iup-stack/iup/srcscintilla/scintilla375/lexers/LexVB.cxx

320 lines
10 KiB
C++
Executable File

// Scintilla source code edit control
/** @file LexVB.cxx
** Lexer for Visual Basic and VBScript.
**/
// Copyright 1998-2005 by Neil Hodgson <neilh@scintilla.org>
// 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"
#ifdef SCI_NAMESPACE
using namespace Scintilla;
#endif
// Internal state, highlighted as number
#define SCE_B_FILENUMBER SCE_B_DEFAULT+100
static bool IsVBComment(Accessor &styler, Sci_Position pos, Sci_Position len) {
return len > 0 && styler[pos] == '\'';
}
static inline bool IsTypeCharacter(int ch) {
return ch == '%' || ch == '&' || ch == '@' || ch == '!' || ch == '#' || ch == '$';
}
// Extended to accept accented characters
static inline bool IsAWordChar(int ch) {
return ch >= 0x80 ||
(isalnum(ch) || ch == '.' || ch == '_');
}
static inline bool IsAWordStart(int ch) {
return ch >= 0x80 ||
(isalpha(ch) || ch == '_');
}
static inline bool IsANumberChar(int ch) {
// Not exactly following number definition (several dots are seen as OK, etc.)
// but probably enough in most cases.
return (ch < 0x80) &&
(isdigit(ch) || toupper(ch) == 'E' ||
ch == '.' || ch == '-' || ch == '+');
}
static void ColouriseVBDoc(Sci_PositionU startPos, Sci_Position length, int initStyle,
WordList *keywordlists[], Accessor &styler, bool vbScriptSyntax) {
WordList &keywords = *keywordlists[0];
WordList &keywords2 = *keywordlists[1];
WordList &keywords3 = *keywordlists[2];
WordList &keywords4 = *keywordlists[3];
styler.StartAt(startPos);
int visibleChars = 0;
int fileNbDigits = 0;
// Do not leak onto next line
if (initStyle == SCE_B_STRINGEOL || initStyle == SCE_B_COMMENT || initStyle == SCE_B_PREPROCESSOR) {
initStyle = SCE_B_DEFAULT;
}
StyleContext sc(startPos, length, initStyle, styler);
for (; sc.More(); sc.Forward()) {
if (sc.state == SCE_B_OPERATOR) {
sc.SetState(SCE_B_DEFAULT);
} else if (sc.state == SCE_B_IDENTIFIER) {
if (!IsAWordChar(sc.ch)) {
// In Basic (except VBScript), a variable name or a function name
// can end with a special character indicating the type of the value
// held or returned.
bool skipType = false;
if (!vbScriptSyntax && IsTypeCharacter(sc.ch)) {
sc.Forward(); // Skip it
skipType = true;
}
if (sc.ch == ']') {
sc.Forward();
}
char s[100];
sc.GetCurrentLowered(s, sizeof(s));
if (skipType) {
s[strlen(s) - 1] = '\0';
}
if (strcmp(s, "rem") == 0) {
sc.ChangeState(SCE_B_COMMENT);
} else {
if (keywords.InList(s)) {
sc.ChangeState(SCE_B_KEYWORD);
} else if (keywords2.InList(s)) {
sc.ChangeState(SCE_B_KEYWORD2);
} else if (keywords3.InList(s)) {
sc.ChangeState(SCE_B_KEYWORD3);
} else if (keywords4.InList(s)) {
sc.ChangeState(SCE_B_KEYWORD4);
} // Else, it is really an identifier...
sc.SetState(SCE_B_DEFAULT);
}
}
} else if (sc.state == SCE_B_NUMBER) {
// We stop the number definition on non-numerical non-dot non-eE non-sign char
// Also accepts A-F for hex. numbers
if (!IsANumberChar(sc.ch) && !(tolower(sc.ch) >= 'a' && tolower(sc.ch) <= 'f')) {
sc.SetState(SCE_B_DEFAULT);
}
} else if (sc.state == SCE_B_STRING) {
// VB doubles quotes to preserve them, so just end this string
// state now as a following quote will start again
if (sc.ch == '\"') {
if (sc.chNext == '\"') {
sc.Forward();
} else {
if (tolower(sc.chNext) == 'c') {
sc.Forward();
}
sc.ForwardSetState(SCE_B_DEFAULT);
}
} else if (sc.atLineEnd) {
visibleChars = 0;
sc.ChangeState(SCE_B_STRINGEOL);
sc.ForwardSetState(SCE_B_DEFAULT);
}
} else if (sc.state == SCE_B_COMMENT) {
if (sc.atLineEnd) {
visibleChars = 0;
sc.ForwardSetState(SCE_B_DEFAULT);
}
} else if (sc.state == SCE_B_PREPROCESSOR) {
if (sc.atLineEnd) {
visibleChars = 0;
sc.ForwardSetState(SCE_B_DEFAULT);
}
} else if (sc.state == SCE_B_FILENUMBER) {
if (IsADigit(sc.ch)) {
fileNbDigits++;
if (fileNbDigits > 3) {
sc.ChangeState(SCE_B_DATE);
}
} else if (sc.ch == '\r' || sc.ch == '\n' || sc.ch == ',') {
// Regular uses: Close #1; Put #1, ...; Get #1, ... etc.
// Too bad if date is format #27, Oct, 2003# or something like that...
// Use regular number state
sc.ChangeState(SCE_B_NUMBER);
sc.SetState(SCE_B_DEFAULT);
} else if (sc.ch == '#') {
sc.ChangeState(SCE_B_DATE);
sc.ForwardSetState(SCE_B_DEFAULT);
} else {
sc.ChangeState(SCE_B_DATE);
}
if (sc.state != SCE_B_FILENUMBER) {
fileNbDigits = 0;
}
} else if (sc.state == SCE_B_DATE) {
if (sc.atLineEnd) {
visibleChars = 0;
sc.ChangeState(SCE_B_STRINGEOL);
sc.ForwardSetState(SCE_B_DEFAULT);
} else if (sc.ch == '#') {
sc.ForwardSetState(SCE_B_DEFAULT);
}
}
if (sc.state == SCE_B_DEFAULT) {
if (sc.ch == '\'') {
sc.SetState(SCE_B_COMMENT);
} else if (sc.ch == '\"') {
sc.SetState(SCE_B_STRING);
} else if (sc.ch == '#' && visibleChars == 0) {
// Preprocessor commands are alone on their line
sc.SetState(SCE_B_PREPROCESSOR);
} else if (sc.ch == '#') {
// It can be a date literal, ending with #, or a file number, from 1 to 511
// The date literal depends on the locale, so anything can go between #'s.
// Can be #January 1, 1993# or #1 Jan 93# or #05/11/2003#, etc.
// So we set the FILENUMBER state, and switch to DATE if it isn't a file number
sc.SetState(SCE_B_FILENUMBER);
} else if (sc.ch == '&' && tolower(sc.chNext) == 'h') {
// Hexadecimal number
sc.SetState(SCE_B_NUMBER);
sc.Forward();
} else if (sc.ch == '&' && tolower(sc.chNext) == 'o') {
// Octal number
sc.SetState(SCE_B_NUMBER);
sc.Forward();
} else if (IsADigit(sc.ch) || (sc.ch == '.' && IsADigit(sc.chNext))) {
sc.SetState(SCE_B_NUMBER);
} else if (IsAWordStart(sc.ch) || (sc.ch == '[')) {
sc.SetState(SCE_B_IDENTIFIER);
} else if (isoperator(static_cast<char>(sc.ch)) || (sc.ch == '\\')) { // Integer division
sc.SetState(SCE_B_OPERATOR);
}
}
if (sc.atLineEnd) {
visibleChars = 0;
}
if (!IsASpace(sc.ch)) {
visibleChars++;
}
}
if (sc.state == SCE_B_IDENTIFIER && !IsAWordChar(sc.ch)) {
// In Basic (except VBScript), a variable name or a function name
// can end with a special character indicating the type of the value
// held or returned.
bool skipType = false;
if (!vbScriptSyntax && IsTypeCharacter(sc.ch)) {
sc.Forward(); // Skip it
skipType = true;
}
if (sc.ch == ']') {
sc.Forward();
}
char s[100];
sc.GetCurrentLowered(s, sizeof(s));
if (skipType) {
s[strlen(s) - 1] = '\0';
}
if (strcmp(s, "rem") == 0) {
sc.ChangeState(SCE_B_COMMENT);
} else {
if (keywords.InList(s)) {
sc.ChangeState(SCE_B_KEYWORD);
} else if (keywords2.InList(s)) {
sc.ChangeState(SCE_B_KEYWORD2);
} else if (keywords3.InList(s)) {
sc.ChangeState(SCE_B_KEYWORD3);
} else if (keywords4.InList(s)) {
sc.ChangeState(SCE_B_KEYWORD4);
} // Else, it is really an identifier...
sc.SetState(SCE_B_DEFAULT);
}
}
sc.Complete();
}
static void FoldVBDoc(Sci_PositionU startPos, Sci_Position length, int,
WordList *[], Accessor &styler) {
Sci_Position endPos = startPos + length;
// Backtrack to previous line in case need to fix its fold status
Sci_Position lineCurrent = styler.GetLine(startPos);
if (startPos > 0) {
if (lineCurrent > 0) {
lineCurrent--;
startPos = styler.LineStart(lineCurrent);
}
}
int spaceFlags = 0;
int indentCurrent = styler.IndentAmount(lineCurrent, &spaceFlags, IsVBComment);
char chNext = styler[startPos];
for (Sci_Position i = startPos; i < endPos; i++) {
char ch = chNext;
chNext = styler.SafeGetCharAt(i + 1);
if ((ch == '\r' && chNext != '\n') || (ch == '\n') || (i == endPos)) {
int lev = indentCurrent;
int indentNext = styler.IndentAmount(lineCurrent + 1, &spaceFlags, IsVBComment);
if (!(indentCurrent & SC_FOLDLEVELWHITEFLAG)) {
// Only non whitespace lines can be headers
if ((indentCurrent & SC_FOLDLEVELNUMBERMASK) < (indentNext & SC_FOLDLEVELNUMBERMASK)) {
lev |= SC_FOLDLEVELHEADERFLAG;
} else if (indentNext & SC_FOLDLEVELWHITEFLAG) {
// Line after is blank so check the next - maybe should continue further?
int spaceFlags2 = 0;
int indentNext2 = styler.IndentAmount(lineCurrent + 2, &spaceFlags2, IsVBComment);
if ((indentCurrent & SC_FOLDLEVELNUMBERMASK) < (indentNext2 & SC_FOLDLEVELNUMBERMASK)) {
lev |= SC_FOLDLEVELHEADERFLAG;
}
}
}
indentCurrent = indentNext;
styler.SetLevel(lineCurrent, lev);
lineCurrent++;
}
}
}
static void ColouriseVBNetDoc(Sci_PositionU startPos, Sci_Position length, int initStyle,
WordList *keywordlists[], Accessor &styler) {
ColouriseVBDoc(startPos, length, initStyle, keywordlists, styler, false);
}
static void ColouriseVBScriptDoc(Sci_PositionU startPos, Sci_Position length, int initStyle,
WordList *keywordlists[], Accessor &styler) {
ColouriseVBDoc(startPos, length, initStyle, keywordlists, styler, true);
}
static const char * const vbWordListDesc[] = {
"Keywords",
"user1",
"user2",
"user3",
0
};
LexerModule lmVB(SCLEX_VB, ColouriseVBNetDoc, "vb", FoldVBDoc, vbWordListDesc);
LexerModule lmVBScript(SCLEX_VBSCRIPT, ColouriseVBScriptDoc, "vbscript", FoldVBDoc, vbWordListDesc);