diff options
Diffstat (limited to 'Userland/Libraries/LibWeb/CSS/Parser/Tokenizer.cpp')
-rw-r--r-- | Userland/Libraries/LibWeb/CSS/Parser/Tokenizer.cpp | 76 |
1 files changed, 38 insertions, 38 deletions
diff --git a/Userland/Libraries/LibWeb/CSS/Parser/Tokenizer.cpp b/Userland/Libraries/LibWeb/CSS/Parser/Tokenizer.cpp index 8a94c251d5..afad5ed6a5 100644 --- a/Userland/Libraries/LibWeb/CSS/Parser/Tokenizer.cpp +++ b/Userland/Libraries/LibWeb/CSS/Parser/Tokenizer.cpp @@ -262,14 +262,14 @@ U32Triplet Tokenizer::peek_triplet() const return values; } -Token Tokenizer::create_new_token(Token::TokenType type) +Token Tokenizer::create_new_token(Token::Type type) { Token token = {}; token.m_type = type; return token; } -Token Tokenizer::create_value_token(Token::TokenType type, String value) +Token Tokenizer::create_value_token(Token::Type type, String value) { Token token; token.m_type = type; @@ -277,7 +277,7 @@ Token Tokenizer::create_value_token(Token::TokenType type, String value) return token; } -Token Tokenizer::create_value_token(Token::TokenType type, u32 value) +Token Tokenizer::create_value_token(Token::Type type, u32 value) { Token token = {}; token.m_type = type; @@ -342,7 +342,7 @@ Token Tokenizer::consume_an_ident_like_token() auto next_two = peek_twin(); // if one of these ", ', ' "', " '" if (is_quotation_mark(next_two.first) || is_apostrophe(next_two.first) || (is_whitespace(next_two.first) && (is_quotation_mark(next_two.second) || is_apostrophe(next_two.second)))) { - return create_value_token(Token::TokenType::Function, string); + return create_value_token(Token::Type::Function, string); } return consume_a_url_token(); @@ -351,10 +351,10 @@ Token Tokenizer::consume_an_ident_like_token() if (is_left_paren(peek_code_point())) { (void)next_code_point(); - return create_value_token(Token::TokenType::Function, string); + return create_value_token(Token::Type::Function, string); } - return create_value_token(Token::TokenType::Ident, string); + return create_value_token(Token::Type::Ident, string); } CSSNumber Tokenizer::consume_a_number() @@ -447,7 +447,7 @@ String Tokenizer::consume_a_name() } Token Tokenizer::consume_a_url_token() { - auto token = create_new_token(Token::TokenType::Url); + auto token = create_new_token(Token::Type::Url); for (;;) { if (!is_whitespace(peek_code_point())) { break; @@ -488,14 +488,14 @@ Token Tokenizer::consume_a_url_token() } consume_the_remnants_of_a_bad_url(); - return create_new_token(Token::TokenType::BadUrl); + return create_new_token(Token::Type::BadUrl); } if (is_quotation_mark(input) || is_apostrophe(input) || is_left_paren(input) || is_non_printable(input)) { log_parse_error(); (void)next_code_point(); consume_the_remnants_of_a_bad_url(); - return create_new_token(Token::TokenType::BadUrl); + return create_new_token(Token::Type::BadUrl); } if (is_reverse_solidus(input)) { @@ -505,7 +505,7 @@ Token Tokenizer::consume_a_url_token() log_parse_error(); (void)next_code_point(); consume_the_remnants_of_a_bad_url(); - return create_new_token(Token::TokenType::BadUrl); + return create_new_token(Token::Type::BadUrl); } } @@ -546,7 +546,7 @@ Token Tokenizer::consume_a_numeric_token() { auto number = consume_a_number(); if (would_start_an_identifier()) { - auto token = create_new_token(Token::TokenType::Dimension); + auto token = create_new_token(Token::Type::Dimension); token.m_value.append(number.value); token.m_number_type = number.type; @@ -559,12 +559,12 @@ Token Tokenizer::consume_a_numeric_token() if (is_percent(peek_code_point())) { (void)next_code_point(); - auto token = create_new_token(Token::TokenType::Percentage); + auto token = create_new_token(Token::Type::Percentage); token.m_value.append(number.value); return token; } - auto token = create_new_token(Token::TokenType::Number); + auto token = create_new_token(Token::Type::Number); token.m_value.append(number.value); token.m_number_type = number.type; return token; @@ -642,7 +642,7 @@ bool Tokenizer::would_start_an_identifier(U32Triplet values) Token Tokenizer::consume_string_token(u32 ending_code_point) { - auto token = create_new_token(Token::TokenType::String); + auto token = create_new_token(Token::Type::String); for (;;) { auto input = next_code_point(); @@ -657,7 +657,7 @@ Token Tokenizer::consume_string_token(u32 ending_code_point) if (is_newline(input)) { reconsume_current_input_code_point(); - return create_new_token(Token::TokenType::BadString); + return create_new_token(Token::Type::BadString); } if (is_reverse_solidus(input)) { @@ -712,7 +712,7 @@ Token Tokenizer::consume_a_token() auto input = next_code_point(); if (is_eof(input)) { - return create_new_token(Token::TokenType::EndOfFile); + return create_new_token(Token::Type::EndOfFile); } if (is_whitespace(input)) { @@ -724,7 +724,7 @@ Token Tokenizer::consume_a_token() next = peek_code_point(); } - return create_new_token(Token::TokenType::Whitespace); + return create_new_token(Token::Type::Whitespace); } if (is_quotation_mark(input)) { @@ -739,7 +739,7 @@ Token Tokenizer::consume_a_token() auto maybe_escape = peek_twin(); if (is_name_code_point(next_input) || is_valid_escape_sequence(maybe_escape)) { - auto token = create_new_token(Token::TokenType::Hash); + auto token = create_new_token(Token::Type::Hash); if (would_start_an_identifier()) token.m_hash_type = Token::HashType::Id; @@ -750,7 +750,7 @@ Token Tokenizer::consume_a_token() return token; } - return create_value_token(Token::TokenType::Delim, input); + return create_value_token(Token::Type::Delim, input); } if (is_apostrophe(input)) { @@ -760,12 +760,12 @@ Token Tokenizer::consume_a_token() if (is_left_paren(input)) { dbgln_if(CSS_TOKENIZER_TRACE, "is left paren"); - return create_new_token(Token::TokenType::OpenParen); + return create_new_token(Token::Type::OpenParen); } if (is_right_paren(input)) { dbgln_if(CSS_TOKENIZER_TRACE, "is right paren"); - return create_new_token(Token::TokenType::CloseParen); + return create_new_token(Token::Type::CloseParen); } if (is_plus_sign(input)) { @@ -775,12 +775,12 @@ Token Tokenizer::consume_a_token() return consume_a_numeric_token(); } - return create_value_token(Token::TokenType::Delim, input); + return create_value_token(Token::Type::Delim, input); } if (is_comma(input)) { dbgln_if(CSS_TOKENIZER_TRACE, "is comma"); - return create_new_token(Token::TokenType::Comma); + return create_new_token(Token::Type::Comma); } if (is_hyphen_minus(input)) { @@ -795,7 +795,7 @@ Token Tokenizer::consume_a_token() (void)next_code_point(); (void)next_code_point(); - return create_new_token(Token::TokenType::CDC); + return create_new_token(Token::Type::CDC); } if (would_start_an_identifier()) { @@ -803,7 +803,7 @@ Token Tokenizer::consume_a_token() return consume_an_ident_like_token(); } - return create_value_token(Token::TokenType::Delim, input); + return create_value_token(Token::Type::Delim, input); } if (is_full_stop(input)) { @@ -813,17 +813,17 @@ Token Tokenizer::consume_a_token() return consume_a_numeric_token(); } - return create_value_token(Token::TokenType::Delim, input); + return create_value_token(Token::Type::Delim, input); } if (is_colon(input)) { dbgln_if(CSS_TOKENIZER_TRACE, "is colon"); - return create_new_token(Token::TokenType::Colon); + return create_new_token(Token::Type::Colon); } if (is_semicolon(input)) { dbgln_if(CSS_TOKENIZER_TRACE, "is semicolon"); - return create_new_token(Token::TokenType::Semicolon); + return create_new_token(Token::Type::Semicolon); } if (is_less_than_sign(input)) { @@ -835,10 +835,10 @@ Token Tokenizer::consume_a_token() (void)next_code_point(); (void)next_code_point(); - return create_new_token(Token::TokenType::CDO); + return create_new_token(Token::Type::CDO); } - return create_value_token(Token::TokenType::Delim, input); + return create_value_token(Token::Type::Delim, input); } if (is_at(input)) { @@ -846,15 +846,15 @@ Token Tokenizer::consume_a_token() if (would_start_an_identifier()) { auto name = consume_a_name(); - return create_value_token(Token::TokenType::AtKeyword, input); + return create_value_token(Token::Type::AtKeyword, input); } - return create_value_token(Token::TokenType::Delim, input); + return create_value_token(Token::Type::Delim, input); } if (is_open_square_bracket(input)) { dbgln_if(CSS_TOKENIZER_TRACE, "is open square"); - return create_new_token(Token::TokenType::OpenSquare); + return create_new_token(Token::Type::OpenSquare); } if (is_reverse_solidus(input)) { @@ -865,22 +865,22 @@ Token Tokenizer::consume_a_token() } log_parse_error(); - return create_value_token(Token::TokenType::Delim, input); + return create_value_token(Token::Type::Delim, input); } if (is_closed_square_bracket(input)) { dbgln_if(CSS_TOKENIZER_TRACE, "is closed square"); - return create_new_token(Token::TokenType::CloseSquare); + return create_new_token(Token::Type::CloseSquare); } if (is_open_curly_bracket(input)) { dbgln_if(CSS_TOKENIZER_TRACE, "is open curly"); - return create_new_token(Token::TokenType::OpenCurly); + return create_new_token(Token::Type::OpenCurly); } if (is_closed_curly_bracket(input)) { dbgln_if(CSS_TOKENIZER_TRACE, "is closed curly"); - return create_new_token(Token::TokenType::CloseCurly); + return create_new_token(Token::Type::CloseCurly); } if (is_ascii_digit(input)) { @@ -896,7 +896,7 @@ Token Tokenizer::consume_a_token() } dbgln_if(CSS_TOKENIZER_TRACE, "is delimiter"); - return create_value_token(Token::TokenType::Delim, input); + return create_value_token(Token::Type::Delim, input); } } |