You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
notepad-plus-plus/lexilla/lexers/LexGDScript.cxx

774 lines
24 KiB

// Scintilla source code edit control
/** @file LexGDScript.cxx
** Lexer for GDScript.
**/
// Copyright 1998-2002 by Neil Hodgson <neilh@scintilla.org>
// Heavily modified later for GDScript
// The License.txt file describes the conditions under which this software may be distributed.
#include <cstdlib>
#include <cassert>
#include <cstring>
#include <string>
#include <string_view>
#include <vector>
#include <map>
#include <algorithm>
#include <functional>
#include "ILexer.h"
#include "Scintilla.h"
#include "SciLexer.h"
#include "StringCopy.h"
#include "WordList.h"
#include "LexAccessor.h"
#include "Accessor.h"
#include "StyleContext.h"
#include "CharacterSet.h"
#include "CharacterCategory.h"
#include "LexerModule.h"
#include "OptionSet.h"
#include "SubStyles.h"
#include "DefaultLexer.h"
using namespace Scintilla;
using namespace Lexilla;
namespace {
enum kwType { kwOther, kwClass, kwDef, kwExtends};
constexpr int indicatorWhitespace = 1;
bool IsGDStringStart(int ch) {
return (ch == '\'' || ch == '"');
}
bool IsGDComment(Accessor &styler, Sci_Position pos, Sci_Position len) {
return len > 0 && styler[pos] == '#';
}
constexpr bool IsGDSingleQuoteStringState(int st) noexcept {
return ((st == SCE_GD_CHARACTER) || (st == SCE_GD_STRING));
}
constexpr bool IsGDTripleQuoteStringState(int st) noexcept {
return ((st == SCE_GD_TRIPLE) || (st == SCE_GD_TRIPLEDOUBLE));
}
char GetGDStringQuoteChar(int st) noexcept {
if ((st == SCE_GD_CHARACTER) || (st == SCE_GD_TRIPLE))
return '\'';
if ((st == SCE_GD_STRING) || (st == SCE_GD_TRIPLEDOUBLE))
return '"';
return '\0';
}
/* Return the state to use for the string starting at i; *nextIndex will be set to the first index following the quote(s) */
int GetGDStringState(Accessor &styler, Sci_Position i, Sci_PositionU *nextIndex) {
char ch = styler.SafeGetCharAt(i);
char chNext = styler.SafeGetCharAt(i + 1);
if (ch != '"' && ch != '\'') {
*nextIndex = i + 1;
return SCE_GD_DEFAULT;
}
if (ch == chNext && ch == styler.SafeGetCharAt(i + 2)) {
*nextIndex = i + 3;
if (ch == '"')
return SCE_GD_TRIPLEDOUBLE;
else
return SCE_GD_TRIPLE;
} else {
*nextIndex = i + 1;
if (ch == '"')
return SCE_GD_STRING;
else
return SCE_GD_CHARACTER;
}
}
int GetGDStringState(int ch) {
if (ch != '"' && ch != '\'')
return SCE_GD_DEFAULT;
if (ch == '"')
return SCE_GD_STRING;
else
return SCE_GD_CHARACTER;
}
inline bool IsAWordChar(int ch, bool unicodeIdentifiers) {
if (IsASCII(ch))
return (IsAlphaNumeric(ch) || ch == '.' || ch == '_');
if (!unicodeIdentifiers)
return false;
return IsXidContinue(ch);
}
inline bool IsANodePathChar(int ch, bool unicodeIdentifiers) {
if (IsASCII(ch))
return (IsAlphaNumeric(ch) || ch == '_' || ch == '/' || ch =='%');
if (!unicodeIdentifiers)
return false;
return IsXidContinue(ch);
}
inline bool IsAWordStart(int ch, bool unicodeIdentifiers) {
if (IsASCII(ch))
return (IsUpperOrLowerCase(ch) || ch == '_');
if (!unicodeIdentifiers)
return false;
return IsXidStart(ch);
}
bool IsFirstNonWhitespace(Sci_Position pos, Accessor &styler) {
const Sci_Position line = styler.GetLine(pos);
const Sci_Position start_pos = styler.LineStart(line);
for (Sci_Position i = start_pos; i < pos; i++) {
const char ch = styler[i];
if (!(ch == ' ' || ch == '\t'))
return false;
}
return true;
}
// Options used for LexerGDScript
struct OptionsGDScript {
int whingeLevel;
bool base2or8Literals;
bool stringsOverNewline;
bool keywords2NoSubIdentifiers;
bool fold;
bool foldQuotes;
bool foldCompact;
bool unicodeIdentifiers;
OptionsGDScript() noexcept {
whingeLevel = 0;
base2or8Literals = true;
stringsOverNewline = false;
keywords2NoSubIdentifiers = false;
fold = false;
foldQuotes = false;
foldCompact = false;
unicodeIdentifiers = true;
}
};
const char *const gdscriptWordListDesc[] = {
"Keywords",
"Highlighted identifiers",
nullptr
};
struct OptionSetGDScript : public OptionSet<OptionsGDScript> {
OptionSetGDScript() {
DefineProperty("lexer.gdscript.whinge.level", &OptionsGDScript::whingeLevel,
"For GDScript code, checks whether indenting is consistent. "
"The default, 0 turns off indentation checking, "
"1 checks whether each line is potentially inconsistent with the previous line, "
"2 checks whether any space characters occur before a tab character in the indentation, "
"3 checks whether any spaces are in the indentation, and "
"4 checks for any tab characters in the indentation. "
"1 is a good level to use.");
DefineProperty("lexer.gdscript.literals.binary", &OptionsGDScript::base2or8Literals,
"Set to 0 to not recognise binary and octal literals: 0b1011 0o712.");
DefineProperty("lexer.gdscript.strings.over.newline", &OptionsGDScript::stringsOverNewline,
"Set to 1 to allow strings to span newline characters.");
DefineProperty("lexer.gdscript.keywords2.no.sub.identifiers", &OptionsGDScript::keywords2NoSubIdentifiers,
"When enabled, it will not style keywords2 items that are used as a sub-identifier. "
"Example: when set, will not highlight \"foo.open\" when \"open\" is a keywords2 item.");
DefineProperty("fold", &OptionsGDScript::fold);
DefineProperty("fold.gdscript.quotes", &OptionsGDScript::foldQuotes,
"This option enables folding multi-line quoted strings when using the GDScript lexer.");
DefineProperty("fold.compact", &OptionsGDScript::foldCompact);
DefineProperty("lexer.gdscript.unicode.identifiers", &OptionsGDScript::unicodeIdentifiers,
"Set to 0 to not recognise Unicode identifiers.");
DefineWordListSets(gdscriptWordListDesc);
}
};
const char styleSubable[] = { SCE_GD_IDENTIFIER, 0 };
LexicalClass lexicalClasses[] = {
// Lexer GDScript SCLEX_GDSCRIPT SCE_GD_:
0, "SCE_GD_DEFAULT", "default", "White space",
1, "SCE_GD_COMMENTLINE", "comment line", "Comment",
2, "SCE_GD_NUMBER", "literal numeric", "Number",
3, "SCE_GD_STRING", "literal string", "String",
4, "SCE_GD_CHARACTER", "literal string", "Single quoted string",
5, "SCE_GD_WORD", "keyword", "Keyword",
6, "SCE_GD_TRIPLE", "literal string", "Triple quotes",
7, "SCE_GD_TRIPLEDOUBLE", "literal string", "Triple double quotes",
8, "SCE_GD_CLASSNAME", "identifier", "Class name definition",
9, "SCE_GD_FUNCNAME", "identifier", "Function or method name definition",
10, "SCE_GD_OPERATOR", "operator", "Operators",
11, "SCE_GD_IDENTIFIER", "identifier", "Identifiers",
12, "SCE_GD_COMMENTBLOCK", "comment", "Comment-blocks",
13, "SCE_GD_STRINGEOL", "error literal string", "End of line where string is not closed",
14, "SCE_GD_WORD2", "identifier", "Highlighted identifiers",
15, "SCE_GD_ANNOTATION", "annotation", "Annotations",
16, "SCE_GD_NODEPATH", "path", "Node path",
};
}
class LexerGDScript : public DefaultLexer {
WordList keywords;
WordList keywords2;
OptionsGDScript options;
OptionSetGDScript osGDScript;
enum { ssIdentifier };
SubStyles subStyles;
public:
explicit LexerGDScript() :
DefaultLexer("gdscript", SCLEX_GDSCRIPT, lexicalClasses, ELEMENTS(lexicalClasses)),
subStyles(styleSubable, 0x80, 0x40, 0) {
}
~LexerGDScript() override {
}
void SCI_METHOD Release() override {
delete this;
}
int SCI_METHOD Version() const override {
return lvRelease5;
}
const char *SCI_METHOD PropertyNames() override {
return osGDScript.PropertyNames();
}
int SCI_METHOD PropertyType(const char *name) override {
return osGDScript.PropertyType(name);
}
const char *SCI_METHOD DescribeProperty(const char *name) override {
return osGDScript.DescribeProperty(name);
}
Sci_Position SCI_METHOD PropertySet(const char *key, const char *val) override;
const char * SCI_METHOD PropertyGet(const char *key) override {
return osGDScript.PropertyGet(key);
}
const char *SCI_METHOD DescribeWordListSets() override {
return osGDScript.DescribeWordListSets();
}
Sci_Position SCI_METHOD WordListSet(int n, const char *wl) override;
void SCI_METHOD Lex(Sci_PositionU startPos, Sci_Position length, int initStyle, IDocument *pAccess) override;
void SCI_METHOD Fold(Sci_PositionU startPos, Sci_Position length, int initStyle, IDocument *pAccess) override;
void *SCI_METHOD PrivateCall(int, void *) override {
return nullptr;
}
int SCI_METHOD LineEndTypesSupported() override {
return SC_LINE_END_TYPE_UNICODE;
}
int SCI_METHOD AllocateSubStyles(int styleBase, int numberStyles) override {
return subStyles.Allocate(styleBase, numberStyles);
}
int SCI_METHOD SubStylesStart(int styleBase) override {
return subStyles.Start(styleBase);
}
int SCI_METHOD SubStylesLength(int styleBase) override {
return subStyles.Length(styleBase);
}
int SCI_METHOD StyleFromSubStyle(int subStyle) override {
const int styleBase = subStyles.BaseStyle(subStyle);
return styleBase;
}
int SCI_METHOD PrimaryStyleFromStyle(int style) override {
return style;
}
void SCI_METHOD FreeSubStyles() override {
subStyles.Free();
}
void SCI_METHOD SetIdentifiers(int style, const char *identifiers) override {
subStyles.SetIdentifiers(style, identifiers);
}
int SCI_METHOD DistanceToSecondaryStyles() override {
return 0;
}
const char *SCI_METHOD GetSubStyleBases() override {
return styleSubable;
}
static ILexer5 *LexerFactoryGDScript() {
return new LexerGDScript();
}
private:
void ProcessLineEnd(StyleContext &sc, bool &inContinuedString);
};
Sci_Position SCI_METHOD LexerGDScript::PropertySet(const char *key, const char *val) {
if (osGDScript.PropertySet(&options, key, val)) {
return 0;
}
return -1;
}
Sci_Position SCI_METHOD LexerGDScript::WordListSet(int n, const char *wl) {
WordList *wordListN = nullptr;
switch (n) {
case 0:
wordListN = &keywords;
break;
case 1:
wordListN = &keywords2;
break;
default:
break;
}
Sci_Position firstModification = -1;
if (wordListN) {
WordList wlNew;
wlNew.Set(wl);
if (*wordListN != wlNew) {
wordListN->Set(wl);
firstModification = 0;
}
}
return firstModification;
}
void LexerGDScript::ProcessLineEnd(StyleContext &sc, bool &inContinuedString) {
if ((sc.state == SCE_GD_DEFAULT)
|| IsGDTripleQuoteStringState(sc.state)) {
// Perform colourisation of white space and triple quoted strings at end of each line to allow
// tab marking to work inside white space and triple quoted strings
sc.SetState(sc.state);
}
if (IsGDSingleQuoteStringState(sc.state)) {
if (inContinuedString || options.stringsOverNewline) {
inContinuedString = false;
} else {
sc.ChangeState(SCE_GD_STRINGEOL);
sc.ForwardSetState(SCE_GD_DEFAULT);
}
}
}
void SCI_METHOD LexerGDScript::Lex(Sci_PositionU startPos, Sci_Position length, int initStyle, IDocument *pAccess) {
Accessor styler(pAccess, nullptr);
const Sci_Position endPos = startPos + length;
// Backtrack to previous line in case need to fix its tab whinging
Sci_Position lineCurrent = styler.GetLine(startPos);
if (startPos > 0) {
if (lineCurrent > 0) {
lineCurrent--;
// Look for backslash-continued lines
while (lineCurrent > 0) {
const Sci_Position eolPos = styler.LineStart(lineCurrent) - 1;
const int eolStyle = styler.StyleAt(eolPos);
if (eolStyle == SCE_GD_STRING || eolStyle == SCE_GD_CHARACTER
|| eolStyle == SCE_GD_STRINGEOL) {
lineCurrent -= 1;
} else {
break;
}
}
startPos = styler.LineStart(lineCurrent);
}
initStyle = startPos == 0 ? SCE_GD_DEFAULT : styler.StyleAt(startPos - 1);
}
initStyle = initStyle & 31;
if (initStyle == SCE_GD_STRINGEOL) {
initStyle = SCE_GD_DEFAULT;
}
kwType kwLast = kwOther;
int spaceFlags = 0;
styler.IndentAmount(lineCurrent, &spaceFlags, IsGDComment);
bool base_n_number = false;
const WordClassifier &classifierIdentifiers = subStyles.Classifier(SCE_GD_IDENTIFIER);
StyleContext sc(startPos, endPos - startPos, initStyle, styler);
bool indentGood = true;
Sci_Position startIndicator = sc.currentPos;
bool inContinuedString = false;
bool percentIsNodePath = false;
int nodePathStringState = SCE_GD_DEFAULT;
for (; sc.More(); sc.Forward()) {
if (sc.atLineStart) {
styler.IndentAmount(lineCurrent, &spaceFlags, IsGDComment);
indentGood = true;
if (options.whingeLevel == 1) {
indentGood = (spaceFlags & wsInconsistent) == 0;
} else if (options.whingeLevel == 2) {
indentGood = (spaceFlags & wsSpaceTab) == 0;
} else if (options.whingeLevel == 3) {
indentGood = (spaceFlags & wsSpace) == 0;
} else if (options.whingeLevel == 4) {
indentGood = (spaceFlags & wsTab) == 0;
}
if (!indentGood) {
styler.IndicatorFill(startIndicator, sc.currentPos, indicatorWhitespace, 0);
startIndicator = sc.currentPos;
}
}
if (sc.atLineEnd) {
percentIsNodePath = false;
ProcessLineEnd(sc, inContinuedString);
lineCurrent++;
if (!sc.More())
break;
}
bool needEOLCheck = false;
if (sc.state == SCE_GD_OPERATOR) {
kwLast = kwOther;
sc.SetState(SCE_GD_DEFAULT);
} else if (sc.state == SCE_GD_NUMBER) {
if (!IsAWordChar(sc.ch, false) &&
!(!base_n_number && ((sc.ch == '+' || sc.ch == '-') && (sc.chPrev == 'e' || sc.chPrev == 'E')))) {
sc.SetState(SCE_GD_DEFAULT);
}
} else if (sc.state == SCE_GD_IDENTIFIER) {
if ((sc.ch == '.') || (!IsAWordChar(sc.ch, options.unicodeIdentifiers))) {
char s[100];
sc.GetCurrent(s, sizeof(s));
int style = SCE_GD_IDENTIFIER;
if (keywords.InList(s)) {
style = SCE_GD_WORD;
} else if (kwLast == kwClass) {
style = SCE_GD_CLASSNAME;
} else if (kwLast == kwDef) {
style = SCE_GD_FUNCNAME;
} else if (keywords2.InList(s)) {
if (options.keywords2NoSubIdentifiers) {
// We don't want to highlight keywords2
// that are used as a sub-identifier,
// i.e. not open in "foo.open".
const Sci_Position pos = styler.GetStartSegment() - 1;
if (pos < 0 || (styler.SafeGetCharAt(pos, '\0') != '.'))
style = SCE_GD_WORD2;
} else {
style = SCE_GD_WORD2;
}
} else {
const int subStyle = classifierIdentifiers.ValueFor(s);
if (subStyle >= 0) {
style = subStyle;
}
}
sc.ChangeState(style);
sc.SetState(SCE_GD_DEFAULT);
if (style == SCE_GD_WORD) {
if (0 == strcmp(s, "class"))
kwLast = kwClass;
else if (0 == strcmp(s, "func"))
kwLast = kwDef;
else if (0 == strcmp(s, "extends"))
kwLast = kwExtends;
else
kwLast = kwOther;
} else {
kwLast = kwOther;
}
}
} else if ((sc.state == SCE_GD_COMMENTLINE) || (sc.state == SCE_GD_COMMENTBLOCK)) {
if (sc.ch == '\r' || sc.ch == '\n') {
sc.SetState(SCE_GD_DEFAULT);
}
} else if (sc.state == SCE_GD_ANNOTATION) {
if (!IsAWordStart(sc.ch, options.unicodeIdentifiers)) {
sc.SetState(SCE_GD_DEFAULT);
}
} else if (sc.state == SCE_GD_NODEPATH) {
if (nodePathStringState != SCE_GD_DEFAULT) {
if (sc.ch == GetGDStringQuoteChar(nodePathStringState) ) {
nodePathStringState = SCE_GD_DEFAULT;
}
} else {
if (IsGDStringStart(sc.ch)) {
nodePathStringState = GetGDStringState(sc.ch);
} else if (!IsANodePathChar(sc.ch, options.unicodeIdentifiers)) {
sc.SetState(SCE_GD_DEFAULT);
}
}
} else if (IsGDSingleQuoteStringState(sc.state)) {
if (sc.ch == '\\') {
if ((sc.chNext == '\r') && (sc.GetRelative(2) == '\n')) {
sc.Forward();
}
if (sc.chNext == '\n' || sc.chNext == '\r') {
inContinuedString = true;
} else {
// Don't roll over the newline.
sc.Forward();
}
} else if (sc.ch == GetGDStringQuoteChar(sc.state)) {
sc.ForwardSetState(SCE_GD_DEFAULT);
needEOLCheck = true;
}
} else if (sc.state == SCE_GD_TRIPLE) {
if (sc.ch == '\\') {
sc.Forward();
} else if (sc.Match(R"(''')")) {
sc.Forward();
sc.Forward();
sc.ForwardSetState(SCE_GD_DEFAULT);
needEOLCheck = true;
}
} else if (sc.state == SCE_GD_TRIPLEDOUBLE) {
if (sc.ch == '\\') {
sc.Forward();
} else if (sc.Match(R"(""")")) {
sc.Forward();
sc.Forward();
sc.ForwardSetState(SCE_GD_DEFAULT);
needEOLCheck = true;
}
}
if (!indentGood && !IsASpaceOrTab(sc.ch)) {
styler.IndicatorFill(startIndicator, sc.currentPos, indicatorWhitespace, 1);
startIndicator = sc.currentPos;
indentGood = true;
}
// State exit code may have moved on to end of line
if (needEOLCheck && sc.atLineEnd) {
ProcessLineEnd(sc, inContinuedString);
lineCurrent++;
styler.IndentAmount(lineCurrent, &spaceFlags, IsGDComment);
if (!sc.More())
break;
}
// Check for a new state starting character
if (sc.state == SCE_GD_DEFAULT) {
if (IsADigit(sc.ch) || (sc.ch == '.' && IsADigit(sc.chNext))) {
if (sc.ch == '0' && (sc.chNext == 'x' || sc.chNext == 'X')) {
base_n_number = true;
sc.SetState(SCE_GD_NUMBER);
} else if (sc.ch == '0' &&
(sc.chNext == 'o' || sc.chNext == 'O' || sc.chNext == 'b' || sc.chNext == 'B')) {
if (options.base2or8Literals) {
base_n_number = true;
sc.SetState(SCE_GD_NUMBER);
} else {
sc.SetState(SCE_GD_NUMBER);
sc.ForwardSetState(SCE_GD_IDENTIFIER);
}
} else {
base_n_number = false;
sc.SetState(SCE_GD_NUMBER);
}
} else if ((sc.ch == '$') || (sc.ch == '%' && (percentIsNodePath || IsFirstNonWhitespace(sc.currentPos, styler)))) {
percentIsNodePath = false;
sc.SetState(SCE_GD_NODEPATH);
} else if (isoperator(sc.ch) || sc.ch == '`') {
percentIsNodePath = !((sc.ch == ')') || (sc.ch == ']') || (sc.ch == '}'));
sc.SetState(SCE_GD_OPERATOR);
} else if (sc.ch == '#') {
sc.SetState(sc.chNext == '#' ? SCE_GD_COMMENTBLOCK : SCE_GD_COMMENTLINE);
} else if (sc.ch == '@') {
if (IsFirstNonWhitespace(sc.currentPos, styler))
sc.SetState(SCE_GD_ANNOTATION);
else
sc.SetState(SCE_GD_OPERATOR);
} else if (IsGDStringStart(sc.ch)) {
Sci_PositionU nextIndex = 0;
sc.SetState(GetGDStringState(styler, sc.currentPos, &nextIndex));
while (nextIndex > (sc.currentPos + 1) && sc.More()) {
sc.Forward();
}
} else if (IsAWordStart(sc.ch, options.unicodeIdentifiers)) {
sc.SetState(SCE_GD_IDENTIFIER);
}
}
}
styler.IndicatorFill(startIndicator, sc.currentPos, indicatorWhitespace, 0);
sc.Complete();
}
static bool IsCommentLine(Sci_Position line, Accessor &styler) {
const Sci_Position pos = styler.LineStart(line);
const Sci_Position eol_pos = styler.LineStart(line + 1) - 1;
for (Sci_Position i = pos; i < eol_pos; i++) {
const char ch = styler[i];
if (ch == '#')
return true;
else if (ch != ' ' && ch != '\t')
return false;
}
return false;
}
static bool IsQuoteLine(Sci_Position line, const Accessor &styler) {
const int style = styler.StyleAt(styler.LineStart(line)) & 31;
return IsGDTripleQuoteStringState(style);
}
void SCI_METHOD LexerGDScript::Fold(Sci_PositionU startPos, Sci_Position length, int /*initStyle - unused*/, IDocument *pAccess) {
if (!options.fold)
return;
Accessor styler(pAccess, nullptr);
const Sci_Position maxPos = startPos + length;
const Sci_Position maxLines = (maxPos == styler.Length()) ? styler.GetLine(maxPos) : styler.GetLine(maxPos - 1); // Requested last line
const Sci_Position docLines = styler.GetLine(styler.Length()); // Available last line
// Backtrack to previous non-blank line so we can determine indent level
// for any white space lines (needed esp. within triple quoted strings)
// and so we can fix any preceding fold level (which is why we go back
// at least one line in all cases)
int spaceFlags = 0;
Sci_Position lineCurrent = styler.GetLine(startPos);
int indentCurrent = styler.IndentAmount(lineCurrent, &spaceFlags, nullptr);
while (lineCurrent > 0) {
lineCurrent--;
indentCurrent = styler.IndentAmount(lineCurrent, &spaceFlags, nullptr);
if (!(indentCurrent & SC_FOLDLEVELWHITEFLAG) &&
(!IsCommentLine(lineCurrent, styler)) &&
(!IsQuoteLine(lineCurrent, styler)))
break;
}
int indentCurrentLevel = indentCurrent & SC_FOLDLEVELNUMBERMASK;
// Set up initial loop state
startPos = styler.LineStart(lineCurrent);
int prev_state = SCE_GD_DEFAULT & 31;
if (lineCurrent >= 1)
prev_state = styler.StyleAt(startPos - 1) & 31;
int prevQuote = options.foldQuotes && IsGDTripleQuoteStringState(prev_state);
// Process all characters to end of requested range or end of any triple quote
//that hangs over the end of the range. Cap processing in all cases
// to end of document (in case of unclosed quote at end).
while ((lineCurrent <= docLines) && ((lineCurrent <= maxLines) || prevQuote)) {
// Gather info
int lev = indentCurrent;
Sci_Position lineNext = lineCurrent + 1;
int indentNext = indentCurrent;
int quote = false;
if (lineNext <= docLines) {
// Information about next line is only available if not at end of document
indentNext = styler.IndentAmount(lineNext, &spaceFlags, nullptr);
const Sci_Position lookAtPos = (styler.LineStart(lineNext) == styler.Length()) ? styler.Length() - 1 : styler.LineStart(lineNext);
const int style = styler.StyleAt(lookAtPos) & 31;
quote = options.foldQuotes && IsGDTripleQuoteStringState(style);
}
const bool quote_start = (quote && !prevQuote);
const bool quote_continue = (quote && prevQuote);
if (!quote || !prevQuote)
indentCurrentLevel = indentCurrent & SC_FOLDLEVELNUMBERMASK;
if (quote)
indentNext = indentCurrentLevel;
if (indentNext & SC_FOLDLEVELWHITEFLAG)
indentNext = SC_FOLDLEVELWHITEFLAG | indentCurrentLevel;
if (quote_start) {
// Place fold point at start of triple quoted string
lev |= SC_FOLDLEVELHEADERFLAG;
} else if (quote_continue || prevQuote) {
// Add level to rest of lines in the string
lev = lev + 1;
}
// Skip past any blank lines for next indent level info; we skip also
// comments (all comments, not just those starting in column 0)
// which effectively folds them into surrounding code rather
// than screwing up folding. If comments end file, use the min
// comment indent as the level after
int minCommentLevel = indentCurrentLevel;
while (!quote &&
(lineNext < docLines) &&
((indentNext & SC_FOLDLEVELWHITEFLAG) || (IsCommentLine(lineNext, styler)))) {
if (IsCommentLine(lineNext, styler) && indentNext < minCommentLevel) {
minCommentLevel = indentNext;
}
lineNext++;
indentNext = styler.IndentAmount(lineNext, &spaceFlags, nullptr);
}
const int levelAfterComments = ((lineNext < docLines) ? indentNext & SC_FOLDLEVELNUMBERMASK : minCommentLevel);
const int levelBeforeComments = std::max(indentCurrentLevel, levelAfterComments);
// Now set all the indent levels on the lines we skipped
// Do this from end to start. Once we encounter one line
// which is indented more than the line after the end of
// the comment-block, use the level of the block before
Sci_Position skipLine = lineNext;
int skipLevel = levelAfterComments;
while (--skipLine > lineCurrent) {
const int skipLineIndent = styler.IndentAmount(skipLine, &spaceFlags, nullptr);
if (options.foldCompact) {
if ((skipLineIndent & SC_FOLDLEVELNUMBERMASK) > levelAfterComments)
skipLevel = levelBeforeComments;
const int whiteFlag = skipLineIndent & SC_FOLDLEVELWHITEFLAG;
styler.SetLevel(skipLine, skipLevel | whiteFlag);
} else {
if ((skipLineIndent & SC_FOLDLEVELNUMBERMASK) > levelAfterComments &&
!(skipLineIndent & SC_FOLDLEVELWHITEFLAG) &&
!IsCommentLine(skipLine, styler))
skipLevel = levelBeforeComments;
styler.SetLevel(skipLine, skipLevel);
}
}
// Set fold header on non-quote line
if (!quote && !(indentCurrent & SC_FOLDLEVELWHITEFLAG)) {
if ((indentCurrent & SC_FOLDLEVELNUMBERMASK) < (indentNext & SC_FOLDLEVELNUMBERMASK))
lev |= SC_FOLDLEVELHEADERFLAG;
}
// Keep track of triple quote state of previous line
prevQuote = quote;
// Set fold level for this line and move to next line
styler.SetLevel(lineCurrent, options.foldCompact ? lev : lev & ~SC_FOLDLEVELWHITEFLAG);
indentCurrent = indentNext;
lineCurrent = lineNext;
}
// NOTE: Cannot set level of last line here because indentCurrent doesn't have
// header flag set; the loop above is crafted to take care of this case!
//styler.SetLevel(lineCurrent, indentCurrent);
}
LexerModule lmGDScript(SCLEX_GDSCRIPT, LexerGDScript::LexerFactoryGDScript, "gdscript",
gdscriptWordListDesc);