fun_ref->parameters) {
cum_arg_width.push_back(total_arg_width += param.declared_type->calc_width_on_stack());
cum_arg_width.push_back(total_arg_width += param.declared_type->get_width_on_stack());
}
std::vector<int> arg_order;
for (int i = 0; i < fun_ref->get_num_params(); ++i) {
@ -1062,7 +1337,7 @@ public:
// ret_order is a shuffled range 0...N
// validate N: a function should return value and mutated arguments onto a stack
if (!fun_ref->ret_order.empty()) {
size_t expected_width = fun_ref->inferred_return_type->calc_width_on_stack() + total_arg_mutate_width;
size_t expected_width = fun_ref->inferred_return_type->get_width_on_stack() + total_arg_mutate_width;
if (expected_width != fun_ref->ret_order.size()) {
v_function->get_body()->error("ret_order (after ->) expected to contain " + std::to_string(expected_width) + " numbers");
}
@ -1072,11 +1347,11 @@ public:
class ConvertASTToLegacyOpVisitor final {
public:
static bool should_visit_function(const FunctionData* fun_ref) {
static bool should_visit_function(FunctionPtr fun_ref) {
return !fun_ref->is_generic_function();
}
static void start_visiting_function(const FunctionData* fun_ref, V<ast_function_declaration>) {
static void start_visiting_function(FunctionPtr fun_ref, V<ast_function_declaration>) {
tolk_assert(fun_ref->is_type_inferring_done());
if (fun_ref->is_code_function()) {
convert_function_body_to_CodeBlob(fun_ref, std::get<FunctionBodyCode*>(fun_ref->body));

View file

@ -177,6 +177,18 @@ class CalculateRvalueLvalueVisitor final : public ASTVisitorFunctionBody {
parent::visit(v->get_expr()); // leave lvalue state unchanged, for `mutate (t.0 as int)` both `t.0 as int` and `t.0` are lvalue
}
void visit(V<ast_not_null_operator> v) override {
mark_vertex_cur_or_rvalue(v);
parent::visit(v->get_expr()); // leave lvalue state unchanged, for `mutate x!` both `x!` and `x` are lvalue
}
void visit(V<ast_is_null_check> v) override {
mark_vertex_cur_or_rvalue(v);
MarkingState saved = enter_state(MarkingState::RValue);
parent::visit(v->get_expr());
restore_state(saved);
}
void visit(V<ast_local_var_lhs> v) override {
tolk_assert(cur_state == MarkingState::LValue);
mark_vertex_cur_or_rvalue(v);
@ -198,7 +210,7 @@ class CalculateRvalueLvalueVisitor final : public ASTVisitorFunctionBody {
}
public:
bool should_visit_function(const FunctionData* fun_ref) override {
bool should_visit_function(FunctionPtr fun_ref) override {
return fun_ref->is_code_function() && !fun_ref->is_generic_function();
}
};
@ -207,7 +219,7 @@ void pipeline_calculate_rvalue_lvalue() {
visit_ast_of_all_functions<CalculateRvalueLvalueVisitor>();
}
void pipeline_calculate_rvalue_lvalue(const FunctionData* fun_ref) {
void pipeline_calculate_rvalue_lvalue(FunctionPtr fun_ref) {
CalculateRvalueLvalueVisitor visitor;
if (visitor.should_visit_function(fun_ref)) {
visitor.start_visiting_function(fun_ref, fun_ref->ast_root->as<ast_function_declaration>());

View file

@ -34,7 +34,7 @@ static void fire_error_impure_operation_inside_pure_function(AnyV v) {
class CheckImpureOperationsInPureFunctionVisitor final : public ASTVisitorFunctionBody {
static void fire_if_global_var(AnyExprV v) {
if (auto v_ident = v->try_as<ast_reference>()) {
if (v_ident->sym->try_as<GlobalVarData>()) {
if (v_ident->sym->try_as<GlobalVarPtr>()) {
fire_error_impure_operation_inside_pure_function(v);
}
}
@ -81,7 +81,7 @@ class CheckImpureOperationsInPureFunctionVisitor final : public ASTVisitorFuncti
}
public:
bool should_visit_function(const FunctionData* fun_ref) override {
bool should_visit_function(FunctionPtr fun_ref) override {
return fun_ref->is_code_function() && !fun_ref->is_generic_function() && fun_ref->is_marked_as_pure();
}
};

View file

@ -37,7 +37,7 @@ static void fire_error_cannot_be_used_as_lvalue(AnyV v, const std::string& detai
}
GNU_ATTRIBUTE_NORETURN GNU_ATTRIBUTE_COLD
static void fire_error_modifying_immutable_variable(AnyExprV v, const LocalVarData* var_ref) {
static void fire_error_modifying_immutable_variable(AnyExprV v, LocalVarPtr var_ref) {
if (var_ref->param_idx == 0 && var_ref->name == "self") {
v->error("modifying `self`, which is immutable by default; probably, you want to declare `mutate self`");
} else {
@ -47,7 +47,7 @@ static void fire_error_modifying_immutable_variable(AnyExprV v, const LocalVarDa
// validate a function used as rvalue, like `var cb = f`
// it's not a generic function (ensured earlier at type inferring) and has some more restrictions
static void validate_function_used_as_noncall(AnyExprV v, const FunctionData* fun_ref) {
static void validate_function_used_as_noncall(AnyExprV v, FunctionPtr fun_ref) {
if (!fun_ref->arg_order.empty() || !fun_ref->ret_order.empty()) {
v->error("saving `" + fun_ref->name + "` into a variable will most likely lead to invalid usage, since it changes the order of variables on the stack");
}
@ -97,6 +97,18 @@ class CheckRValueLvalueVisitor final : public ASTVisitorFunctionBody {
parent::visit(v->get_expr());
}
void visit(V<ast_not_null_operator> v) override {
// if `x!` is lvalue, then `x` is also lvalue, so check that `x` is ok
parent::visit(v->get_expr());
}
void visit(V<ast_is_null_check> v) override {
if (v->is_lvalue) {
fire_error_cannot_be_used_as_lvalue(v, v->is_negated ? "operator !=" : "operator ==");
}
parent::visit(v->get_expr());
}
void visit(V<ast_int_const> v) override {
if (v->is_lvalue) {
fire_error_cannot_be_used_as_lvalue(v, "literal");
@ -124,7 +136,7 @@ class CheckRValueLvalueVisitor final : public ASTVisitorFunctionBody {
void visit(V<ast_dot_access> v) override {
// a reference to a method used as rvalue, like `var v = t.tupleAt`
if (v->is_rvalue && v->is_target_fun_ref()) {
validate_function_used_as_noncall(v, std::get<const FunctionData*>(v->target));
validate_function_used_as_noncall(v, std::get<FunctionPtr>(v->target));
}
}
@ -158,17 +170,17 @@ class CheckRValueLvalueVisitor final : public ASTVisitorFunctionBody {
void visit(V<ast_reference> v) override {
if (v->is_lvalue) {
tolk_assert(v->sym);
if (const auto* var_ref = v->sym->try_as<LocalVarData>(); var_ref && var_ref->is_immutable()) {
if (LocalVarPtr var_ref = v->sym->try_as<LocalVarPtr>(); var_ref && var_ref->is_immutable()) {
fire_error_modifying_immutable_variable(v, var_ref);
} else if (v->sym->try_as<GlobalConstData>()) {
} else if (v->sym->try_as<GlobalConstPtr>()) {
v->error("modifying immutable constant");
} else if (v->sym->try_as<FunctionData>()) {
} else if (v->sym->try_as<FunctionPtr>()) {
v->error("function can't be used as lvalue");
}
}
// a reference to a function used as rvalue, like `var v = someFunction`
if (const FunctionData* fun_ref = v->sym->try_as<FunctionData>(); fun_ref && v->is_rvalue) {
if (FunctionPtr fun_ref = v->sym->try_as<FunctionPtr>(); fun_ref && v->is_rvalue) {
validate_function_used_as_noncall(v, fun_ref);
}
}
@ -186,7 +198,7 @@ class CheckRValueLvalueVisitor final : public ASTVisitorFunctionBody {
}
public:
bool should_visit_function(const FunctionData* fun_ref) override {
bool should_visit_function(FunctionPtr fun_ref) override {
return fun_ref->is_code_function() && !fun_ref->is_generic_function();
}
};

View file

@ -88,8 +88,19 @@ class ConstantFoldingReplacer final : public ASTReplacerInFunctionBody {
return v;
}
AnyExprV replace(V<ast_is_null_check> v) override {
parent::replace(v);
// `null == null` / `null != null`
if (v->get_expr()->type == ast_null_keyword) {
return create_bool_const(v->loc, !v->is_negated);
}
return v;
}
public:
bool should_visit_function(const FunctionData* fun_ref) override {
bool should_visit_function(FunctionPtr fun_ref) override {
return fun_ref->is_code_function() && !fun_ref->is_generic_function();
}
};

View file

@ -111,11 +111,11 @@ class UnreachableStatementsDetectVisitor final {
}
public:
static bool should_visit_function(const FunctionData* fun_ref) {
static bool should_visit_function(FunctionPtr fun_ref) {
return fun_ref->is_code_function() && !fun_ref->is_generic_function();
}
void start_visiting_function(const FunctionData* fun_ref, V<ast_function_declaration> v_function) {
void start_visiting_function(FunctionPtr fun_ref, V<ast_function_declaration> v_function) {
bool control_flow_reaches_end = !always_returns(v_function->get_body()->as<ast_sequence>());
if (control_flow_reaches_end) {
fun_ref->mutate()->assign_is_implicit_return();
@ -128,7 +128,7 @@ void pipeline_detect_unreachable_statements() {
visit_ast_of_all_functions<UnreachableStatementsDetectVisitor>();
}
void pipeline_detect_unreachable_statements(const FunctionData* fun_ref) {
void pipeline_detect_unreachable_statements(FunctionPtr fun_ref) {
UnreachableStatementsDetectVisitor visitor;
if (UnreachableStatementsDetectVisitor::should_visit_function(fun_ref)) {
visitor.start_visiting_function(fun_ref, fun_ref->ast_root->as<ast_function_declaration>());

View file

@ -36,7 +36,7 @@ namespace tolk {
static void mark_function_used_dfs(const std::unique_ptr<Op>& op);
static void mark_function_used(const FunctionData* fun_ref) {
static void mark_function_used(FunctionPtr fun_ref) {
if (!fun_ref->is_code_function() || fun_ref->is_really_used()) { // already handled
return;
}
@ -45,7 +45,7 @@ static void mark_function_used(const FunctionData* fun_ref) {
mark_function_used_dfs(std::get<FunctionBodyCode*>(fun_ref->body)->code->ops);
}
static void mark_global_var_used(const GlobalVarData* glob_ref) {
static void mark_global_var_used(GlobalVarPtr glob_ref) {
glob_ref->mutate()->assign_is_really_used();
}
@ -66,7 +66,7 @@ static void mark_function_used_dfs(const std::unique_ptr<Op>& op) {
}
void pipeline_find_unused_symbols() {
for (const FunctionData* fun_ref : G.all_functions) {
for (FunctionPtr fun_ref : G.all_functions) {
if (fun_ref->is_method_id_not_empty()) { // get methods, main and other entrypoints, regular functions with @method_id
mark_function_used(fun_ref);
}

View file

@ -39,7 +39,7 @@ void FunctionBodyAsm::set_code(std::vector<AsmOp>&& code) {
}
static void generate_output_func(const FunctionData* fun_ref) {
static void generate_output_func(FunctionPtr fun_ref) {
tolk_assert(fun_ref->is_code_function());
if (G.is_verbosity(2)) {
std::cerr << "\n\n=========================\nfunction " << fun_ref->name << " : " << fun_ref->inferred_return_type << std::endl;
@ -119,7 +119,7 @@ void pipeline_generate_fif_output_to_std_cout() {
std::cout << "PROGRAM{\n";
bool has_main_procedure = false;
for (const FunctionData* fun_ref : G.all_functions) {
for (FunctionPtr fun_ref : G.all_functions) {
if (!fun_ref->does_need_codegen()) {
if (G.is_verbosity(2) && fun_ref->is_code_function()) {
std::cerr << fun_ref->name << ": code not generated, function does not need codegen\n";
@ -143,7 +143,7 @@ void pipeline_generate_fif_output_to_std_cout() {
throw Fatal("the contract has no entrypoint; forgot `fun onInternalMessage(...)`?");
}
for (const GlobalVarData* var_ref : G.all_global_vars) {
for (GlobalVarPtr var_ref : G.all_global_vars) {
if (!var_ref->is_really_used() && G.settings.remove_unused_functions) {
if (G.is_verbosity(2)) {
std::cerr << var_ref->name << ": variable not generated, it's unused\n";
@ -154,7 +154,7 @@ void pipeline_generate_fif_output_to_std_cout() {
std::cout << std::string(2, ' ') << "DECLGLOBVAR " << var_ref->name << "\n";
}
for (const FunctionData* fun_ref : G.all_functions) {
for (FunctionPtr fun_ref : G.all_functions) {
if (!fun_ref->does_need_codegen()) {
continue;
}

View file

@ -63,9 +63,9 @@
namespace tolk {
static void infer_and_save_return_type_of_function(const FunctionData* fun_ref);
static void infer_and_save_return_type_of_function(FunctionPtr fun_ref);
static TypePtr get_or_infer_return_type(const FunctionData* fun_ref) {
static TypePtr get_or_infer_return_type(FunctionPtr fun_ref) {
if (!fun_ref->inferred_return_type) {
infer_and_save_return_type_of_function(fun_ref);
}
@ -83,12 +83,7 @@ static std::string to_string(AnyExprV v_with_type) {
}
GNU_ATTRIBUTE_NOINLINE
static std::string to_string(const LocalVarData& var_ref) {
return "`" + var_ref.declared_type->as_human_readable() + "`";
}
GNU_ATTRIBUTE_NOINLINE
static std::string to_string(const FunctionData* fun_ref) {
static std::string to_string(FunctionPtr fun_ref) {
return "`" + fun_ref->as_human_readable() + "`";
}
@ -96,8 +91,8 @@ static std::string to_string(const FunctionData* fun_ref) {
// asm functions generally can't handle it, they expect T to be a TVM primitive
// (in FunC, `forall` type just couldn't be unified with non-primitives; in Tolk, generic T is expectedly inferred)
GNU_ATTRIBUTE_NORETURN GNU_ATTRIBUTE_COLD
static void fire_error_calling_asm_function_with_non1_stack_width_arg(SrcLocation loc, const FunctionData* fun_ref, const std::vector<TypePtr>& substitutions, int arg_idx) {
throw ParseError(loc, "can not call `" + fun_ref->as_human_readable() + "` with " + fun_ref->genericTs->get_nameT(arg_idx) + "=" + substitutions[arg_idx]->as_human_readable() + ", because it occupies " + std::to_string(substitutions[arg_idx]->calc_width_on_stack()) + " stack slots in TVM, not 1");
static void fire_error_calling_asm_function_with_non1_stack_width_arg(SrcLocation loc, FunctionPtr fun_ref, const std::vector<TypePtr>& substitutions, int arg_idx) {
throw ParseError(loc, "can not call `" + fun_ref->as_human_readable() + "` with " + fun_ref->genericTs->get_nameT(arg_idx) + "=" + substitutions[arg_idx]->as_human_readable() + ", because it occupies " + std::to_string(substitutions[arg_idx]->get_width_on_stack()) + " stack slots in TVM, not 1");
}
// fire an error on `var n = null`
@ -105,7 +100,7 @@ static void fire_error_calling_asm_function_with_non1_stack_width_arg(SrcLocatio
// so, it's better to see an error on assignment, that later, on `n` usage and types mismatch
// (most common is situation above, but generally, `var (x,n) = xn` where xn is a tensor with 2-nd always-null, can be)
GNU_ATTRIBUTE_NORETURN GNU_ATTRIBUTE_COLD
static void fire_error_assign_always_null_to_variable(SrcLocation loc, const LocalVarData* assigned_var, bool is_assigned_null_literal) {
static void fire_error_assign_always_null_to_variable(SrcLocation loc, LocalVarPtr assigned_var, bool is_assigned_null_literal) {
std::string var_name = assigned_var->name;
throw ParseError(loc, "can not infer type of `" + var_name + "`, it's always null; specify its type with `" + var_name + ": <type>`" + (is_assigned_null_literal ? " or use `null as <type>`" : ""));
}
@ -134,34 +129,26 @@ static void fire_error_cannot_deduce_untyped_tuple_access(SrcLocation loc, int i
// fire an error on `untypedTupleVar.0` when inferred as (int,int), or `[int, (int,int)]`, or other non-1 width in a tuple
GNU_ATTRIBUTE_NORETURN GNU_ATTRIBUTE_COLD
static void fire_error_tuple_cannot_have_non1_stack_width_elem(SrcLocation loc, TypePtr inferred_type) {
throw ParseError(loc, "a tuple can not have " + to_string(inferred_type) + " inside, because it occupies " + std::to_string(inferred_type->calc_width_on_stack()) + " stack slots in TVM, not 1");
throw ParseError(loc, "a tuple can not have " + to_string(inferred_type) + " inside, because it occupies " + std::to_string(inferred_type->get_width_on_stack()) + " stack slots in TVM, not 1");
}
// check correctness of called arguments counts and their type matching
static void check_function_arguments(const FunctionData* fun_ref, V<ast_argument_list> v, AnyExprV lhs_of_dot_call) {
int delta_self = lhs_of_dot_call ? 1 : 0;
int n_arguments = v->size() + delta_self;
int n_parameters = fun_ref->get_num_params();
// Tolk doesn't have optional parameters currently, so just compare counts
if (!n_parameters && lhs_of_dot_call) {
v->error("`" + fun_ref->name + "` has no parameters and can not be called as method");
}
if (n_parameters < n_arguments) {
v->error("too many arguments in call to `" + fun_ref->name + "`, expected " + std::to_string(n_parameters - delta_self) + ", have " + std::to_string(n_arguments - delta_self));
}
if (n_arguments < n_parameters) {
v->error("too few arguments in call to `" + fun_ref->name + "`, expected " + std::to_string(n_parameters - delta_self) + ", have " + std::to_string(n_arguments - delta_self));
}
if (lhs_of_dot_call) {
if (!fun_ref->parameters[0].declared_type->can_rhs_be_assigned(lhs_of_dot_call->inferred_type)) {
lhs_of_dot_call->error("can not call method for " + to_string(fun_ref->parameters[0]) + " with object of type " + to_string(lhs_of_dot_call));
// check type correctness of a passed argument when calling a function/method
static void check_function_argument(TypePtr param_type, bool is_mutate_param, AnyExprV ith_arg, bool is_obj_of_dot_call) {
// given `f(x: int)` and a call `f(expr)`, check that expr_type is assignable to `int`
if (!param_type->can_rhs_be_assigned(ith_arg->inferred_type)) {
if (is_obj_of_dot_call) {
ith_arg->error("can not call method for " + to_string(param_type) + " with object of type " + to_string(ith_arg));
} else {
ith_arg->error("can not pass " + to_string(ith_arg) + " to " + to_string(param_type));
}
}
for (int i = 0; i < v->size(); ++i) {
if (!fun_ref->parameters[i + delta_self].declared_type->can_rhs_be_assigned(v->get_arg(i)->inferred_type)) {
v->get_arg(i)->error("can not pass " + to_string(v->get_arg(i)) + " to " + to_string(fun_ref->parameters[i + delta_self]));
// given `f(x: mutate int?)` and a call `f(expr)`, check that `int?` is assignable to expr_type
// (for instance, can't call such a function with `f(mutate intVal)`, since f can potentially assign null to it)
if (is_mutate_param && !ith_arg->inferred_type->can_rhs_be_assigned(param_type)) {
if (is_obj_of_dot_call) {
ith_arg->error("can not call method for mutate " + to_string(param_type) + " with object of type " + to_string(ith_arg) + ", because mutation is not type compatible");
} else {
ith_arg->error("can not pass " + to_string(ith_arg) + " to mutate " + to_string(param_type) + ", because mutation is not type compatible");
}
}
}
@ -189,6 +176,13 @@ class TypeInferringUnifyStrategy {
return t2;
}
if (t1 == TypeDataNullLiteral::create()) {
return TypeDataNullable::create(t2);
}
if (t2 == TypeDataNullLiteral::create()) {
return TypeDataNullable::create(t1);
}
const auto* tensor1 = t1->try_as<TypeDataTensor>();
const auto* tensor2 = t2->try_as<TypeDataTensor>();
if (tensor1 && tensor2 && tensor1->size() == tensor2->size()) {
@ -256,8 +250,8 @@ public:
// handle __expect_type(expr, "type") call
// this is used in compiler tests
GNU_ATTRIBUTE_NOINLINE GNU_ATTRIBUTE_COLD
static void handle_possible_compiler_internal_call(const FunctionData* current_function, V<ast_function_call> v) {
const FunctionData* fun_ref = v->fun_maybe;
static void handle_possible_compiler_internal_call(FunctionPtr current_function, V<ast_function_call> v) {
FunctionPtr fun_ref = v->fun_maybe;
tolk_assert(fun_ref && fun_ref->is_builtin_function());
static_cast<void>(current_function);
@ -279,7 +273,7 @@ static void handle_possible_compiler_internal_call(const FunctionData* current_f
* 2) easy to maintain a hint (see comments at the top of the file)
*/
class InferCheckTypesAndCallsAndFieldsVisitor final {
const FunctionData* current_function = nullptr;
FunctionPtr current_function = nullptr;
TypeInferringUnifyStrategy return_unifier;
GNU_ATTRIBUTE_ALWAYS_INLINE
@ -298,14 +292,14 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
dst->mutate()->assign_inferred_type(inferred_type);
}
static void assign_inferred_type(const LocalVarData* local_var_or_param, TypePtr inferred_type) {
static void assign_inferred_type(LocalVarPtr local_var_or_param, TypePtr inferred_type) {
#ifdef TOLK_DEBUG
tolk_assert(inferred_type != nullptr && !inferred_type->has_unresolved_inside() && !inferred_type->has_genericT_inside());
#endif
local_var_or_param->mutate()->assign_inferred_type(inferred_type);
}
static void assign_inferred_type(const FunctionData* fun_ref, TypePtr inferred_return_type, TypePtr inferred_full_type) {
static void assign_inferred_type(FunctionPtr fun_ref, TypePtr inferred_return_type, TypePtr inferred_full_type) {
#ifdef TOLK_DEBUG
tolk_assert(inferred_return_type != nullptr && !inferred_return_type->has_unresolved_inside() && !inferred_return_type->has_genericT_inside());
#endif
@ -365,6 +359,10 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
return infer_ternary_operator(v->as<ast_ternary_operator>(), hint);
case ast_cast_as_operator:
return infer_cast_as_operator(v->as<ast_cast_as_operator>());
case ast_not_null_operator:
return infer_not_null_operator(v->as<ast_not_null_operator>());
case ast_is_null_check:
return infer_is_null_check(v->as<ast_is_null_check>());
case ast_parenthesized_expression:
return infer_parenthesized(v->as<ast_parenthesized_expression>(), hint);
case ast_reference:
@ -388,14 +386,29 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
}
}
static TypePtr unwrap_nullable(TypePtr type) {
while (const TypeDataNullable* as_nullable = type->try_as<TypeDataNullable>()) {
type = as_nullable->inner;
}
return type;
}
static bool expect_integer(AnyExprV v_inferred) {
return v_inferred->inferred_type == TypeDataInt::create();
}
static bool expect_integer(TypePtr inferred_type) {
return inferred_type == TypeDataInt::create();
}
static bool expect_boolean(AnyExprV v_inferred) {
return v_inferred->inferred_type == TypeDataBool::create();
}
static bool expect_boolean(TypePtr inferred_type) {
return inferred_type == TypeDataBool::create();
}
static void infer_int_const(V<ast_int_const> v) {
assign_inferred_type(v, TypeDataInt::create());
}
@ -467,7 +480,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
}
} else {
if (rhs_type == TypeDataNullLiteral::create()) {
fire_error_assign_always_null_to_variable(err_loc->loc, lhs_var->var_ref->try_as<LocalVarData>(), corresponding_maybe_rhs && corresponding_maybe_rhs->type == ast_null_keyword);
fire_error_assign_always_null_to_variable(err_loc->loc, lhs_var->var_ref->try_as<LocalVarPtr>(), corresponding_maybe_rhs && corresponding_maybe_rhs->type == ast_null_keyword);
}
assign_inferred_type(lhs_var, rhs_type);
assign_inferred_type(lhs_var->var_ref, rhs_type);
@ -520,7 +533,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
// check `untypedTuple.0 = rhs_tensor` and other non-1 width elements
if (auto lhs_dot = lhs->try_as<ast_dot_access>()) {
if (lhs_dot->is_target_indexed_access() && lhs_dot->get_obj()->inferred_type == TypeDataTuple::create()) {
if (rhs_type->calc_width_on_stack() != 1) {
if (rhs_type->get_width_on_stack() != 1) {
fire_error_tuple_cannot_have_non1_stack_width_elem(err_loc->loc, rhs_type);
}
}
@ -563,8 +576,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
assign_inferred_type(v, lhs);
if (!builtin_func.empty()) {
const FunctionData* builtin_sym = lookup_global_symbol("_" + static_cast<std::string>(builtin_func) + "_")->as<FunctionData>();
tolk_assert(builtin_sym);
FunctionPtr builtin_sym = lookup_global_symbol("_" + static_cast<std::string>(builtin_func) + "_")->try_as<FunctionPtr>();
v->mutate()->assign_fun_ref(builtin_sym);
}
}
@ -598,8 +610,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
}
if (!builtin_func.empty()) {
const FunctionData* builtin_sym = lookup_global_symbol(static_cast<std::string>(builtin_func) + "_")->as<FunctionData>();
tolk_assert(builtin_sym);
FunctionPtr builtin_sym = lookup_global_symbol(static_cast<std::string>(builtin_func) + "_")->try_as<FunctionPtr>();
v->mutate()->assign_fun_ref(builtin_sym);
}
}
@ -617,8 +628,8 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
// == != can compare both integers and booleans, (int == bool) is NOT allowed
case tok_eq:
case tok_neq: {
bool both_int = expect_integer(lhs) && expect_integer(rhs);
bool both_bool = expect_boolean(lhs) && expect_boolean(rhs);
bool both_int = expect_integer(unwrap_nullable(lhs->inferred_type)) && expect_integer(unwrap_nullable(rhs->inferred_type));
bool both_bool = expect_boolean(unwrap_nullable(lhs->inferred_type)) && expect_boolean(unwrap_nullable(rhs->inferred_type));
if (!both_int && !both_bool) {
if (lhs->inferred_type == rhs->inferred_type) { // compare slice with slice
v->error("type " + to_string(lhs) + " can not be compared with `== !=`");
@ -674,8 +685,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
}
if (!builtin_func.empty()) {
const FunctionData* builtin_sym = lookup_global_symbol("_" + static_cast<std::string>(builtin_func) + "_")->as<FunctionData>();
tolk_assert(builtin_sym);
FunctionPtr builtin_sym = lookup_global_symbol("_" + static_cast<std::string>(builtin_func) + "_")->try_as<FunctionPtr>();
v->mutate()->assign_fun_ref(builtin_sym);
}
}
@ -706,22 +716,38 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
assign_inferred_type(v, v->cast_to_type);
}
void infer_is_null_check(V<ast_is_null_check> v) {
infer_any_expr(v->get_expr());
assign_inferred_type(v, TypeDataBool::create());
}
void infer_not_null_operator(V<ast_not_null_operator> v) {
infer_any_expr(v->get_expr());
if (const auto* as_nullable = v->get_expr()->inferred_type->try_as<TypeDataNullable>()) {
// operator `!` used for `T?`, leave `T`
assign_inferred_type(v, as_nullable->inner);
} else {
// operator `!` used for non-nullable, probably a warning should be printed
assign_inferred_type(v, v->get_expr());
}
}
void infer_parenthesized(V<ast_parenthesized_expression> v, TypePtr hint) {
infer_any_expr(v->get_expr(), hint);
assign_inferred_type(v, v->get_expr());
}
static void infer_reference(V<ast_reference> v) {
if (const auto* var_ref = v->sym->try_as<LocalVarData>()) {
if (LocalVarPtr var_ref = v->sym->try_as<LocalVarPtr>()) {
assign_inferred_type(v, var_ref->declared_type);
} else if (const auto* const_ref = v->sym->try_as<GlobalConstData>()) {
} else if (GlobalConstPtr const_ref = v->sym->try_as<GlobalConstPtr>()) {
assign_inferred_type(v, const_ref->is_int_const() ? TypeDataInt::create() : TypeDataSlice::create());
} else if (const auto* glob_ref = v->sym->try_as<GlobalVarData>()) {
} else if (GlobalVarPtr glob_ref = v->sym->try_as<GlobalVarPtr>()) {
assign_inferred_type(v, glob_ref->declared_type);
} else if (const auto* fun_ref = v->sym->try_as<FunctionData>()) {
} else if (FunctionPtr fun_ref = v->sym->try_as<FunctionPtr>()) {
// it's `globalF` / `globalF<int>` - references to functions used as non-call
V<ast_instantiationT_list> v_instantiationTs = v->get_instantiationTs();
@ -758,7 +784,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
// given `genericF<int, slice>` / `t.tupleFirst<cell>` (the user manually specified instantiation Ts),
// validate and collect them
// returns: [int, slice] / [cell]
static std::vector<TypePtr> collect_fun_generic_substitutions_from_manually_specified(SrcLocation loc, const FunctionData* fun_ref, V<ast_instantiationT_list> instantiationT_list) {
static std::vector<TypePtr> collect_fun_generic_substitutions_from_manually_specified(SrcLocation loc, FunctionPtr fun_ref, V<ast_instantiationT_list> instantiationT_list) {
if (fun_ref->genericTs->size() != instantiationT_list->get_items().size()) {
throw ParseError(loc, "wrong count of generic T: expected " + std::to_string(fun_ref->genericTs->size()) + ", got " + std::to_string(instantiationT_list->size()));
}
@ -778,11 +804,11 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
// example: was `t.tuplePush<slice>(2)`, read <slice>, instantiate `tuplePush<slice>` (will later fail type check)
// example: was `var cb = t.tupleFirst<int>;` (used as reference, as non-call), instantiate `tupleFirst<int>`
// returns fun_ref to instantiated function
static const FunctionData* check_and_instantiate_generic_function(SrcLocation loc, const FunctionData* fun_ref, std::vector<TypePtr>&& substitutionTs) {
static FunctionPtr check_and_instantiate_generic_function(SrcLocation loc, FunctionPtr fun_ref, std::vector<TypePtr>&& substitutionTs) {
// T for asm function must be a TVM primitive (width 1), otherwise, asm would act incorrectly
if (fun_ref->is_asm_function() || fun_ref->is_builtin_function()) {
for (int i = 0; i < static_cast<int>(substitutionTs.size()); ++i) {
if (substitutionTs[i]->calc_width_on_stack() != 1) {
if (substitutionTs[i]->get_width_on_stack() != 1) {
fire_error_calling_asm_function_with_non1_stack_width_arg(loc, fun_ref, substitutionTs, i);
}
}
@ -836,7 +862,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
if (hint == nullptr) {
fire_error_cannot_deduce_untyped_tuple_access(v->loc, index_at);
}
if (hint->calc_width_on_stack() != 1) {
if (hint->get_width_on_stack() != 1) {
fire_error_tuple_cannot_have_non1_stack_width_elem(v->loc, hint);
}
item_type = hint;
@ -850,14 +876,14 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
// for now, Tolk doesn't have fields and object-scoped methods; `t.tupleSize` is a global function `tupleSize`
const Symbol* sym = lookup_global_symbol(field_name);
const FunctionData* fun_ref = sym ? sym->try_as<FunctionData>() : nullptr;
FunctionPtr fun_ref = sym ? sym->try_as<FunctionPtr>() : nullptr;
if (!fun_ref) {
v_ident->error("non-existing field `" + static_cast<std::string>(field_name) + "` of type " + to_string(obj_type));
}
// `t.tupleSize` is ok, `cs.tupleSize` not
if (!fun_ref->parameters[0].declared_type->can_rhs_be_assigned(obj_type)) {
v_ident->error("referencing a method for " + to_string(fun_ref->parameters[0]) + " with object of type " + to_string(obj_type));
v_ident->error("referencing a method for " + to_string(fun_ref->parameters[0].declared_type) + " with object of type " + to_string(obj_type));
}
if (fun_ref->is_generic_function() && !v_instantiationTs) {
@ -886,12 +912,12 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
// v is `globalF(args)` / `globalF<int>(args)` / `obj.method(args)` / `local_var(args)` / `getF()(args)`
int delta_self = 0;
AnyExprV dot_obj = nullptr;
const FunctionData* fun_ref = nullptr;
FunctionPtr fun_ref = nullptr;
V<ast_instantiationT_list> v_instantiationTs = nullptr;
if (auto v_ref = callee->try_as<ast_reference>()) {
// `globalF()` / `globalF<int>()` / `local_var()` / `SOME_CONST()`
fun_ref = v_ref->sym->try_as<FunctionData>(); // not null for `globalF`
fun_ref = v_ref->sym->try_as<FunctionPtr>(); // not null for `globalF`
v_instantiationTs = v_ref->get_instantiationTs(); // present for `globalF<int>()`
} else if (auto v_dot = callee->try_as<ast_dot_access>()) {
@ -910,7 +936,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
} else {
// for now, Tolk doesn't have fields and object-scoped methods; `t.tupleSize` is a global function `tupleSize`
const Symbol* sym = lookup_global_symbol(field_name);
fun_ref = sym ? sym->try_as<FunctionData>() : nullptr;
fun_ref = sym ? sym->try_as<FunctionPtr>() : nullptr;
if (!fun_ref) {
v_dot->get_identifier()->error("non-existing method `" + static_cast<std::string>(field_name) + "` of type " + to_string(dot_obj));
}
@ -921,30 +947,26 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
// fun_ref remains nullptr
}
// infer argument types, looking at fun_ref's parameters as hints
for (int i = 0; i < v->get_num_args(); ++i) {
TypePtr param_type = fun_ref && i < fun_ref->get_num_params() - delta_self ? fun_ref->parameters[delta_self + i].declared_type : nullptr;
auto arg_i = v->get_arg(i);
infer_any_expr(arg_i->get_expr(), param_type && !param_type->has_genericT_inside() ? param_type : nullptr);
assign_inferred_type(arg_i, arg_i->get_expr());
}
// handle `local_var()` / `getF()()` / `5()` / `SOME_CONST()` / `obj.method()()()` / `tensorVar.0()`
if (!fun_ref) {
// treat callee like a usual expression, which must have "callable" inferred type
infer_any_expr(callee);
const TypeDataFunCallable* f_callable = callee->inferred_type->try_as<TypeDataFunCallable>();
if (!f_callable) { // `5()` / `SOME_CONST()` / `null()`
v->error("calling a non-function");
v->error("calling a non-function " + to_string(callee->inferred_type));
}
// check arguments count and their types
if (v->get_num_args() != static_cast<int>(f_callable->params_types.size())) {
v->error("expected " + std::to_string(f_callable->params_types.size()) + " arguments, got " + std::to_string(v->get_arg_list()->size()));
}
for (int i = 0; i < v->get_num_args(); ++i) {
if (!f_callable->params_types[i]->can_rhs_be_assigned(v->get_arg(i)->inferred_type)) {
v->get_arg(i)->error("can not pass " + to_string(v->get_arg(i)) + " to " + to_string(f_callable->params_types[i]));
auto arg_i = v->get_arg(i)->get_expr();
TypePtr param_type = f_callable->params_types[i];
infer_any_expr(arg_i, param_type);
if (!param_type->can_rhs_be_assigned(arg_i->inferred_type)) {
arg_i->error("can not pass " + to_string(arg_i) + " to " + to_string(param_type));
}
assign_inferred_type(v->get_arg(i), arg_i);
}
v->mutate()->assign_fun_ref(nullptr); // no fun_ref to a global function
assign_inferred_type(v, f_callable->return_type);
@ -952,30 +974,75 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
}
// so, we have a call `f(args)` or `obj.f(args)`, f is a global function (fun_ref) (code / asm / builtin)
// we're going to iterate over passed arguments, check type compatibility, and (if generic) infer substitutionTs
// at first, check arguments count (Tolk doesn't have optional parameters, so just compare counts)
int n_arguments = v->get_num_args() + delta_self;
int n_parameters = fun_ref->get_num_params();
if (!n_parameters && dot_obj) {
v->error("`" + fun_ref->name + "` has no parameters and can not be called as method");
}
if (n_parameters < n_arguments) {
v->error("too many arguments in call to `" + fun_ref->name + "`, expected " + std::to_string(n_parameters - delta_self) + ", have " + std::to_string(n_arguments - delta_self));
}
if (n_arguments < n_parameters) {
v->error("too few arguments in call to `" + fun_ref->name + "`, expected " + std::to_string(n_parameters - delta_self) + ", have " + std::to_string(n_arguments - delta_self));
}
// now, for every passed argument, we need to infer its type, and check it against parameter type
// for regular functions, it's obvious
// but for generic functions, we need to infer type arguments (substitutionTs) on the fly
// (unless Ts are specified by a user like `f<int>(args)` / `t.tupleAt<slice>()`, take them)
GenericSubstitutionsDeduceForCall* deducingTs = fun_ref->is_generic_function() ? new GenericSubstitutionsDeduceForCall(fun_ref) : nullptr;
if (deducingTs && v_instantiationTs) {
deducingTs->provide_manually_specified(collect_fun_generic_substitutions_from_manually_specified(v->loc, fun_ref, v_instantiationTs));
}
// loop over every argument, for `obj.method()` obj is the first one
// if genericT deducing has a conflict, ParseError is thrown
// note, that deducing Ts one by one is important to manage control flow (mutate params work like assignments)
// a corner case, e.g. `f<T>(v1:T?, v2:T?)` and `f(null,2)` will fail on first argument, won't try the second one
if (dot_obj) {
const LocalVarData& param_0 = fun_ref->parameters[0];
TypePtr param_type = param_0.declared_type;
if (param_type->has_genericT_inside()) {
param_type = deducingTs->auto_deduce_from_argument(dot_obj->loc, param_type, dot_obj->inferred_type);
}
check_function_argument(param_type, param_0.is_mutate_parameter(), dot_obj, true);
}
for (int i = 0; i < v->get_num_args(); ++i) {
const LocalVarData& param_i = fun_ref->parameters[delta_self + i];
AnyExprV arg_i = v->get_arg(i)->get_expr();
TypePtr param_type = param_i.declared_type;
if (param_type->has_genericT_inside() && deducingTs->is_manually_specified()) { // `f<int>(a)`
param_type = deducingTs->replace_by_manually_specified(param_type);
}
if (param_type->has_genericT_inside()) { // `f(a)` where f is generic: use `a` to infer param type
infer_any_expr(arg_i); // then arg_i is inferred without any hint
param_type = deducingTs->auto_deduce_from_argument(arg_i->loc, param_type, arg_i->inferred_type);
} else {
infer_any_expr(arg_i, param_type); // param_type is hint, helps infer arg_i
}
assign_inferred_type(v->get_arg(i), arg_i); // arg itself is an expression
check_function_argument(param_type, param_i.is_mutate_parameter(), arg_i, false);
}
// if it's a generic function `f<T>`, we need to instantiate it, like `f<int>`
// same for generic methods `t.tupleAt<T>`, need to achieve `t.tupleAt<int>`
if (fun_ref->is_generic_function() && v_instantiationTs) {
// if Ts are specified by a user like `f<int>(args)` / `t.tupleAt<slice>()`, take them
std::vector<TypePtr> substitutions = collect_fun_generic_substitutions_from_manually_specified(v->loc, fun_ref, v_instantiationTs);
fun_ref = check_and_instantiate_generic_function(v->loc, fun_ref, std::move(substitutions));
} else if (fun_ref->is_generic_function()) {
// if `f<T>` called like `f(args)`, deduce T from arg types
std::vector<TypePtr> arg_types;
arg_types.reserve(delta_self + v->get_num_args());
if (dot_obj) {
arg_types.push_back(dot_obj->inferred_type);
if (fun_ref->is_generic_function()) {
// if `f(args)` was called, Ts were inferred; check that all of them are known
int idx = deducingTs->get_first_not_deduced_idx();
if (idx != -1 && hint && fun_ref->declared_return_type->has_genericT_inside()) {
// example: `t.tupleFirst()`, T doesn't depend on arguments, but is determined by return type
// if used like `var x: int = t.tupleFirst()` / `t.tupleFirst() as int` / etc., use hint
deducingTs->auto_deduce_from_argument(v->loc, fun_ref->declared_return_type, hint);
idx = deducingTs->get_first_not_deduced_idx();
}
for (int i = 0; i < v->get_num_args(); ++i) {
arg_types.push_back(v->get_arg(i)->inferred_type);
if (idx != -1) {
v->error("can not deduce " + fun_ref->genericTs->get_nameT(idx));
}
td::Result<std::vector<TypePtr>> deduced = deduce_substitutionTs_on_generic_func_call(fun_ref, std::move(arg_types), hint);
if (deduced.is_error()) {
v->error(deduced.error().message().str() + " for generic function " + to_string(fun_ref));
}
fun_ref = check_and_instantiate_generic_function(v->loc, fun_ref, deduced.move_as_ok());
fun_ref = check_and_instantiate_generic_function(v->loc, fun_ref, deducingTs->flush());
delete deducingTs;
} else if (UNLIKELY(v_instantiationTs != nullptr)) {
// non-generic function/method called with type arguments, like `c.cellHash<int>()` / `beginCell<builder>()`
@ -988,8 +1055,6 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
v->get_callee()->as<ast_dot_access>()->mutate()->assign_target(fun_ref);
v->get_callee()->as<ast_dot_access>()->mutate()->assign_inferred_type(fun_ref->inferred_full_type);
}
// check arguments count and their types
check_function_arguments(fun_ref, v->get_arg_list(), dot_obj);
// get return type either from user-specified declaration or infer here on demand traversing its body
get_or_infer_return_type(fun_ref);
TypePtr inferred_type = dot_obj && fun_ref->does_return_self() ? dot_obj->inferred_type : fun_ref->inferred_return_type;
@ -1020,7 +1085,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
for (int i = 0; i < v->size(); ++i) {
AnyExprV item = v->get_item(i);
infer_any_expr(item, tuple_hint && i < tuple_hint->size() ? tuple_hint->items[i] : nullptr);
if (item->inferred_type->calc_width_on_stack() != 1) {
if (item->inferred_type->get_width_on_stack() != 1) {
fire_error_tuple_cannot_have_non1_stack_width_elem(v->get_item(i)->loc, item->inferred_type);
}
types_list.emplace_back(item->inferred_type);
@ -1134,7 +1199,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
v->get_thrown_code()->error("excNo of `throw` must be an integer, got " + to_string(v->get_thrown_code()));
}
infer_any_expr(v->get_thrown_arg());
if (v->has_thrown_arg() && v->get_thrown_arg()->inferred_type->calc_width_on_stack() != 1) {
if (v->has_thrown_arg() && v->get_thrown_arg()->inferred_type->get_width_on_stack() != 1) {
v->get_thrown_arg()->error("can not throw " + to_string(v->get_thrown_arg()) + ", exception arg must occupy exactly 1 stack slot");
}
}
@ -1153,7 +1218,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
static void process_catch_variable(AnyExprV catch_var, TypePtr catch_var_type) {
if (auto v_ref = catch_var->try_as<ast_reference>(); v_ref && v_ref->sym) { // not underscore
assign_inferred_type(v_ref->sym->as<LocalVarData>(), catch_var_type);
assign_inferred_type(v_ref->sym->try_as<LocalVarPtr>(), catch_var_type);
}
assign_inferred_type(catch_var, catch_var_type);
}
@ -1163,7 +1228,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
// `catch` has exactly 2 variables: excNo and arg (when missing, they are implicit underscores)
// `arg` is a curious thing, it can be any TVM primitive, so assign unknown to it
// hence, using `fInt(arg)` (int from parameter is a hint) or `arg as slice` works well
// hence, using `fInt(arg)` (int from parameter is a target type) or `arg as slice` works well
// it's not truly correct, because `arg as (int,int)` also compiles, but can never happen, but let it be user responsibility
tolk_assert(v->get_catch_expr()->size() == 2);
std::vector<TypePtr> types_list = {TypeDataInt::create(), TypeDataUnknown::create()};
@ -1175,7 +1240,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final {
}
public:
static void assign_fun_full_type(const FunctionData* fun_ref, TypePtr inferred_return_type) {
static void assign_fun_full_type(FunctionPtr fun_ref, TypePtr inferred_return_type) {
// calculate function full type `fun(params) -> ret_type`
std::vector<TypePtr> params_types;
params_types.reserve(fun_ref->get_num_params());
@ -1185,7 +1250,7 @@ public:
assign_inferred_type(fun_ref, inferred_return_type, TypeDataFunCallable::create(std::move(params_types), inferred_return_type));
}
void start_visiting_function(const FunctionData* fun_ref, V<ast_function_declaration> v_function) {
void start_visiting_function(FunctionPtr fun_ref, V<ast_function_declaration> v_function) {
if (fun_ref->is_code_function()) {
current_function = fun_ref;
process_any_statement(v_function->get_body());
@ -1212,12 +1277,12 @@ public:
class LaunchInferTypesAndMethodsOnce final {
public:
static bool should_visit_function(const FunctionData* fun_ref) {
static bool should_visit_function(FunctionPtr fun_ref) {
// since inferring can be requested on demand, prevent second execution from a regular pipeline launcher
return !fun_ref->is_type_inferring_done() && !fun_ref->is_generic_function();
}
static void start_visiting_function(const FunctionData* fun_ref, V<ast_function_declaration> v_function) {
static void start_visiting_function(FunctionPtr fun_ref, V<ast_function_declaration> v_function) {
InferCheckTypesAndCallsAndFieldsVisitor visitor;
visitor.start_visiting_function(fun_ref, v_function);
}
@ -1227,8 +1292,8 @@ public:
// example: `fun f() { return g(); } fun g() { ... }`
// when analyzing `f()`, we need to infer what fun_ref=g returns
// (if `g` is generic, it was already instantiated, so fun_ref=g<int> is here)
static void infer_and_save_return_type_of_function(const FunctionData* fun_ref) {
static std::vector<const FunctionData*> called_stack;
static void infer_and_save_return_type_of_function(FunctionPtr fun_ref) {
static std::vector<FunctionPtr> called_stack;
tolk_assert(!fun_ref->is_generic_function() && !fun_ref->is_type_inferring_done());
// if `g` has return type declared, like `fun g(): int { ... }`, don't traverse its body
@ -1255,7 +1320,7 @@ void pipeline_infer_types_and_calls_and_fields() {
visit_ast_of_all_functions<LaunchInferTypesAndMethodsOnce>();
}
void pipeline_infer_types_and_calls_and_fields(const FunctionData* fun_ref) {
void pipeline_infer_types_and_calls_and_fields(FunctionPtr fun_ref) {
InferCheckTypesAndCallsAndFieldsVisitor visitor;
visitor.start_visiting_function(fun_ref, fun_ref->ast_root->as<ast_function_declaration>());
}

View file

@ -53,7 +53,7 @@ struct OptimizerBooleanExpressionsReplacer final : ASTReplacerInFunctionBody {
auto v_not = createV<ast_unary_operator>(loc, "!", tok_logical_not, rhs);
v_not->assign_inferred_type(TypeDataBool::create());
v_not->assign_rvalue_true();
v_not->assign_fun_ref(lookup_global_symbol("!b_")->as<FunctionData>());
v_not->assign_fun_ref(lookup_global_symbol("!b_")->try_as<FunctionPtr>());
return v_not;
}
@ -75,7 +75,7 @@ protected:
auto v_neq = createV<ast_binary_operator>(v->loc, "!=", tok_neq, cond_not_not, v_zero);
v_neq->mutate()->assign_rvalue_true();
v_neq->mutate()->assign_inferred_type(TypeDataBool::create());
v_neq->mutate()->assign_fun_ref(lookup_global_symbol("_!=_")->as<FunctionData>());
v_neq->mutate()->assign_fun_ref(lookup_global_symbol("_!=_")->try_as<FunctionPtr>());
return v_neq;
}
}
@ -117,12 +117,17 @@ protected:
}
v = createV<ast_if_statement>(v->loc, !v->is_ifnot, v_cond_unary->get_rhs(), v->get_if_body(), v->get_else_body());
}
// `if (x != null)` -> ifnot(x == null)
if (auto v_cond_isnull = v->get_cond()->try_as<ast_is_null_check>(); v_cond_isnull && v_cond_isnull->is_negated) {
v_cond_isnull->mutate()->assign_is_negated(!v_cond_isnull->is_negated);
v = createV<ast_if_statement>(v->loc, !v->is_ifnot, v_cond_isnull, v->get_if_body(), v->get_else_body());
}
return v;
}
public:
bool should_visit_function(const FunctionData* fun_ref) override {
bool should_visit_function(FunctionPtr fun_ref) override {
return fun_ref->is_code_function() && !fun_ref->is_generic_function();
}
};

View file

@ -34,7 +34,7 @@
namespace tolk {
GNU_ATTRIBUTE_NORETURN GNU_ATTRIBUTE_COLD
static void fire_error_invalid_mutate_arg_passed(AnyExprV v, const FunctionData* fun_ref, const LocalVarData& p_sym, bool called_as_method, bool arg_passed_as_mutate, AnyV arg_expr) {
static void fire_error_invalid_mutate_arg_passed(AnyExprV v, FunctionPtr fun_ref, const LocalVarData& p_sym, bool called_as_method, bool arg_passed_as_mutate, AnyV arg_expr) {
std::string arg_str(arg_expr->type == ast_reference ? arg_expr->as<ast_reference>()->get_name() : "obj");
// case: `loadInt(cs, 32)`; suggest: `cs.loadInt(32)`
@ -60,7 +60,7 @@ static void fire_error_invalid_mutate_arg_passed(AnyExprV v, const FunctionData*
class RefineLvalueForMutateArgumentsVisitor final : public ASTVisitorFunctionBody {
void visit(V<ast_function_call> v) override {
// v is `globalF(args)` / `globalF<int>(args)` / `obj.method(args)` / `local_var(args)` / `getF()(args)`
const FunctionData* fun_ref = v->fun_maybe;
FunctionPtr fun_ref = v->fun_maybe;
if (!fun_ref) {
parent::visit(v);
for (int i = 0; i < v->get_num_args(); ++i) {
@ -86,6 +86,8 @@ class RefineLvalueForMutateArgumentsVisitor final : public ASTVisitorFunctionBod
leftmost_obj = as_par->get_expr();
} else if (auto as_cast = leftmost_obj->try_as<ast_cast_as_operator>()) {
leftmost_obj = as_cast->get_expr();
} else if (auto as_nn = leftmost_obj->try_as<ast_not_null_operator>()) {
leftmost_obj = as_nn->get_expr();
} else {
break;
}
@ -114,7 +116,7 @@ class RefineLvalueForMutateArgumentsVisitor final : public ASTVisitorFunctionBod
}
public:
bool should_visit_function(const FunctionData* fun_ref) override {
bool should_visit_function(FunctionPtr fun_ref) override {
return fun_ref->is_code_function() && !fun_ref->is_generic_function();
}
};

View file

@ -176,8 +176,8 @@ static void register_function(V<ast_function_declaration> v) {
genericTs = construct_genericTs(v->genericsT_list);
}
if (v->is_builtin_function()) {
const Symbol* builtin_func = lookup_global_symbol(func_name);
const FunctionData* fun_ref = builtin_func ? builtin_func->as<FunctionData>() : nullptr;
const Symbol* sym = lookup_global_symbol(func_name);
FunctionPtr fun_ref = sym ? sym->try_as<FunctionPtr>() : nullptr;
if (!fun_ref || !fun_ref->is_builtin_function()) {
v->error("`builtin` used for non-builtin function");
}
@ -202,7 +202,7 @@ static void register_function(V<ast_function_declaration> v) {
f_sym->method_id = static_cast<int>(v->method_id->to_long());
} else if (v->flags & FunctionData::flagGetMethod) {
f_sym->method_id = calculate_method_id_by_func_name(func_name);
for (const FunctionData* other : G.all_get_methods) {
for (FunctionPtr other : G.all_get_methods) {
if (other->method_id == f_sym->method_id) {
v->error(PSTRING() << "GET methods hash collision: `" << other->name << "` and `" << f_sym->name << "` produce the same hash. Consider renaming one of these functions.");
}

View file

@ -119,7 +119,7 @@ struct NameAndScopeResolver {
return G.symtable.lookup(name);
}
void add_local_var(const LocalVarData* v_sym) {
void add_local_var(LocalVarPtr v_sym) {
if (UNLIKELY(scopes.empty())) {
throw Fatal("unexpected scope_level = 0");
}
@ -168,9 +168,9 @@ static TypePtr finalize_type_data(TypePtr type_data, const GenericsDeclaration*
class AssignSymInsideFunctionVisitor final : public ASTVisitorFunctionBody {
// more correctly this field shouldn't be static, but currently there is no need to make it a part of state
static NameAndScopeResolver current_scope;
static const FunctionData* current_function;
static FunctionPtr current_function;
static const LocalVarData* create_local_var_sym(std::string_view name, SrcLocation loc, TypePtr declared_type, bool immutable) {
static LocalVarPtr create_local_var_sym(std::string_view name, SrcLocation loc, TypePtr declared_type, bool immutable) {
LocalVarData* v_sym = new LocalVarData(static_cast<std::string>(name), loc, declared_type, immutable * LocalVarData::flagImmutable, -1);
current_scope.add_local_var(v_sym);
return v_sym;
@ -178,7 +178,7 @@ class AssignSymInsideFunctionVisitor final : public ASTVisitorFunctionBody {
static void process_catch_variable(AnyExprV catch_var) {
if (auto v_ref = catch_var->try_as<ast_reference>()) {
const LocalVarData* var_ref = create_local_var_sym(v_ref->get_name(), catch_var->loc, nullptr, true);
LocalVarPtr var_ref = create_local_var_sym(v_ref->get_name(), catch_var->loc, nullptr, true);
v_ref->mutate()->assign_sym(var_ref);
}
}
@ -190,14 +190,14 @@ protected:
if (sym == nullptr) {
v->error("`redef` for unknown variable");
}
const LocalVarData* var_ref = sym->try_as<LocalVarData>();
LocalVarPtr var_ref = sym->try_as<LocalVarPtr>();
if (!var_ref) {
v->error("`redef` for unknown variable");
}
v->mutate()->assign_var_ref(var_ref);
} else {
TypePtr declared_type = finalize_type_data(v->declared_type, current_function->genericTs);
const LocalVarData* var_ref = create_local_var_sym(v->get_name(), v->loc, declared_type, v->is_immutable);
LocalVarPtr var_ref = create_local_var_sym(v->get_name(), v->loc, declared_type, v->is_immutable);
v->mutate()->assign_resolved_type(declared_type);
v->mutate()->assign_var_ref(var_ref);
}
@ -216,7 +216,7 @@ protected:
v->mutate()->assign_sym(sym);
// for global functions, global vars and constants, `import` must exist
if (!sym->try_as<LocalVarData>()) {
if (!sym->try_as<LocalVarPtr>()) {
check_import_exists_when_using_sym(v, sym);
}
@ -276,14 +276,14 @@ protected:
}
public:
bool should_visit_function(const FunctionData* fun_ref) override {
bool should_visit_function(FunctionPtr fun_ref) override {
// this pipe is done just after parsing
// visit both asm and code functions, resolve identifiers in parameter/return types everywhere
// for generic functions, unresolved "T" will be replaced by TypeDataGenericT
return true;
}
void start_visiting_function(const FunctionData* fun_ref, V<ast_function_declaration> v) override {
void start_visiting_function(FunctionPtr fun_ref, V<ast_function_declaration> v) override {
current_function = fun_ref;
for (int i = 0; i < v->get_num_params(); ++i) {
@ -313,7 +313,7 @@ public:
};
NameAndScopeResolver AssignSymInsideFunctionVisitor::current_scope;
const FunctionData* AssignSymInsideFunctionVisitor::current_function = nullptr;
FunctionPtr AssignSymInsideFunctionVisitor::current_function = nullptr;
void pipeline_resolve_identifiers_and_assign_symbols() {
AssignSymInsideFunctionVisitor visitor;
@ -337,7 +337,7 @@ void pipeline_resolve_identifiers_and_assign_symbols() {
}
}
void pipeline_resolve_identifiers_and_assign_symbols(const FunctionData* fun_ref) {
void pipeline_resolve_identifiers_and_assign_symbols(FunctionPtr fun_ref) {
AssignSymInsideFunctionVisitor visitor;
if (visitor.should_visit_function(fun_ref)) {
visitor.start_visiting_function(fun_ref, fun_ref->ast_root->as<ast_function_declaration>());

View file

@ -49,10 +49,10 @@ void pipeline_generate_fif_output_to_std_cout();
// these pipes also can be called per-function individually
// they are called for instantiated generics functions, when `f<T>` is deeply cloned as `f<int>`
void pipeline_resolve_identifiers_and_assign_symbols(const FunctionData*);
void pipeline_calculate_rvalue_lvalue(const FunctionData*);
void pipeline_detect_unreachable_statements(const FunctionData*);
void pipeline_infer_types_and_calls_and_fields(const FunctionData*);
void pipeline_resolve_identifiers_and_assign_symbols(FunctionPtr);
void pipeline_calculate_rvalue_lvalue(FunctionPtr);
void pipeline_detect_unreachable_statements(FunctionPtr);
void pipeline_infer_types_and_calls_and_fields(FunctionPtr);
} // namespace tolk

View file

@ -120,7 +120,7 @@ static void fire_error_redefinition_of_symbol(SrcLocation loc, const Symbol* pre
throw ParseError(loc, "redefinition of built-in symbol");
}
void GlobalSymbolTable::add_function(const FunctionData* f_sym) {
void GlobalSymbolTable::add_function(FunctionPtr f_sym) {
auto key = key_hash(f_sym->name);
auto [it, inserted] = entries.emplace(key, f_sym);
if (!inserted) {
@ -128,7 +128,7 @@ void GlobalSymbolTable::add_function(const FunctionData* f_sym) {
}
}
void GlobalSymbolTable::add_global_var(const GlobalVarData* g_sym) {
void GlobalSymbolTable::add_global_var(GlobalVarPtr g_sym) {
auto key = key_hash(g_sym->name);
auto [it, inserted] = entries.emplace(key, g_sym);
if (!inserted) {
@ -136,7 +136,7 @@ void GlobalSymbolTable::add_global_var(const GlobalVarData* g_sym) {
}
}
void GlobalSymbolTable::add_global_const(const GlobalConstData* c_sym) {
void GlobalSymbolTable::add_global_const(GlobalConstPtr c_sym) {
auto key = key_hash(c_sym->name);
auto [it, inserted] = entries.emplace(key, c_sym);
if (!inserted) {

View file

@ -37,17 +37,12 @@ struct Symbol {
virtual ~Symbol() = default;
template<class T>
const T* as() const {
template<class ConstTPtr>
ConstTPtr try_as() const {
#ifdef TOLK_DEBUG
assert(dynamic_cast<const T*>(this) != nullptr);
assert(this != nullptr);
#endif
return dynamic_cast<const T*>(this);
}
template<class T>
const T* try_as() const {
return dynamic_cast<const T*>(this);
return dynamic_cast<ConstTPtr>(this);
}
};
@ -229,9 +224,9 @@ class GlobalSymbolTable {
}
public:
void add_function(const FunctionData* f_sym);
void add_global_var(const GlobalVarData* g_sym);
void add_global_const(const GlobalConstData* c_sym);
void add_function(FunctionPtr f_sym);
void add_global_var(GlobalVarPtr g_sym);
void add_global_const(GlobalConstPtr c_sym);
const Symbol* lookup(std::string_view name) const {
const auto it = entries.find(key_hash(name));

View file

@ -45,7 +45,7 @@ typedef int const_idx_t;
struct TmpVar {
var_idx_t ir_idx; // every var in IR represents 1 stack slot
TypePtr v_type; // calc_width_on_stack() is 1
TypePtr v_type; // get_width_on_stack() is 1
std::string name; // "x" for vars originated from user sources; "x.0" for tensor components; empty for implicitly created tmp vars
SrcLocation loc; // location of var declaration in sources or where a tmp var was originated
#ifdef TOLK_DEBUG
@ -283,8 +283,8 @@ struct Op {
enum { _Disabled = 1, _NoReturn = 4, _Impure = 24 };
int flags;
std::unique_ptr<Op> next;
const FunctionData* f_sym = nullptr;
const GlobalVarData* g_sym = nullptr;
FunctionPtr f_sym = nullptr;
GlobalVarPtr g_sym = nullptr;
SrcLocation where;
VarDescrList var_info;
std::vector<VarDescr> args;
@ -313,19 +313,19 @@ struct Op {
: cl(_cl), flags(0), f_sym(nullptr), where(_where), left(std::move(_left)), right(std::move(_right)) {
}
Op(SrcLocation _where, OpKind _cl, const std::vector<var_idx_t>& _left, const std::vector<var_idx_t>& _right,
const FunctionData* _fun)
FunctionPtr _fun)
: cl(_cl), flags(0), f_sym(_fun), where(_where), left(_left), right(_right) {
}
Op(SrcLocation _where, OpKind _cl, std::vector<var_idx_t>&& _left, std::vector<var_idx_t>&& _right,
const FunctionData* _fun)
FunctionPtr _fun)
: cl(_cl), flags(0), f_sym(_fun), where(_where), left(std::move(_left)), right(std::move(_right)) {
}
Op(SrcLocation _where, OpKind _cl, const std::vector<var_idx_t>& _left, const std::vector<var_idx_t>& _right,
const GlobalVarData* _gvar)
GlobalVarPtr _gvar)
: cl(_cl), flags(0), g_sym(_gvar), where(_where), left(_left), right(_right) {
}
Op(SrcLocation _where, OpKind _cl, std::vector<var_idx_t>&& _left, std::vector<var_idx_t>&& _right,
const GlobalVarData* _gvar)
GlobalVarPtr _gvar)
: cl(_cl), flags(0), g_sym(_gvar), where(_where), left(std::move(_left)), right(std::move(_right)) {
}
@ -1083,7 +1083,7 @@ struct FunctionBodyAsm {
struct CodeBlob {
int var_cnt, in_var_cnt;
const FunctionData* fun_ref;
FunctionPtr fun_ref;
std::string name;
SrcLocation loc;
std::vector<TmpVar> vars;
@ -1094,7 +1094,7 @@ struct CodeBlob {
#endif
std::stack<std::unique_ptr<Op>*> cur_ops_stack;
bool require_callxargs = false;
CodeBlob(std::string name, SrcLocation loc, const FunctionData* fun_ref)
CodeBlob(std::string name, SrcLocation loc, FunctionPtr fun_ref)
: var_cnt(0), in_var_cnt(0), fun_ref(fun_ref), name(std::move(name)), loc(loc), cur_ops(&ops) {
}
template <typename... Args>

View file

@ -108,6 +108,19 @@ void type_system_init() {
// and creates an object only if it isn't found in a global hashtable
//
TypePtr TypeDataNullable::create(TypePtr inner) {
TypeDataTypeIdCalculation hash(1774084920039440885ULL);
hash.feed_child(inner);
if (TypePtr existing = hash.get_existing()) {
return existing;
}
// most types (int?, slice?, etc.), when nullable, still occupy 1 stack slot (holding TVM NULL at runtime)
// but for example for `(int, int)` we need an extra stack slot "null flag"
int width_on_stack = inner->can_hold_tvm_null_instead() ? 1 : inner->get_width_on_stack() + 1;
return hash.register_unique(new TypeDataNullable(hash.type_id(), hash.children_flags(), width_on_stack, inner));
}
TypePtr TypeDataFunCallable::create(std::vector<TypePtr>&& params_types, TypePtr return_type) {
TypeDataTypeIdCalculation hash(3184039965511020991ULL);
for (TypePtr param : params_types) {
@ -143,7 +156,11 @@ TypePtr TypeDataTensor::create(std::vector<TypePtr>&& items) {
if (TypePtr existing = hash.get_existing()) {
return existing;
}
return hash.register_unique(new TypeDataTensor(hash.type_id(), hash.children_flags(), std::move(items)));
int width_on_stack = 0;
for (TypePtr item : items) {
width_on_stack += item->get_width_on_stack();
}
return hash.register_unique(new TypeDataTensor(hash.type_id(), hash.children_flags(), width_on_stack, std::move(items)));
}
TypePtr TypeDataTypedTuple::create(std::vector<TypePtr>&& items) {
@ -178,6 +195,12 @@ TypePtr TypeDataUnresolved::create(std::string&& text, SrcLocation loc) {
// only non-trivial implementations are here; trivial are defined in .h file
//
std::string TypeDataNullable::as_human_readable() const {
std::string nested = inner->as_human_readable();
bool embrace = inner->try_as<TypeDataFunCallable>();
return embrace ? "(" + nested + ")?" : nested + "?";
}
std::string TypeDataFunCallable::as_human_readable() const {
std::string result = "(";
for (TypePtr param : params_types) {
@ -223,6 +246,11 @@ std::string TypeDataTypedTuple::as_human_readable() const {
// only non-trivial implementations are here; by default (no children), `callback(this)` is executed
//
void TypeDataNullable::traverse(const TraverserCallbackT& callback) const {
callback(this);
inner->traverse(callback);
}
void TypeDataFunCallable::traverse(const TraverserCallbackT& callback) const {
callback(this);
for (TypePtr param : params_types) {
@ -254,6 +282,10 @@ void TypeDataTypedTuple::traverse(const TraverserCallbackT& callback) const {
// only non-trivial implementations are here; by default (no children), `return callback(this)` is executed
//
TypePtr TypeDataNullable::replace_children_custom(const ReplacerCallbackT& callback) const {
return callback(create(inner->replace_children_custom(callback)));
}
TypePtr TypeDataFunCallable::replace_children_custom(const ReplacerCallbackT& callback) const {
std::vector<TypePtr> mapped;
mapped.reserve(params_types.size());
@ -282,53 +314,17 @@ TypePtr TypeDataTypedTuple::replace_children_custom(const ReplacerCallbackT& cal
}
// --------------------------------------------
// calc_width_on_stack()
//
// returns the number of stack slots occupied by a variable of this type
// only non-trivial implementations are here; by default (most types) occupy 1 stack slot
//
int TypeDataGenericT::calc_width_on_stack() const {
// this function is invoked only in functions with generics already instantiated
assert(false);
return -999999;
}
int TypeDataTensor::calc_width_on_stack() const {
int sum = 0;
for (TypePtr item : items) {
sum += item->calc_width_on_stack();
}
return sum;
}
int TypeDataUnresolved::calc_width_on_stack() const {
// since early pipeline stages, no unresolved types left
assert(false);
return -999999;
}
int TypeDataVoid::calc_width_on_stack() const {
return 0;
}
// --------------------------------------------
// can_rhs_be_assigned()
//
// on `var lhs: <lhs_type> = rhs`, having inferred rhs_type, check that it can be assigned without any casts
// the same goes for passing arguments, returning values, etc. — where the "receiver" (lhs) checks "applier" (rhs)
// for now, `null` can be assigned to any TVM primitive, be later we'll have T? types and null safety
//
bool TypeDataInt::can_rhs_be_assigned(TypePtr rhs) const {
if (rhs == this) {
return true;
}
if (rhs == TypeDataNullLiteral::create()) {
return true;
}
return false;
}
@ -336,9 +332,6 @@ bool TypeDataBool::can_rhs_be_assigned(TypePtr rhs) const {
if (rhs == this) {
return true;
}
if (rhs == TypeDataNullLiteral::create()) {
return true;
}
return false;
}
@ -346,9 +339,6 @@ bool TypeDataCell::can_rhs_be_assigned(TypePtr rhs) const {
if (rhs == this) {
return true;
}
if (rhs == TypeDataNullLiteral::create()) {
return true;
}
return false;
}
@ -356,9 +346,6 @@ bool TypeDataSlice::can_rhs_be_assigned(TypePtr rhs) const {
if (rhs == this) {
return true;
}
if (rhs == TypeDataNullLiteral::create()) {
return true;
}
return false;
}
@ -366,9 +353,6 @@ bool TypeDataBuilder::can_rhs_be_assigned(TypePtr rhs) const {
if (rhs == this) {
return true;
}
if (rhs == TypeDataNullLiteral::create()) {
return true;
}
return false;
}
@ -376,9 +360,6 @@ bool TypeDataTuple::can_rhs_be_assigned(TypePtr rhs) const {
if (rhs == this) {
return true;
}
if (rhs == TypeDataNullLiteral::create()) {
return true;
}
return false;
}
@ -386,9 +367,6 @@ bool TypeDataContinuation::can_rhs_be_assigned(TypePtr rhs) const {
if (rhs == this) {
return true;
}
if (rhs == TypeDataNullLiteral::create()) {
return true;
}
return false;
}
@ -396,6 +374,19 @@ bool TypeDataNullLiteral::can_rhs_be_assigned(TypePtr rhs) const {
return rhs == this;
}
bool TypeDataNullable::can_rhs_be_assigned(TypePtr rhs) const {
if (rhs == this) {
return true;
}
if (rhs == TypeDataNullLiteral::create()) {
return true;
}
if (const TypeDataNullable* rhs_nullable = rhs->try_as<TypeDataNullable>()) {
return inner->can_rhs_be_assigned(rhs_nullable->inner);
}
return inner->can_rhs_be_assigned(rhs);
}
bool TypeDataFunCallable::can_rhs_be_assigned(TypePtr rhs) const {
return rhs == this;
}
@ -414,7 +405,6 @@ bool TypeDataTensor::can_rhs_be_assigned(TypePtr rhs) const {
}
return true;
}
// note, that tensors can not accept null
return false;
}
@ -427,9 +417,6 @@ bool TypeDataTypedTuple::can_rhs_be_assigned(TypePtr rhs) const {
}
return true;
}
if (rhs == TypeDataNullLiteral::create()) {
return true;
}
return false;
}
@ -455,41 +442,69 @@ bool TypeDataVoid::can_rhs_be_assigned(TypePtr rhs) const {
//
bool TypeDataInt::can_be_casted_with_as_operator(TypePtr cast_to) const {
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) { // `int` as `int?`
return can_be_casted_with_as_operator(to_nullable->inner);
}
return cast_to == this;
}
bool TypeDataBool::can_be_casted_with_as_operator(TypePtr cast_to) const {
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) {
return can_be_casted_with_as_operator(to_nullable->inner);
}
return cast_to == this || cast_to == TypeDataInt::create();
}
bool TypeDataCell::can_be_casted_with_as_operator(TypePtr cast_to) const {
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) {
return can_be_casted_with_as_operator(to_nullable->inner);
}
return cast_to == this;
}
bool TypeDataSlice::can_be_casted_with_as_operator(TypePtr cast_to) const {
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) {
return can_be_casted_with_as_operator(to_nullable->inner);
}
return cast_to == this;
}
bool TypeDataBuilder::can_be_casted_with_as_operator(TypePtr cast_to) const {
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) {
return can_be_casted_with_as_operator(to_nullable->inner);
}
return cast_to == this;
}
bool TypeDataTuple::can_be_casted_with_as_operator(TypePtr cast_to) const {
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) {
return can_be_casted_with_as_operator(to_nullable->inner);
}
return cast_to == this;
}
bool TypeDataContinuation::can_be_casted_with_as_operator(TypePtr cast_to) const {
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) {
return can_be_casted_with_as_operator(to_nullable->inner);
}
return cast_to == this;
}
bool TypeDataNullLiteral::can_be_casted_with_as_operator(TypePtr cast_to) const {
return cast_to == this
|| cast_to == TypeDataInt::create() || cast_to == TypeDataBool::create() || cast_to == TypeDataCell::create() || cast_to == TypeDataSlice::create()
|| cast_to == TypeDataBuilder::create() || cast_to == TypeDataContinuation::create() || cast_to == TypeDataTuple::create()
|| cast_to->try_as<TypeDataTypedTuple>();
return cast_to == this || cast_to->try_as<TypeDataNullable>();
}
bool TypeDataNullable::can_be_casted_with_as_operator(TypePtr cast_to) const {
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) {
return inner->can_be_casted_with_as_operator(to_nullable->inner);
}
return false;
}
bool TypeDataFunCallable::can_be_casted_with_as_operator(TypePtr cast_to) const {
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) {
return can_be_casted_with_as_operator(to_nullable->inner);
}
return this == cast_to;
}
@ -506,6 +521,9 @@ bool TypeDataTensor::can_be_casted_with_as_operator(TypePtr cast_to) const {
}
return true;
}
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) {
return can_be_casted_with_as_operator(to_nullable->inner);
}
return false;
}
@ -518,14 +536,15 @@ bool TypeDataTypedTuple::can_be_casted_with_as_operator(TypePtr cast_to) const {
}
return true;
}
if (const auto* to_nullable = cast_to->try_as<TypeDataNullable>()) {
return can_be_casted_with_as_operator(to_nullable->inner);
}
return false;
}
bool TypeDataUnknown::can_be_casted_with_as_operator(TypePtr cast_to) const {
// 'unknown' can be cast to any type
// (though it's not valid for exception arguments when casting them to non-1 stack width,
// but to ensure it, we need a special type "unknown TVM primitive", which is overwhelming I think)
return true;
// 'unknown' can be cast to any TVM value
return cast_to->get_width_on_stack() == 1;
}
bool TypeDataUnresolved::can_be_casted_with_as_operator(TypePtr cast_to) const {
@ -537,12 +556,45 @@ bool TypeDataVoid::can_be_casted_with_as_operator(TypePtr cast_to) const {
}
// --------------------------------------------
// can_hold_tvm_null_instead()
//
// assigning `null` to a primitive variable like `int?` / `cell?` can store TVM NULL inside the same slot
// (that's why the default implementation is just "return true", and most of types occupy 1 slot)
// but for complex variables, like `(int, int)?`, "null presence" is kept in a separate slot (UTag for union types)
// though still, tricky situations like `(int, ())?` can still "embed" TVM NULL in parallel with original value
//
bool TypeDataNullable::can_hold_tvm_null_instead() const {
if (get_width_on_stack() != 1) { // `(int, int)?` / `()?` can not hold null instead
return false; // only `int?` / `cell?` / `StructWith1IntField?` can
} // and some tricky situations like `(int, ())?`, but not `(int?, ())?`
return !inner->can_hold_tvm_null_instead();
}
bool TypeDataTensor::can_hold_tvm_null_instead() const {
if (get_width_on_stack() != 1) { // `(int, int)` / `()` can not hold null instead, since null is 1 slot
return false; // only `((), int)` and similar can:
} // one item is width 1 (and not nullable), others are 0
for (TypePtr item : items) {
if (item->get_width_on_stack() == 1 && !item->can_hold_tvm_null_instead()) {
return false;
}
}
return true;
}
bool TypeDataVoid::can_hold_tvm_null_instead() const {
return false;
}
// --------------------------------------------
// parsing type from tokens
//
// here we implement parsing types (mostly after colon) to TypeData
// example: `var v: int` is TypeDataInt
// example: `var v: (builder, [cell])` is TypeDataTensor(TypeDataBuilder, TypeDataTypedTuple(TypeDataCell))
// example: `var v: (builder?, [cell])` is TypeDataTensor(TypeDataNullable(TypeDataBuilder), TypeDataTypedTuple(TypeDataCell))
// example: `fun f(): ()` is TypeDataTensor() (an empty one)
//
// note, that unrecognized type names (MyEnum, MyStruct, T) are parsed as TypeDataUnresolved,
@ -633,7 +685,8 @@ static TypePtr parse_type_nullable(Lexer& lex) {
TypePtr result = parse_simple_type(lex);
if (lex.tok() == tok_question) {
lex.error("nullable types are not supported yet");
lex.next();
result = TypeDataNullable::create(result);
}
return result;

View file

@ -50,6 +50,8 @@ class TypeData {
const uint64_t type_id;
// bits of flag_mask, to store often-used properties and return them without tree traversing
const int flags;
// how many slots on a stack this type occupies (calculated on creation), e.g. `int`=1, `(int,int)`=2, `(int,int)?`=3
const int width_on_stack;
friend class TypeDataTypeIdCalculation;
@ -60,9 +62,10 @@ protected:
flag_contains_unresolved_inside = 1 << 3,
};
explicit TypeData(uint64_t type_id, int flags_with_children)
explicit TypeData(uint64_t type_id, int flags_with_children, int width_on_stack)
: type_id(type_id)
, flags(flags_with_children) {
, flags(flags_with_children)
, width_on_stack(width_on_stack) {
}
public:
@ -74,6 +77,7 @@ public:
}
uint64_t get_type_id() const { return type_id; }
int get_width_on_stack() const { return width_on_stack; }
bool has_unknown_inside() const { return flags & flag_contains_unknown_inside; }
bool has_genericT_inside() const { return flags & flag_contains_genericT_inside; }
@ -86,6 +90,10 @@ public:
virtual bool can_rhs_be_assigned(TypePtr rhs) const = 0;
virtual bool can_be_casted_with_as_operator(TypePtr cast_to) const = 0;
virtual bool can_hold_tvm_null_instead() const {
return true;
}
virtual void traverse(const TraverserCallbackT& callback) const {
callback(this);
}
@ -93,17 +101,13 @@ public:
virtual TypePtr replace_children_custom(const ReplacerCallbackT& callback) const {
return callback(this);
}
virtual int calc_width_on_stack() const {
return 1;
}
};
/*
* `int` is TypeDataInt, representation of TVM int.
*/
class TypeDataInt final : public TypeData {
TypeDataInt() : TypeData(1ULL, 0) {}
TypeDataInt() : TypeData(1ULL, 0, 1) {}
static TypePtr singleton;
friend void type_system_init();
@ -121,7 +125,7 @@ public:
* From the type system point of view, int and bool are different, not-autocastable types.
*/
class TypeDataBool final : public TypeData {
TypeDataBool() : TypeData(2ULL, 0) {}
TypeDataBool() : TypeData(2ULL, 0, 1) {}
static TypePtr singleton;
friend void type_system_init();
@ -138,7 +142,7 @@ public:
* `cell` is TypeDataCell, representation of TVM cell.
*/
class TypeDataCell final : public TypeData {
TypeDataCell() : TypeData(3ULL, 0) {}
TypeDataCell() : TypeData(3ULL, 0, 1) {}
static TypePtr singleton;
friend void type_system_init();
@ -155,7 +159,7 @@ public:
* `slice` is TypeDataSlice, representation of TVM slice.
*/
class TypeDataSlice final : public TypeData {
TypeDataSlice() : TypeData(4ULL, 0) {}
TypeDataSlice() : TypeData(4ULL, 0, 1) {}
static TypePtr singleton;
friend void type_system_init();
@ -172,7 +176,7 @@ public:
* `builder` is TypeDataBuilder, representation of TVM builder.
*/
class TypeDataBuilder final : public TypeData {
TypeDataBuilder() : TypeData(5ULL, 0) {}
TypeDataBuilder() : TypeData(5ULL, 0, 1) {}
static TypePtr singleton;
friend void type_system_init();
@ -191,7 +195,7 @@ public:
* so getting its element results in TypeDataUnknown (which must be assigned/cast explicitly).
*/
class TypeDataTuple final : public TypeData {
TypeDataTuple() : TypeData(6ULL, 0) {}
TypeDataTuple() : TypeData(6ULL, 0, 1) {}
static TypePtr singleton;
friend void type_system_init();
@ -209,7 +213,7 @@ public:
* It's like "untyped callable", not compatible with other types.
*/
class TypeDataContinuation final : public TypeData {
TypeDataContinuation() : TypeData(7ULL, 0) {}
TypeDataContinuation() : TypeData(7ULL, 0, 1) {}
static TypePtr singleton;
friend void type_system_init();
@ -224,12 +228,12 @@ public:
/*
* `null` has TypeDataNullLiteral type.
* Currently, it can be assigned to int/slice/etc., but later Tolk will have T? types and null safety.
* It can be assigned only to nullable types (`int?`, etc.), to ensure null safety.
* Note, that `var i = null`, though valid (i would be constant null), fires an "always-null" compilation error
* (it's much better for user to see an error here than when he passes this variable somewhere).
*/
class TypeDataNullLiteral final : public TypeData {
TypeDataNullLiteral() : TypeData(8ULL, 0) {}
TypeDataNullLiteral() : TypeData(8ULL, 0, 1) {}
static TypePtr singleton;
friend void type_system_init();
@ -242,6 +246,30 @@ public:
bool can_be_casted_with_as_operator(TypePtr cast_to) const override;
};
/*
* `T?` is "nullable T".
* It can be converted to T either with ! (non-null assertion operator) or with smart casts.
*/
class TypeDataNullable final : public TypeData {
TypeDataNullable(uint64_t type_id, int children_flags, int width_on_stack, TypePtr inner)
: TypeData(type_id, children_flags, width_on_stack)
, inner(inner) {}
public:
const TypePtr inner;
static TypePtr create(TypePtr inner);
bool is_primitive_nullable() const { return get_width_on_stack() == 1 && inner->get_width_on_stack() == 1; }
std::string as_human_readable() const override;
bool can_rhs_be_assigned(TypePtr rhs) const override;
bool can_be_casted_with_as_operator(TypePtr cast_to) const override;
void traverse(const TraverserCallbackT& callback) const override;
TypePtr replace_children_custom(const ReplacerCallbackT& callback) const override;
bool can_hold_tvm_null_instead() const override;
};
/*
* `fun(int, int) -> void` is TypeDataFunCallable, think of is as a typed continuation.
* A type of function `fun f(x: int) { return x; }` is actually `fun(int) -> int`.
@ -249,7 +277,7 @@ public:
*/
class TypeDataFunCallable final : public TypeData {
TypeDataFunCallable(uint64_t type_id, int children_flags, std::vector<TypePtr>&& params_types, TypePtr return_type)
: TypeData(type_id, children_flags)
: TypeData(type_id, children_flags, 1)
, params_types(std::move(params_types))
, return_type(return_type) {}
@ -275,7 +303,7 @@ public:
*/
class TypeDataGenericT final : public TypeData {
TypeDataGenericT(uint64_t type_id, std::string&& nameT)
: TypeData(type_id, flag_contains_genericT_inside)
: TypeData(type_id, flag_contains_genericT_inside, -999999) // width undefined until instantiated
, nameT(std::move(nameT)) {}
public:
@ -286,7 +314,6 @@ public:
std::string as_human_readable() const override { return nameT; }
bool can_rhs_be_assigned(TypePtr rhs) const override;
bool can_be_casted_with_as_operator(TypePtr cast_to) const override;
int calc_width_on_stack() const override;
};
/*
@ -296,8 +323,8 @@ public:
* A tensor can be empty.
*/
class TypeDataTensor final : public TypeData {
TypeDataTensor(uint64_t type_id, int children_flags, std::vector<TypePtr>&& items)
: TypeData(type_id, children_flags)
TypeDataTensor(uint64_t type_id, int children_flags, int width_on_stack, std::vector<TypePtr>&& items)
: TypeData(type_id, children_flags, width_on_stack)
, items(std::move(items)) {}
public:
@ -312,7 +339,7 @@ public:
bool can_be_casted_with_as_operator(TypePtr cast_to) const override;
void traverse(const TraverserCallbackT& callback) const override;
TypePtr replace_children_custom(const ReplacerCallbackT& callback) const override;
int calc_width_on_stack() const override;
bool can_hold_tvm_null_instead() const override;
};
/*
@ -322,7 +349,7 @@ public:
*/
class TypeDataTypedTuple final : public TypeData {
TypeDataTypedTuple(uint64_t type_id, int children_flags, std::vector<TypePtr>&& items)
: TypeData(type_id, children_flags)
: TypeData(type_id, children_flags, 1)
, items(std::move(items)) {}
public:
@ -346,7 +373,7 @@ public:
* The only thing available to do with unknown is to cast it: `catch (excNo, arg) { var i = arg as int; }`
*/
class TypeDataUnknown final : public TypeData {
TypeDataUnknown() : TypeData(20ULL, flag_contains_unknown_inside) {}
TypeDataUnknown() : TypeData(20ULL, flag_contains_unknown_inside, 1) {}
static TypePtr singleton;
friend void type_system_init();
@ -367,7 +394,7 @@ public:
*/
class TypeDataUnresolved final : public TypeData {
TypeDataUnresolved(uint64_t type_id, std::string&& text, SrcLocation loc)
: TypeData(type_id, flag_contains_unresolved_inside)
: TypeData(type_id, flag_contains_unresolved_inside, -999999)
, text(std::move(text))
, loc(loc) {}
@ -380,7 +407,6 @@ public:
std::string as_human_readable() const override { return text + "*"; }
bool can_rhs_be_assigned(TypePtr rhs) const override;
bool can_be_casted_with_as_operator(TypePtr cast_to) const override;
int calc_width_on_stack() const override;
};
/*
@ -389,7 +415,7 @@ public:
* Empty tensor is not compatible with void, although at IR level they are similar, 0 stack slots.
*/
class TypeDataVoid final : public TypeData {
TypeDataVoid() : TypeData(10ULL, 0) {}
TypeDataVoid() : TypeData(10ULL, 0, 0) {}
static TypePtr singleton;
friend void type_system_init();
@ -400,7 +426,7 @@ public:
std::string as_human_readable() const override { return "void"; }
bool can_rhs_be_assigned(TypePtr rhs) const override;
bool can_be_casted_with_as_operator(TypePtr cast_to) const override;
int calc_width_on_stack() const override;
bool can_hold_tvm_null_instead() const override;
};