mirror of
https://github.com/LadybirdBrowser/ladybird.git
synced 2024-11-22 15:40:19 +00:00
1682f0b760
SPDX License Identifiers are a more compact / standardized way of representing file license information. See: https://spdx.dev/resources/use/#identifiers This was done with the `ambr` search and replace tool. ambr --no-parent-ignore --key-from-file --rep-from-file key.txt rep.txt *
229 lines
8.8 KiB
C++
229 lines
8.8 KiB
C++
/*
|
|
* Copyright (c) 2020, Stephan Unverwerth <s.unverwerth@gmx.de>
|
|
*
|
|
* SPDX-License-Identifier: BSD-2-Clause
|
|
*/
|
|
|
|
#pragma once
|
|
|
|
#include <AK/HashTable.h>
|
|
#include <AK/NonnullRefPtr.h>
|
|
#include <AK/StringBuilder.h>
|
|
#include <LibJS/AST.h>
|
|
#include <LibJS/Lexer.h>
|
|
#include <LibJS/SourceRange.h>
|
|
#include <stdio.h>
|
|
|
|
namespace JS {
|
|
|
|
enum class Associativity {
|
|
Left,
|
|
Right
|
|
};
|
|
|
|
struct FunctionNodeParseOptions {
|
|
enum {
|
|
CheckForFunctionAndName = 1 << 0,
|
|
AllowSuperPropertyLookup = 1 << 1,
|
|
AllowSuperConstructorCall = 1 << 2,
|
|
IsGetterFunction = 1 << 3,
|
|
IsSetterFunction = 1 << 4,
|
|
IsArrowFunction = 1 << 5,
|
|
};
|
|
};
|
|
|
|
class Parser {
|
|
public:
|
|
explicit Parser(Lexer lexer);
|
|
|
|
NonnullRefPtr<Program> parse_program();
|
|
|
|
template<typename FunctionNodeType>
|
|
NonnullRefPtr<FunctionNodeType> parse_function_node(u8 parse_options = FunctionNodeParseOptions::CheckForFunctionAndName);
|
|
Vector<FunctionNode::Parameter> parse_function_parameters(int& function_length, u8 parse_options = 0);
|
|
|
|
NonnullRefPtr<Declaration> parse_declaration();
|
|
NonnullRefPtr<Statement> parse_statement();
|
|
NonnullRefPtr<BlockStatement> parse_block_statement();
|
|
NonnullRefPtr<BlockStatement> parse_block_statement(bool& is_strict);
|
|
NonnullRefPtr<ReturnStatement> parse_return_statement();
|
|
NonnullRefPtr<VariableDeclaration> parse_variable_declaration(bool for_loop_variable_declaration = false);
|
|
NonnullRefPtr<Statement> parse_for_statement();
|
|
NonnullRefPtr<Statement> parse_for_in_of_statement(NonnullRefPtr<ASTNode> lhs);
|
|
NonnullRefPtr<IfStatement> parse_if_statement();
|
|
NonnullRefPtr<ThrowStatement> parse_throw_statement();
|
|
NonnullRefPtr<TryStatement> parse_try_statement();
|
|
NonnullRefPtr<CatchClause> parse_catch_clause();
|
|
NonnullRefPtr<SwitchStatement> parse_switch_statement();
|
|
NonnullRefPtr<SwitchCase> parse_switch_case();
|
|
NonnullRefPtr<BreakStatement> parse_break_statement();
|
|
NonnullRefPtr<ContinueStatement> parse_continue_statement();
|
|
NonnullRefPtr<DoWhileStatement> parse_do_while_statement();
|
|
NonnullRefPtr<WhileStatement> parse_while_statement();
|
|
NonnullRefPtr<WithStatement> parse_with_statement();
|
|
NonnullRefPtr<DebuggerStatement> parse_debugger_statement();
|
|
NonnullRefPtr<ConditionalExpression> parse_conditional_expression(NonnullRefPtr<Expression> test);
|
|
NonnullRefPtr<Expression> parse_expression(int min_precedence, Associativity associate = Associativity::Right, const Vector<TokenType>& forbidden = {});
|
|
NonnullRefPtr<Expression> parse_primary_expression();
|
|
NonnullRefPtr<Expression> parse_unary_prefixed_expression();
|
|
NonnullRefPtr<RegExpLiteral> parse_regexp_literal();
|
|
NonnullRefPtr<ObjectExpression> parse_object_expression();
|
|
NonnullRefPtr<ArrayExpression> parse_array_expression();
|
|
NonnullRefPtr<StringLiteral> parse_string_literal(const Token& token, bool in_template_literal = false);
|
|
NonnullRefPtr<TemplateLiteral> parse_template_literal(bool is_tagged);
|
|
NonnullRefPtr<Expression> parse_secondary_expression(NonnullRefPtr<Expression>, int min_precedence, Associativity associate = Associativity::Right);
|
|
NonnullRefPtr<CallExpression> parse_call_expression(NonnullRefPtr<Expression>);
|
|
NonnullRefPtr<NewExpression> parse_new_expression();
|
|
NonnullRefPtr<ClassDeclaration> parse_class_declaration();
|
|
NonnullRefPtr<ClassExpression> parse_class_expression(bool expect_class_name);
|
|
NonnullRefPtr<Expression> parse_property_key();
|
|
NonnullRefPtr<AssignmentExpression> parse_assignment_expression(AssignmentOp, NonnullRefPtr<Expression> lhs, int min_precedence, Associativity);
|
|
|
|
RefPtr<FunctionExpression> try_parse_arrow_function_expression(bool expect_parens);
|
|
RefPtr<Statement> try_parse_labelled_statement();
|
|
RefPtr<MetaProperty> try_parse_new_target_expression();
|
|
|
|
struct Error {
|
|
String message;
|
|
Optional<Position> position;
|
|
|
|
String to_string() const
|
|
{
|
|
if (!position.has_value())
|
|
return message;
|
|
return String::formatted("{} (line: {}, column: {})", message, position.value().line, position.value().column);
|
|
}
|
|
|
|
String source_location_hint(const StringView& source, const char spacer = ' ', const char indicator = '^') const
|
|
{
|
|
if (!position.has_value())
|
|
return {};
|
|
// We need to modify the source to match what the lexer considers one line - normalizing
|
|
// line terminators to \n is easier than splitting using all different LT characters.
|
|
String source_string { source };
|
|
source_string.replace("\r\n", "\n");
|
|
source_string.replace("\r", "\n");
|
|
source_string.replace(LINE_SEPARATOR, "\n");
|
|
source_string.replace(PARAGRAPH_SEPARATOR, "\n");
|
|
StringBuilder builder;
|
|
builder.append(source_string.split_view('\n', true)[position.value().line - 1]);
|
|
builder.append('\n');
|
|
for (size_t i = 0; i < position.value().column - 1; ++i)
|
|
builder.append(spacer);
|
|
builder.append(indicator);
|
|
return builder.build();
|
|
}
|
|
};
|
|
|
|
bool has_errors() const { return m_parser_state.m_errors.size(); }
|
|
const Vector<Error>& errors() const { return m_parser_state.m_errors; }
|
|
void print_errors() const
|
|
{
|
|
for (auto& error : m_parser_state.m_errors) {
|
|
auto hint = error.source_location_hint(m_parser_state.m_lexer.source());
|
|
if (!hint.is_empty())
|
|
warnln("{}", hint);
|
|
warnln("SyntaxError: {}", error.to_string());
|
|
}
|
|
}
|
|
|
|
struct TokenMemoization {
|
|
bool try_parse_arrow_function_expression_failed;
|
|
};
|
|
|
|
private:
|
|
friend class ScopePusher;
|
|
|
|
Associativity operator_associativity(TokenType) const;
|
|
bool match_expression() const;
|
|
bool match_unary_prefixed_expression() const;
|
|
bool match_secondary_expression(const Vector<TokenType>& forbidden = {}) const;
|
|
bool match_statement() const;
|
|
bool match_declaration() const;
|
|
bool match_variable_declaration() const;
|
|
bool match_identifier_name() const;
|
|
bool match_property_key() const;
|
|
bool match(TokenType type) const;
|
|
bool done() const;
|
|
void expected(const char* what);
|
|
void syntax_error(const String& message, Optional<Position> = {});
|
|
Token consume();
|
|
Token consume(TokenType type);
|
|
Token consume_and_validate_numeric_literal();
|
|
void consume_or_insert_semicolon();
|
|
void save_state();
|
|
void load_state();
|
|
void discard_saved_state();
|
|
Position position() const;
|
|
|
|
bool try_parse_arrow_function_expression_failed_at_position(const Position&) const;
|
|
void set_try_parse_arrow_function_expression_failed_at_position(const Position&, bool);
|
|
|
|
struct RulePosition {
|
|
AK_MAKE_NONCOPYABLE(RulePosition);
|
|
AK_MAKE_NONMOVABLE(RulePosition);
|
|
|
|
public:
|
|
RulePosition(Parser& parser, Position position)
|
|
: m_parser(parser)
|
|
, m_position(position)
|
|
{
|
|
m_parser.m_rule_starts.append(position);
|
|
}
|
|
|
|
~RulePosition()
|
|
{
|
|
auto last = m_parser.m_rule_starts.take_last();
|
|
VERIFY(last.line == m_position.line);
|
|
VERIFY(last.column == m_position.column);
|
|
}
|
|
|
|
const Position& position() const { return m_position; }
|
|
|
|
private:
|
|
Parser& m_parser;
|
|
Position m_position;
|
|
};
|
|
|
|
[[nodiscard]] RulePosition push_start() { return { *this, position() }; }
|
|
|
|
struct ParserState {
|
|
Lexer m_lexer;
|
|
Token m_current_token;
|
|
Vector<Error> m_errors;
|
|
Vector<NonnullRefPtrVector<VariableDeclaration>> m_var_scopes;
|
|
Vector<NonnullRefPtrVector<VariableDeclaration>> m_let_scopes;
|
|
Vector<NonnullRefPtrVector<FunctionDeclaration>> m_function_scopes;
|
|
HashTable<StringView> m_labels_in_scope;
|
|
bool m_strict_mode { false };
|
|
bool m_allow_super_property_lookup { false };
|
|
bool m_allow_super_constructor_call { false };
|
|
bool m_in_function_context { false };
|
|
bool m_in_arrow_function_context { false };
|
|
bool m_in_break_context { false };
|
|
bool m_in_continue_context { false };
|
|
bool m_string_legacy_octal_escape_sequence_in_scope { false };
|
|
|
|
explicit ParserState(Lexer);
|
|
};
|
|
|
|
class PositionKeyTraits {
|
|
public:
|
|
static int hash(const Position& position)
|
|
{
|
|
return int_hash(position.line) ^ int_hash(position.column);
|
|
}
|
|
|
|
static bool equals(const Position& a, const Position& b)
|
|
{
|
|
return a.column == b.column && a.line == b.line;
|
|
}
|
|
};
|
|
|
|
Vector<Position> m_rule_starts;
|
|
ParserState m_parser_state;
|
|
FlyString m_filename;
|
|
Vector<ParserState> m_saved_state;
|
|
HashMap<Position, TokenMemoization, PositionKeyTraits> m_token_memoizations;
|
|
};
|
|
}
|