547 lines
20 KiB
C++
547 lines
20 KiB
C++
// Scintilla source code edit control
|
|
/** @file LexOScript.cxx
|
|
** Lexer for OScript sources; ocx files and/or OSpace dumps.
|
|
** OScript is a programming language used to develop applications for the
|
|
** Livelink server platform.
|
|
**/
|
|
// Written by Ferdinand Prantl <prantlf@gmail.com>, inspired by the code from
|
|
// LexVB.cxx and LexPascal.cxx. The License.txt file describes the conditions
|
|
// under which this software may be distributed.
|
|
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
#include <stdio.h>
|
|
#include <stdarg.h>
|
|
#include <assert.h>
|
|
#include <ctype.h>
|
|
|
|
#include "ILexer.h"
|
|
#include "Scintilla.h"
|
|
#include "SciLexer.h"
|
|
|
|
#include "WordList.h"
|
|
#include "LexAccessor.h"
|
|
#include "Accessor.h"
|
|
#include "StyleContext.h"
|
|
#include "CharacterSet.h"
|
|
#include "LexerModule.h"
|
|
|
|
using namespace Scintilla;
|
|
|
|
// -----------------------------------------
|
|
// Functions classifying a single character.
|
|
|
|
// This function is generic and should be probably moved to CharSet.h where
|
|
// IsAlphaNumeric the others reside.
|
|
inline bool IsAlpha(int ch) {
|
|
return (ch >= 'a' && ch <= 'z') || (ch >= 'A' && ch <= 'Z');
|
|
}
|
|
|
|
static inline bool IsIdentifierChar(int ch) {
|
|
// Identifiers cannot contain non-ASCII letters; a word with non-English
|
|
// language-specific characters cannot be an identifier.
|
|
return IsAlphaNumeric(ch) || ch == '_';
|
|
}
|
|
|
|
static inline bool IsIdentifierStart(int ch) {
|
|
// Identifiers cannot contain non-ASCII letters; a word with non-English
|
|
// language-specific characters cannot be an identifier.
|
|
return IsAlpha(ch) || ch == '_';
|
|
}
|
|
|
|
static inline bool IsNumberChar(int ch, int chNext) {
|
|
// Numeric constructs are not checked for lexical correctness. They are
|
|
// expected to look like +1.23-E9 but actually any bunch of the following
|
|
// characters will be styled as number.
|
|
// KNOWN PROBLEM: if you put + or - operators immediately after a number
|
|
// and the next operand starts with the letter E, the operator will not be
|
|
// recognized and it will be styled together with the preceding number.
|
|
// This should not occur; at least not often. The coding style recommends
|
|
// putting spaces around operators.
|
|
return IsADigit(ch) || toupper(ch) == 'E' || ch == '.' ||
|
|
((ch == '-' || ch == '+') && toupper(chNext) == 'E');
|
|
}
|
|
|
|
// This function checks for the start or a natural number without any symbols
|
|
// or operators as a prefix; the IsPrefixedNumberStart should be called
|
|
// immediately after this one to cover all possible numeric constructs.
|
|
static inline bool IsNaturalNumberStart(int ch) {
|
|
return IsADigit(ch) != 0;
|
|
}
|
|
|
|
static inline bool IsPrefixedNumberStart(int ch, int chNext) {
|
|
// KNOWN PROBLEM: if you put + or - operators immediately before a number
|
|
// the operator will not be recognized and it will be styled together with
|
|
// the succeeding number. This should not occur; at least not often. The
|
|
// coding style recommends putting spaces around operators.
|
|
return (ch == '.' || ch == '-' || ch == '+') && IsADigit(chNext);
|
|
}
|
|
|
|
static inline bool IsOperator(int ch) {
|
|
return strchr("%^&*()-+={}[]:;<>,/?!.~|\\", ch) != NULL;
|
|
}
|
|
|
|
// ---------------------------------------------------------------
|
|
// Functions classifying a token currently processed in the lexer.
|
|
|
|
// Checks if the current line starts with the preprocessor directive used
|
|
// usually to introduce documentation comments: #ifdef DOC. This method is
|
|
// supposed to be called if the line has been recognized as a preprocessor
|
|
// directive already.
|
|
static bool IsDocCommentStart(StyleContext &sc) {
|
|
// Check the line back to its start only if the end looks promising.
|
|
if (sc.LengthCurrent() == 10 && !IsAlphaNumeric(sc.ch)) {
|
|
char s[11];
|
|
sc.GetCurrentLowered(s, sizeof(s));
|
|
return strcmp(s, "#ifdef doc") == 0;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
// Checks if the current line starts with the preprocessor directive that
|
|
// is complementary to the #ifdef DOC start: #endif. This method is supposed
|
|
// to be called if the current state point to the documentation comment.
|
|
// QUESTIONAL ASSUMPTION: The complete #endif directive is not checked; just
|
|
// the starting #e. However, there is no other preprocessor directive with
|
|
// the same starting letter and thus this optimization should always work.
|
|
static bool IsDocCommentEnd(StyleContext &sc) {
|
|
return sc.ch == '#' && sc.chNext == 'e';
|
|
}
|
|
|
|
class IdentifierClassifier {
|
|
WordList &keywords; // Passed from keywords property.
|
|
WordList &constants; // Passed from keywords2 property.
|
|
WordList &operators; // Passed from keywords3 property.
|
|
WordList &types; // Passed from keywords4 property.
|
|
WordList &functions; // Passed from keywords5 property.
|
|
WordList &objects; // Passed from keywords6 property.
|
|
|
|
IdentifierClassifier(IdentifierClassifier const&);
|
|
IdentifierClassifier& operator=(IdentifierClassifier const&);
|
|
|
|
public:
|
|
IdentifierClassifier(WordList *keywordlists[]) :
|
|
keywords(*keywordlists[0]), constants(*keywordlists[1]),
|
|
operators(*keywordlists[2]), types(*keywordlists[3]),
|
|
functions(*keywordlists[4]), objects(*keywordlists[5])
|
|
{}
|
|
|
|
void ClassifyIdentifier(StyleContext &sc) {
|
|
// Opening parenthesis following an identifier makes it a possible
|
|
// function call.
|
|
// KNOWN PROBLEM: If some whitespace is inserted between the
|
|
// identifier and the parenthesis they will not be able to be
|
|
// recognized as a function call. This should not occur; at
|
|
// least not often. Such coding style would be weird.
|
|
if (sc.Match('(')) {
|
|
char s[100];
|
|
sc.GetCurrentLowered(s, sizeof(s));
|
|
// Before an opening brace can be control statements and
|
|
// operators too; function call is the last option.
|
|
if (keywords.InList(s)) {
|
|
sc.ChangeState(SCE_OSCRIPT_KEYWORD);
|
|
} else if (operators.InList(s)) {
|
|
sc.ChangeState(SCE_OSCRIPT_OPERATOR);
|
|
} else if (functions.InList(s)) {
|
|
sc.ChangeState(SCE_OSCRIPT_FUNCTION);
|
|
} else {
|
|
sc.ChangeState(SCE_OSCRIPT_METHOD);
|
|
}
|
|
sc.SetState(SCE_OSCRIPT_OPERATOR);
|
|
} else {
|
|
char s[100];
|
|
sc.GetCurrentLowered(s, sizeof(s));
|
|
// A dot following an identifier means an access to an object
|
|
// member. The related object identifier can be special.
|
|
// KNOWN PROBLEM: If there is whitespace between the identifier
|
|
// and the following dot, the identifier will not be recognized
|
|
// as an object in an object member access. If it is one of the
|
|
// listed static objects it will not be styled.
|
|
if (sc.Match('.') && objects.InList(s)) {
|
|
sc.ChangeState(SCE_OSCRIPT_OBJECT);
|
|
sc.SetState(SCE_OSCRIPT_OPERATOR);
|
|
} else {
|
|
if (keywords.InList(s)) {
|
|
sc.ChangeState(SCE_OSCRIPT_KEYWORD);
|
|
} else if (constants.InList(s)) {
|
|
sc.ChangeState(SCE_OSCRIPT_CONSTANT);
|
|
} else if (operators.InList(s)) {
|
|
sc.ChangeState(SCE_OSCRIPT_OPERATOR);
|
|
} else if (types.InList(s)) {
|
|
sc.ChangeState(SCE_OSCRIPT_TYPE);
|
|
} else if (functions.InList(s)) {
|
|
sc.ChangeState(SCE_OSCRIPT_FUNCTION);
|
|
}
|
|
sc.SetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
}
|
|
}
|
|
};
|
|
|
|
// ------------------------------------------------
|
|
// Function colourising an excerpt of OScript code.
|
|
|
|
static void ColouriseOScriptDoc(Sci_PositionU startPos, Sci_Position length,
|
|
int initStyle, WordList *keywordlists[],
|
|
Accessor &styler) {
|
|
// I wonder how whole-line styles ended by EOLN can escape the resetting
|
|
// code in the loop below and overflow to the next line. Let us make sure
|
|
// that a new line does not start with them carried from the previous one.
|
|
// NOTE: An overflowing string is intentionally not checked; it reminds
|
|
// the developer that the string must be ended on the same line.
|
|
if (initStyle == SCE_OSCRIPT_LINE_COMMENT ||
|
|
initStyle == SCE_OSCRIPT_PREPROCESSOR) {
|
|
initStyle = SCE_OSCRIPT_DEFAULT;
|
|
}
|
|
|
|
styler.StartAt(startPos);
|
|
StyleContext sc(startPos, length, initStyle, styler);
|
|
IdentifierClassifier identifierClassifier(keywordlists);
|
|
|
|
// It starts with true at the beginning of a line and changes to false as
|
|
// soon as the first non-whitespace character has been processed.
|
|
bool isFirstToken = true;
|
|
// It starts with true at the beginning of a line and changes to false as
|
|
// soon as the first identifier on the line is passed by.
|
|
bool isFirstIdentifier = true;
|
|
// It becomes false when #ifdef DOC (the preprocessor directive often
|
|
// used to start a documentation comment) is encountered and remain false
|
|
// until the end of the documentation block is not detected. This is done
|
|
// by checking for the complementary #endif preprocessor directive.
|
|
bool endDocComment = false;
|
|
|
|
for (; sc.More(); sc.Forward()) {
|
|
|
|
if (sc.atLineStart) {
|
|
isFirstToken = true;
|
|
isFirstIdentifier = true;
|
|
// Detect the current state is neither whitespace nor identifier. It
|
|
// means that no next identifier can be the first token on the line.
|
|
} else if (isFirstIdentifier && sc.state != SCE_OSCRIPT_DEFAULT &&
|
|
sc.state != SCE_OSCRIPT_IDENTIFIER) {
|
|
isFirstIdentifier = false;
|
|
}
|
|
|
|
// Check if the current state should be changed.
|
|
if (sc.state == SCE_OSCRIPT_OPERATOR) {
|
|
// Multiple-symbol operators are marked by single characters.
|
|
sc.SetState(SCE_OSCRIPT_DEFAULT);
|
|
} else if (sc.state == SCE_OSCRIPT_IDENTIFIER) {
|
|
if (!IsIdentifierChar(sc.ch)) {
|
|
// Colon after an identifier makes it a label if it is the
|
|
// first token on the line.
|
|
// KNOWN PROBLEM: If some whitespace is inserted between the
|
|
// identifier and the colon they will not be recognized as a
|
|
// label. This should not occur; at least not often. It would
|
|
// make the code structure less legible and examples in the
|
|
// Livelink documentation do not show it.
|
|
if (sc.Match(':') && isFirstIdentifier) {
|
|
sc.ChangeState(SCE_OSCRIPT_LABEL);
|
|
sc.ForwardSetState(SCE_OSCRIPT_DEFAULT);
|
|
} else {
|
|
identifierClassifier.ClassifyIdentifier(sc);
|
|
}
|
|
// Avoid a sequence of two words be mistaken for a label. A
|
|
// switch case would be an example.
|
|
isFirstIdentifier = false;
|
|
}
|
|
} else if (sc.state == SCE_OSCRIPT_GLOBAL) {
|
|
if (!IsIdentifierChar(sc.ch)) {
|
|
sc.SetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
} else if (sc.state == SCE_OSCRIPT_PROPERTY) {
|
|
if (!IsIdentifierChar(sc.ch)) {
|
|
// Any member access introduced by the dot operator is
|
|
// initially marked as a property access. If an opening
|
|
// parenthesis is detected later it is changed to method call.
|
|
// KNOWN PROBLEM: The same as at the function call recognition
|
|
// for SCE_OSCRIPT_IDENTIFIER above.
|
|
if (sc.Match('(')) {
|
|
sc.ChangeState(SCE_OSCRIPT_METHOD);
|
|
}
|
|
sc.SetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
} else if (sc.state == SCE_OSCRIPT_NUMBER) {
|
|
if (!IsNumberChar(sc.ch, sc.chNext)) {
|
|
sc.SetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
} else if (sc.state == SCE_OSCRIPT_SINGLEQUOTE_STRING) {
|
|
if (sc.ch == '\'') {
|
|
// Two consequential apostrophes convert to a single one.
|
|
if (sc.chNext == '\'') {
|
|
sc.Forward();
|
|
} else {
|
|
sc.ForwardSetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
} else if (sc.atLineEnd) {
|
|
sc.ForwardSetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
} else if (sc.state == SCE_OSCRIPT_DOUBLEQUOTE_STRING) {
|
|
if (sc.ch == '\"') {
|
|
// Two consequential quotation marks convert to a single one.
|
|
if (sc.chNext == '\"') {
|
|
sc.Forward();
|
|
} else {
|
|
sc.ForwardSetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
} else if (sc.atLineEnd) {
|
|
sc.ForwardSetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
} else if (sc.state == SCE_OSCRIPT_BLOCK_COMMENT) {
|
|
if (sc.Match('*', '/')) {
|
|
sc.Forward();
|
|
sc.ForwardSetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
} else if (sc.state == SCE_OSCRIPT_LINE_COMMENT) {
|
|
if (sc.atLineEnd) {
|
|
sc.ForwardSetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
} else if (sc.state == SCE_OSCRIPT_PREPROCESSOR) {
|
|
if (IsDocCommentStart(sc)) {
|
|
sc.ChangeState(SCE_OSCRIPT_DOC_COMMENT);
|
|
endDocComment = false;
|
|
} else if (sc.atLineEnd) {
|
|
sc.ForwardSetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
} else if (sc.state == SCE_OSCRIPT_DOC_COMMENT) {
|
|
// KNOWN PROBLEM: The first line detected that would close a
|
|
// conditional preprocessor block (#endif) the documentation
|
|
// comment block will end. (Nested #if-#endif blocks are not
|
|
// supported. Hopefully it will not occur often that a line
|
|
// within the text block would stat with #endif.
|
|
if (isFirstToken && IsDocCommentEnd(sc)) {
|
|
endDocComment = true;
|
|
} else if (sc.atLineEnd && endDocComment) {
|
|
sc.ForwardSetState(SCE_OSCRIPT_DEFAULT);
|
|
}
|
|
}
|
|
|
|
// Check what state starts with the current character.
|
|
if (sc.state == SCE_OSCRIPT_DEFAULT) {
|
|
if (sc.Match('\'')) {
|
|
sc.SetState(SCE_OSCRIPT_SINGLEQUOTE_STRING);
|
|
} else if (sc.Match('\"')) {
|
|
sc.SetState(SCE_OSCRIPT_DOUBLEQUOTE_STRING);
|
|
} else if (sc.Match('/', '/')) {
|
|
sc.SetState(SCE_OSCRIPT_LINE_COMMENT);
|
|
sc.Forward();
|
|
} else if (sc.Match('/', '*')) {
|
|
sc.SetState(SCE_OSCRIPT_BLOCK_COMMENT);
|
|
sc.Forward();
|
|
} else if (isFirstToken && sc.Match('#')) {
|
|
sc.SetState(SCE_OSCRIPT_PREPROCESSOR);
|
|
} else if (sc.Match('$')) {
|
|
// Both process-global ($xxx) and thread-global ($$xxx)
|
|
// variables are handled as one global.
|
|
sc.SetState(SCE_OSCRIPT_GLOBAL);
|
|
} else if (IsNaturalNumberStart(sc.ch)) {
|
|
sc.SetState(SCE_OSCRIPT_NUMBER);
|
|
} else if (IsPrefixedNumberStart(sc.ch, sc.chNext)) {
|
|
sc.SetState(SCE_OSCRIPT_NUMBER);
|
|
sc.Forward();
|
|
} else if (sc.Match('.') && IsIdentifierStart(sc.chNext)) {
|
|
// Every object member access is marked as a property access
|
|
// initially. The decision between property and method is made
|
|
// after parsing the identifier and looking what comes then.
|
|
// KNOWN PROBLEM: If there is whitespace between the following
|
|
// identifier and the dot, the dot will not be recognized
|
|
// as a member accessing operator. In turn, the identifier
|
|
// will not be recognizable as a property or a method too.
|
|
sc.SetState(SCE_OSCRIPT_OPERATOR);
|
|
sc.Forward();
|
|
sc.SetState(SCE_OSCRIPT_PROPERTY);
|
|
} else if (IsIdentifierStart(sc.ch)) {
|
|
sc.SetState(SCE_OSCRIPT_IDENTIFIER);
|
|
} else if (IsOperator(sc.ch)) {
|
|
sc.SetState(SCE_OSCRIPT_OPERATOR);
|
|
}
|
|
}
|
|
|
|
if (isFirstToken && !IsASpaceOrTab(sc.ch)) {
|
|
isFirstToken = false;
|
|
}
|
|
}
|
|
|
|
sc.Complete();
|
|
}
|
|
|
|
// ------------------------------------------
|
|
// Functions supporting OScript code folding.
|
|
|
|
static inline bool IsBlockComment(int style) {
|
|
return style == SCE_OSCRIPT_BLOCK_COMMENT;
|
|
}
|
|
|
|
static bool IsLineComment(Sci_Position line, Accessor &styler) {
|
|
Sci_Position pos = styler.LineStart(line);
|
|
Sci_Position eolPos = styler.LineStart(line + 1) - 1;
|
|
for (Sci_Position i = pos; i < eolPos; i++) {
|
|
char ch = styler[i];
|
|
char chNext = styler.SafeGetCharAt(i + 1);
|
|
int style = styler.StyleAt(i);
|
|
if (ch == '/' && chNext == '/' && style == SCE_OSCRIPT_LINE_COMMENT) {
|
|
return true;
|
|
} else if (!IsASpaceOrTab(ch)) {
|
|
return false;
|
|
}
|
|
}
|
|
return false;
|
|
}
|
|
|
|
static inline bool IsPreprocessor(int style) {
|
|
return style == SCE_OSCRIPT_PREPROCESSOR ||
|
|
style == SCE_OSCRIPT_DOC_COMMENT;
|
|
}
|
|
|
|
static void GetRangeLowered(Sci_PositionU start, Sci_PositionU end,
|
|
Accessor &styler, char *s, Sci_PositionU len) {
|
|
Sci_PositionU i = 0;
|
|
while (i < end - start + 1 && i < len - 1) {
|
|
s[i] = static_cast<char>(tolower(styler[start + i]));
|
|
i++;
|
|
}
|
|
s[i] = '\0';
|
|
}
|
|
|
|
static void GetForwardWordLowered(Sci_PositionU start, Accessor &styler,
|
|
char *s, Sci_PositionU len) {
|
|
Sci_PositionU i = 0;
|
|
while (i < len - 1 && IsAlpha(styler.SafeGetCharAt(start + i))) {
|
|
s[i] = static_cast<char>(tolower(styler.SafeGetCharAt(start + i)));
|
|
i++;
|
|
}
|
|
s[i] = '\0';
|
|
}
|
|
|
|
static void UpdatePreprocessorFoldLevel(int &levelCurrent,
|
|
Sci_PositionU startPos, Accessor &styler) {
|
|
char s[7]; // Size of the longest possible keyword + null.
|
|
GetForwardWordLowered(startPos, styler, s, sizeof(s));
|
|
|
|
if (strcmp(s, "ifdef") == 0 ||
|
|
strcmp(s, "ifndef") == 0) {
|
|
levelCurrent++;
|
|
} else if (strcmp(s, "endif") == 0) {
|
|
levelCurrent--;
|
|
if (levelCurrent < SC_FOLDLEVELBASE) {
|
|
levelCurrent = SC_FOLDLEVELBASE;
|
|
}
|
|
}
|
|
}
|
|
|
|
static void UpdateKeywordFoldLevel(int &levelCurrent, Sci_PositionU lastStart,
|
|
Sci_PositionU currentPos, Accessor &styler) {
|
|
char s[9];
|
|
GetRangeLowered(lastStart, currentPos, styler, s, sizeof(s));
|
|
|
|
if (strcmp(s, "if") == 0 || strcmp(s, "for") == 0 ||
|
|
strcmp(s, "switch") == 0 || strcmp(s, "function") == 0 ||
|
|
strcmp(s, "while") == 0 || strcmp(s, "repeat") == 0) {
|
|
levelCurrent++;
|
|
} else if (strcmp(s, "end") == 0 || strcmp(s, "until") == 0) {
|
|
levelCurrent--;
|
|
if (levelCurrent < SC_FOLDLEVELBASE) {
|
|
levelCurrent = SC_FOLDLEVELBASE;
|
|
}
|
|
}
|
|
}
|
|
|
|
// ------------------------------
|
|
// Function folding OScript code.
|
|
|
|
static void FoldOScriptDoc(Sci_PositionU startPos, Sci_Position length, int initStyle,
|
|
WordList *[], Accessor &styler) {
|
|
bool foldComment = styler.GetPropertyInt("fold.comment") != 0;
|
|
bool foldPreprocessor = styler.GetPropertyInt("fold.preprocessor") != 0;
|
|
bool foldCompact = styler.GetPropertyInt("fold.compact", 1) != 0;
|
|
Sci_Position endPos = startPos + length;
|
|
int visibleChars = 0;
|
|
Sci_Position lineCurrent = styler.GetLine(startPos);
|
|
int levelPrev = styler.LevelAt(lineCurrent) & SC_FOLDLEVELNUMBERMASK;
|
|
int levelCurrent = levelPrev;
|
|
char chNext = styler[startPos];
|
|
int styleNext = styler.StyleAt(startPos);
|
|
int style = initStyle;
|
|
Sci_Position lastStart = 0;
|
|
|
|
for (Sci_Position i = startPos; i < endPos; i++) {
|
|
char ch = chNext;
|
|
chNext = styler.SafeGetCharAt(i + 1);
|
|
int stylePrev = style;
|
|
style = styleNext;
|
|
styleNext = styler.StyleAt(i + 1);
|
|
bool atLineEnd = (ch == '\r' && chNext != '\n') || (ch == '\n');
|
|
|
|
if (foldComment && IsBlockComment(style)) {
|
|
if (!IsBlockComment(stylePrev)) {
|
|
levelCurrent++;
|
|
} else if (!IsBlockComment(styleNext) && !atLineEnd) {
|
|
// Comments do not end at end of line and the next character
|
|
// may not be styled.
|
|
levelCurrent--;
|
|
}
|
|
}
|
|
if (foldComment && atLineEnd && IsLineComment(lineCurrent, styler)) {
|
|
if (!IsLineComment(lineCurrent - 1, styler) &&
|
|
IsLineComment(lineCurrent + 1, styler))
|
|
levelCurrent++;
|
|
else if (IsLineComment(lineCurrent - 1, styler) &&
|
|
!IsLineComment(lineCurrent+1, styler))
|
|
levelCurrent--;
|
|
}
|
|
if (foldPreprocessor) {
|
|
if (ch == '#' && IsPreprocessor(style)) {
|
|
UpdatePreprocessorFoldLevel(levelCurrent, i + 1, styler);
|
|
}
|
|
}
|
|
|
|
if (stylePrev != SCE_OSCRIPT_KEYWORD && style == SCE_OSCRIPT_KEYWORD) {
|
|
lastStart = i;
|
|
}
|
|
if (stylePrev == SCE_OSCRIPT_KEYWORD) {
|
|
if(IsIdentifierChar(ch) && !IsIdentifierChar(chNext)) {
|
|
UpdateKeywordFoldLevel(levelCurrent, lastStart, i, styler);
|
|
}
|
|
}
|
|
|
|
if (!IsASpace(ch))
|
|
visibleChars++;
|
|
|
|
if (atLineEnd) {
|
|
int level = levelPrev;
|
|
if (visibleChars == 0 && foldCompact)
|
|
level |= SC_FOLDLEVELWHITEFLAG;
|
|
if ((levelCurrent > levelPrev) && (visibleChars > 0))
|
|
level |= SC_FOLDLEVELHEADERFLAG;
|
|
if (level != styler.LevelAt(lineCurrent)) {
|
|
styler.SetLevel(lineCurrent, level);
|
|
}
|
|
lineCurrent++;
|
|
levelPrev = levelCurrent;
|
|
visibleChars = 0;
|
|
}
|
|
}
|
|
|
|
// If we did not reach EOLN in the previous loop, store the line level and
|
|
// whitespace information. The rest will be filled in later.
|
|
int lev = levelPrev;
|
|
if (visibleChars == 0 && foldCompact)
|
|
lev |= SC_FOLDLEVELWHITEFLAG;
|
|
styler.SetLevel(lineCurrent, lev);
|
|
}
|
|
|
|
// --------------------------------------------
|
|
// Declaration of the OScript lexer descriptor.
|
|
|
|
static const char * const oscriptWordListDesc[] = {
|
|
"Keywords and reserved words",
|
|
"Literal constants",
|
|
"Literal operators",
|
|
"Built-in value and reference types",
|
|
"Built-in global functions",
|
|
"Built-in static objects",
|
|
0
|
|
};
|
|
|
|
LexerModule lmOScript(SCLEX_OSCRIPT, ColouriseOScriptDoc, "oscript", FoldOScriptDoc, oscriptWordListDesc);
|