diff options
Diffstat (limited to 'modules/gdscript/editor/gdscript_highlighter.cpp')
-rw-r--r-- | modules/gdscript/editor/gdscript_highlighter.cpp | 97 |
1 files changed, 52 insertions, 45 deletions
diff --git a/modules/gdscript/editor/gdscript_highlighter.cpp b/modules/gdscript/editor/gdscript_highlighter.cpp index afb59b486c..8b27307d0c 100644 --- a/modules/gdscript/editor/gdscript_highlighter.cpp +++ b/modules/gdscript/editor/gdscript_highlighter.cpp @@ -39,30 +39,32 @@ Dictionary GDScriptSyntaxHighlighter::_get_line_syntax_highlighting_impl(int p_l Type next_type = NONE; Type current_type = NONE; - Type previous_type = NONE; - - String previous_text = ""; - int previous_column = 0; + Type prev_type = NONE; + String prev_text = ""; + int prev_column = 0; bool prev_is_char = false; bool prev_is_digit = false; bool prev_is_binary_op = false; + bool in_keyword = false; bool in_word = false; bool in_number = false; - bool in_function_name = false; - bool in_lambda = false; - bool in_variable_declaration = false; - bool in_signal_declaration = false; - bool in_function_args = false; - bool in_member_variable = false; bool in_node_path = false; bool in_node_ref = false; bool in_annotation = false; bool in_string_name = false; bool is_hex_notation = false; bool is_bin_notation = false; + bool in_member_variable = false; + bool in_lambda = false; + + bool in_function_name = false; + bool in_function_args = false; + bool in_variable_declaration = false; + bool in_signal_declaration = false; bool expect_type = false; + Color keyword_color; Color color; @@ -224,9 +226,9 @@ Dictionary GDScriptSyntaxHighlighter::_get_line_syntax_highlighting_impl(int p_l } } - previous_type = REGION; - previous_text = ""; - previous_column = j; + prev_type = REGION; + prev_text = ""; + prev_column = j; j = from + (end_key_length - 1); if (region_end_index == -1) { color_region_cache[p_line] = in_region; @@ -241,19 +243,22 @@ Dictionary GDScriptSyntaxHighlighter::_get_line_syntax_highlighting_impl(int p_l } } - // A bit of a hack, but couldn't come up with anything better. + // VERY hacky... but couldn't come up with anything better. if (j > 0 && (str[j] == '&' || str[j] == '^' || str[j] == '%' || str[j] == '+' || str[j] == '-' || str[j] == '~' || str[j] == '.')) { - if (!keywords.has(previous_text)) { - if (previous_text == "PI" || previous_text == "TAU" || previous_text == "INF" || previous_text == "NAN") { + int to = j - 1; + // Find what the last text was (prev_text won't work if there's no whitespace, so we need to do it manually). + while (to > 0 && is_whitespace(str[to])) { + to--; + } + int from = to; + while (from > 0 && !is_symbol(str[from])) { + from--; + } + String word = str.substr(from + 1, to - from); + // Keywords need to be exceptions, except for keywords that represent a value. + if (word == "true" || word == "false" || word == "null" || word == "PI" || word == "TAU" || word == "INF" || word == "NAN" || word == "self" || word == "super" || !keywords.has(word)) { + if (!is_symbol(str[to]) || str[to] == '"' || str[to] == '\'' || str[to] == ')' || str[to] == ']' || str[to] == '}') { is_binary_op = true; - } else { - int k = j - 1; - while (k > 0 && is_whitespace(str[k])) { - k--; - } - if (!is_symbol(str[k]) || str[k] == '"' || str[k] == '\'' || str[k] == ')' || str[k] == ']' || str[k] == '}') { - is_binary_op = true; - } } } } @@ -287,16 +292,18 @@ Dictionary GDScriptSyntaxHighlighter::_get_line_syntax_highlighting_impl(int p_l is_hex_notation = true; } else if (!((str[j] == '-' || str[j] == '+') && str[j - 1] == 'e' && !prev_is_digit) && !(str[j] == '_' && (prev_is_digit || str[j - 1] == 'b' || str[j - 1] == 'x' || str[j - 1] == '.')) && - !((str[j] == 'e' || str[j] == '.') && (prev_is_digit || str[j - 1] == '_')) && + !(str[j] == 'e' && (prev_is_digit || str[j - 1] == '_')) && + !(str[j] == '.' && (prev_is_digit || (!prev_is_binary_op && (j > 0 && (str[j - 1] == '-' || str[j - 1] == '+' || str[j - 1] == '~'))))) && !((str[j] == '-' || str[j] == '+' || str[j] == '~') && !prev_is_binary_op && str[j - 1] != 'e')) { - /* 1st row of condition: '+' or '-' after scientific notation; - 2nd row of condition: '_' as a numeric separator; - 3rd row of condition: Scientific notation 'e' and floating points; - 4th row of condition: Multiple unary operators. */ + /* This condition continues Number highlighting in special cases. + 1st row: '+' or '-' after scientific notation; + 2nd row: '_' as a numeric separator; + 3rd row: Scientific notation 'e' and floating points; + 4th row: Floating points inside the number, or leading if after a unary mathematical operator; + 5th row: Multiple unary mathematical operators */ in_number = false; } - } else if ((str[j] == '-' || str[j] == '+' || str[j] == '~' || (str[j] == '.' && str[j + 1] != '.' && (j == 0 || (j > 0 && str[j - 1] != '.')))) && !is_binary_op) { - // Start a number from unary mathematical operators and floating points, except for '..' + } else if (!is_binary_op && (str[j] == '-' || str[j] == '+' || str[j] == '~' || (str[j] == '.' && str[j + 1] != '.' && (j == 0 || (j > 0 && str[j - 1] != '.'))))) { in_number = true; } @@ -318,7 +325,7 @@ Dictionary GDScriptSyntaxHighlighter::_get_line_syntax_highlighting_impl(int p_l Color col = Color(); if (global_functions.has(word)) { // "assert" and "preload" are reserved, so highlight even if not followed by a bracket. - if (word == "assert" || word == "preload") { + if (word == GDScriptTokenizer::get_token_name(GDScriptTokenizer::Token::ASSERT) || word == GDScriptTokenizer::get_token_name(GDScriptTokenizer::Token::PRELOAD)) { col = global_function_color; } else { // For other global functions, check if followed by bracket. @@ -355,7 +362,7 @@ Dictionary GDScriptSyntaxHighlighter::_get_line_syntax_highlighting_impl(int p_l } if (!in_function_name && in_word && !in_keyword) { - if (previous_text == GDScriptTokenizer::get_token_name(GDScriptTokenizer::Token::SIGNAL)) { + if (prev_text == GDScriptTokenizer::get_token_name(GDScriptTokenizer::Token::SIGNAL)) { in_signal_declaration = true; } else { int k = j; @@ -370,12 +377,12 @@ Dictionary GDScriptSyntaxHighlighter::_get_line_syntax_highlighting_impl(int p_l if (str[k] == '(') { in_function_name = true; - } else if (previous_text == GDScriptTokenizer::get_token_name(GDScriptTokenizer::Token::VAR)) { + } else if (prev_text == GDScriptTokenizer::get_token_name(GDScriptTokenizer::Token::VAR)) { in_variable_declaration = true; } // Check for lambda. - if (in_function_name && previous_text == GDScriptTokenizer::get_token_name(GDScriptTokenizer::Token::FUNC)) { + if (in_function_name && prev_text == GDScriptTokenizer::get_token_name(GDScriptTokenizer::Token::FUNC)) { k = j - 1; while (k > 0 && is_whitespace(str[k])) { k--; @@ -408,11 +415,11 @@ Dictionary GDScriptSyntaxHighlighter::_get_line_syntax_highlighting_impl(int p_l in_function_args = false; } - if (expect_type && (prev_is_char || str[j] == '=')) { + if (expect_type && (prev_is_char || str[j] == '=') && str[j] != '[') { expect_type = false; } - if (j > 0 && str[j] == '>' && str[j - 1] == '-') { + if (j > 0 && str[j - 1] == '-' && str[j] == '>') { expect_type = true; } @@ -491,7 +498,7 @@ Dictionary GDScriptSyntaxHighlighter::_get_line_syntax_highlighting_impl(int p_l } else if (in_function_name) { next_type = FUNCTION; - if (!in_lambda && previous_text == GDScriptTokenizer::get_token_name(GDScriptTokenizer::Token::FUNC)) { + if (!in_lambda && prev_text == GDScriptTokenizer::get_token_name(GDScriptTokenizer::Token::FUNC)) { color = function_definition_color; } else { color = function_color; @@ -513,20 +520,20 @@ Dictionary GDScriptSyntaxHighlighter::_get_line_syntax_highlighting_impl(int p_l if (current_type == NONE) { current_type = next_type; } else { - previous_type = current_type; + prev_type = current_type; current_type = next_type; // no need to store regions... - if (previous_type == REGION) { - previous_text = ""; - previous_column = j; + if (prev_type == REGION) { + prev_text = ""; + prev_column = j; } else { - String text = str.substr(previous_column, j - previous_column).strip_edges(); - previous_column = j; + String text = str.substr(prev_column, j - prev_column).strip_edges(); + prev_column = j; // ignore if just whitespace if (!text.is_empty()) { - previous_text = text; + prev_text = text; } } } |