notepad-plus-plus/lexilla/lexers/LexR.cxx

218 lines
6.2 KiB
C++

// Scintilla source code edit control
/** @file LexR.cxx
** Lexer for R, S, SPlus Statistics Program (Heavily derived from CPP Lexer).
**
**/
// Copyright 1998-2002 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 <string>
#include <string_view>
#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 Lexilla;
static inline bool IsAWordChar(const int ch) {
return (ch < 0x80) && (isalnum(ch) || ch == '.' || ch == '_');
}
static inline bool IsAWordStart(const int ch) {
return (ch < 0x80) && (isalnum(ch) || ch == '_');
}
static inline bool IsAnOperator(const int ch) {
if (IsASCII(ch) && isalnum(ch))
return false;
// '.' left out as it is used to make up numbers
if (ch == '-' || ch == '+' || ch == '!' || ch == '~' ||
ch == '?' || ch == ':' || ch == '*' || ch == '/' ||
ch == '^' || ch == '<' || ch == '>' || ch == '=' ||
ch == '&' || ch == '|' || ch == '$' || ch == '(' ||
ch == ')' || ch == '}' || ch == '{' || ch == '[' ||
ch == ']')
return true;
return false;
}
static void ColouriseRDoc(Sci_PositionU startPos, Sci_Position length, int initStyle, WordList *keywordlists[],
Accessor &styler) {
WordList &keywords = *keywordlists[0];
WordList &keywords2 = *keywordlists[1];
WordList &keywords3 = *keywordlists[2];
// Do not leak onto next line
if (initStyle == SCE_R_INFIXEOL)
initStyle = SCE_R_DEFAULT;
StyleContext sc(startPos, length, initStyle, styler);
for (; sc.More(); sc.Forward()) {
if (sc.atLineStart && (sc.state == SCE_R_STRING)) {
// Prevent SCE_R_STRINGEOL from leaking back to previous line
sc.SetState(SCE_R_STRING);
}
// Determine if the current state should terminate.
if (sc.state == SCE_R_OPERATOR) {
sc.SetState(SCE_R_DEFAULT);
} else if (sc.state == SCE_R_NUMBER) {
if (!IsADigit(sc.ch) && !(sc.ch == '.' && IsADigit(sc.chNext))) {
sc.SetState(SCE_R_DEFAULT);
}
} else if (sc.state == SCE_R_IDENTIFIER) {
if (!IsAWordChar(sc.ch)) {
char s[100];
sc.GetCurrent(s, sizeof(s));
if (keywords.InList(s)) {
sc.ChangeState(SCE_R_KWORD);
} else if (keywords2.InList(s)) {
sc.ChangeState(SCE_R_BASEKWORD);
} else if (keywords3.InList(s)) {
sc.ChangeState(SCE_R_OTHERKWORD);
}
sc.SetState(SCE_R_DEFAULT);
}
} else if (sc.state == SCE_R_COMMENT) {
if (sc.ch == '\r' || sc.ch == '\n') {
sc.SetState(SCE_R_DEFAULT);
}
} else if (sc.state == SCE_R_STRING) {
if (sc.ch == '\\') {
if (sc.chNext == '\"' || sc.chNext == '\'' || sc.chNext == '\\') {
sc.Forward();
}
} else if (sc.ch == '\"') {
sc.ForwardSetState(SCE_R_DEFAULT);
}
} else if (sc.state == SCE_R_INFIX) {
if (sc.ch == '%') {
sc.ForwardSetState(SCE_R_DEFAULT);
} else if (sc.atLineEnd) {
sc.ChangeState(SCE_R_INFIXEOL);
sc.ForwardSetState(SCE_R_DEFAULT);
}
}else if (sc.state == SCE_R_STRING2) {
if (sc.ch == '\\') {
if (sc.chNext == '\"' || sc.chNext == '\'' || sc.chNext == '\\') {
sc.Forward();
}
} else if (sc.ch == '\'') {
sc.ForwardSetState(SCE_R_DEFAULT);
}
}
// Determine if a new state should be entered.
if (sc.state == SCE_R_DEFAULT) {
if (IsADigit(sc.ch) || (sc.ch == '.' && IsADigit(sc.chNext))) {
sc.SetState(SCE_R_NUMBER);
} else if (IsAWordStart(sc.ch) ) {
sc.SetState(SCE_R_IDENTIFIER);
} else if (sc.Match('#')) {
sc.SetState(SCE_R_COMMENT);
} else if (sc.ch == '\"') {
sc.SetState(SCE_R_STRING);
} else if (sc.ch == '%') {
sc.SetState(SCE_R_INFIX);
} else if (sc.ch == '\'') {
sc.SetState(SCE_R_STRING2);
} else if (IsAnOperator(sc.ch)) {
sc.SetState(SCE_R_OPERATOR);
}
}
}
sc.Complete();
}
// Store both the current line's fold level and the next lines in the
// level store to make it easy to pick up with each increment
// and to make it possible to fiddle the current level for "} else {".
static void FoldRDoc(Sci_PositionU startPos, Sci_Position length, int, WordList *[],
Accessor &styler) {
bool foldCompact = styler.GetPropertyInt("fold.compact", 1) != 0;
bool foldAtElse = styler.GetPropertyInt("fold.at.else", 0) != 0;
Sci_PositionU endPos = startPos + length;
int visibleChars = 0;
Sci_Position lineCurrent = styler.GetLine(startPos);
int levelCurrent = SC_FOLDLEVELBASE;
if (lineCurrent > 0)
levelCurrent = styler.LevelAt(lineCurrent-1) >> 16;
int levelMinCurrent = levelCurrent;
int levelNext = levelCurrent;
char chNext = styler[startPos];
int styleNext = styler.StyleAt(startPos);
for (Sci_PositionU i = startPos; i < endPos; i++) {
char ch = chNext;
chNext = styler.SafeGetCharAt(i + 1);
int style = styleNext;
styleNext = styler.StyleAt(i + 1);
bool atEOL = (ch == '\r' && chNext != '\n') || (ch == '\n');
if (style == SCE_R_OPERATOR) {
if (ch == '{') {
// Measure the minimum before a '{' to allow
// folding on "} else {"
if (levelMinCurrent > levelNext) {
levelMinCurrent = levelNext;
}
levelNext++;
} else if (ch == '}') {
levelNext--;
}
}
if (atEOL) {
int levelUse = levelCurrent;
if (foldAtElse) {
levelUse = levelMinCurrent;
}
int lev = levelUse | levelNext << 16;
if (visibleChars == 0 && foldCompact)
lev |= SC_FOLDLEVELWHITEFLAG;
if (levelUse < levelNext)
lev |= SC_FOLDLEVELHEADERFLAG;
if (lev != styler.LevelAt(lineCurrent)) {
styler.SetLevel(lineCurrent, lev);
}
lineCurrent++;
levelCurrent = levelNext;
levelMinCurrent = levelCurrent;
visibleChars = 0;
}
if (!isspacechar(ch))
visibleChars++;
}
}
static const char * const RWordLists[] = {
"Language Keywords",
"Base / Default package function",
"Other Package Functions",
"Unused",
"Unused",
0,
};
LexerModule lmR(SCLEX_R, ColouriseRDoc, "r", FoldRDoc, RWordLists);