diff options
Diffstat (limited to 'src/LexAU3.cpp')
-rwxr-xr-x | src/LexAU3.cpp | 891 |
1 files changed, 891 insertions, 0 deletions
diff --git a/src/LexAU3.cpp b/src/LexAU3.cpp new file mode 100755 index 0000000..2bc2e0f --- /dev/null +++ b/src/LexAU3.cpp @@ -0,0 +1,891 @@ +// Scintilla source code edit control +// @file LexAU3.cxx +// Lexer for AutoIt3 http://www.hiddensoft.com/autoit3 +// by Jos van der Zande, jvdzande@yahoo.com +// +// Changes: +// March 28, 2004 - Added the standard Folding code +// April 21, 2004 - Added Preprosessor Table + Syntax Highlighting +// Fixed Number highlighting +// Changed default isoperator to IsAOperator to have a better match to AutoIt3 +// Fixed "#comments_start" -> "#comments-start" +// Fixed "#comments_end" -> "#comments-end" +// Fixed Sendkeys in Strings when not terminated with } +// Added support for Sendkey strings that have second parameter e.g. {UP 5} or {a down} +// April 26, 2004 - Fixed # pre-processor statement inside of comment block would invalidly change the color. +// Added logic for #include <xyz.au3> to treat the <> as string +// Added underscore to IsAOperator. +// May 17, 2004 - Changed the folding logic from indent to keyword folding. +// Added Folding logic for blocks of single-commentlines or commentblock. +// triggered by: fold.comment=1 +// Added Folding logic for preprocessor blocks triggered by fold.preprocessor=1 +// Added Special for #region - #endregion syntax highlight and folding. +// May 30, 2004 - Fixed issue with continuation lines on If statements. +// June 5, 2004 - Added comma to Operators for better readability. +// Added fold.compact support set with fold.compact=1 +// Changed folding inside of #cs-#ce. Default is no keyword folding inside comment blocks when fold.comment=1 +// it will now only happen when fold.comment=2. +// Sep 5, 2004 - Added logic to handle colourizing words on the last line. +// Typed Characters now show as "default" till they match any table. +// Oct 10, 2004 - Added logic to show Comments in "Special" directives. +// Nov 1, 2004 - Added better testing for Numbers supporting x and e notation. +// Nov 28, 2004 - Added logic to handle continuation lines for syntax highlighting. +// Jan 10, 2005 - Added Abbreviations Keyword used for expansion +// Mar 24, 2005 - Updated Abbreviations Keywords to fix when followed by Operator. +// Apr 18, 2005 - Updated #CE/#Comment-End logic to take a linecomment ";" into account +// - Added folding support for With...EndWith +// - Added support for a DOT in variable names +// - Fixed Underscore in CommentBlock +// May 23, 2005 - Fixed the SentKey lexing in case of a missing } +// Aug 11, 2005 - Fixed possible bug with s_save length > 100. +// Aug 23, 2005 - Added Switch/endswitch support to the folding logic. +// Sep 27, 2005 - Fixed the SentKey lexing logic in case of multiple sentkeys. +// Mar 12, 2006 - Fixed issue with <> coloring as String in stead of Operator in rare occasions. +// Apr 8, 2006 - Added support for AutoIt3 Standard UDF library (SCE_AU3_UDF) +// +// Copyright for Scintilla: 1998-2001 by Neil Hodgson <neilh@scintilla.org> +// The License.txt file describes the conditions under which this software may be distributed. +// Scintilla source code edit control + +#include <stdlib.h> +#include <string.h> +#include <ctype.h> +#include <stdio.h> +#include <stdarg.h> + +#include "Platform.h" + +#include "PropSet.h" +#include "Accessor.h" +#include "StyleContext.h" +#include "KeyWords.h" +#include "Scintilla.h" +#include "SciLexer.h" + +static inline bool IsTypeCharacter(const int ch) +{ + return ch == '$'; +} +static inline bool IsAWordChar(const int ch) +{ + return (ch < 0x80) && (isalnum(ch) || ch == '_'); +} + +static inline bool IsAWordStart(const int ch) +{ + return (ch < 0x80) && (isalnum(ch) || ch == '_' || ch == '@' || ch == '#' || ch == '$' || ch == '.'); +} + +static inline bool IsAOperator(char ch) { + if (isascii(ch) && isalnum(ch)) + return false; + if (ch == '+' || ch == '-' || ch == '*' || ch == '/' || + ch == '&' || ch == '^' || ch == '=' || ch == '<' || ch == '>' || + ch == '(' || ch == ')' || ch == '[' || ch == ']' || ch == ',' ) + return true; + return false; +} + +/////////////////////////////////////////////////////////////////////////////// +// GetSendKey() filters the portion before and after a/multiple space(s) +// and return the first portion to be looked-up in the table +// also check if the second portion is valid... (up,down.on.off,toggle or a number) +/////////////////////////////////////////////////////////////////////////////// + +static int GetSendKey(const char *szLine, char *szKey) +{ + int nFlag = 0; + int nStartFound = 0; + int nKeyPos = 0; + int nSpecPos= 0; + int nSpecNum= 1; + int nPos = 0; + char cTemp; + char szSpecial[100]; + + // split the portion of the sendkey in the part before and after the spaces + while ( ( (cTemp = szLine[nPos]) != '\0')) + { + // skip leading Ctrl/Shift/Alt state + if (cTemp == '{') { + nStartFound = 1; + } + // + if (nStartFound == 1) { + if ((cTemp == ' ') && (nFlag == 0) ) // get the stuff till first space + { + nFlag = 1; + // Add } to the end of the first bit for table lookup later. + szKey[nKeyPos++] = '}'; + } + else if (cTemp == ' ') + { + // skip other spaces + } + else if (nFlag == 0) + { + // save first portion into var till space or } is hit + szKey[nKeyPos++] = cTemp; + } + else if ((nFlag == 1) && (cTemp != '}')) + { + // Save second portion into var... + szSpecial[nSpecPos++] = cTemp; + // check if Second portion is all numbers for repeat fuction + if (isdigit(cTemp) == false) {nSpecNum = 0;} + } + } + nPos++; // skip to next char + + } // End While + + + // Check if the second portion is either a number or one of these keywords + szKey[nKeyPos] = '\0'; + szSpecial[nSpecPos] = '\0'; + if (strcmp(szSpecial,"down")== 0 || strcmp(szSpecial,"up")== 0 || + strcmp(szSpecial,"on")== 0 || strcmp(szSpecial,"off")== 0 || + strcmp(szSpecial,"toggle")== 0 || nSpecNum == 1 ) + { + nFlag = 0; + } + else + { + nFlag = 1; + } + return nFlag; // 1 is bad, 0 is good + +} // GetSendKey() + +// +// Routine to check the last "none comment" character on a line to see if its a continuation +// +static bool IsContinuationLine(unsigned int szLine, Accessor &styler) +{ + int nsPos = styler.LineStart(szLine); + int nePos = styler.LineStart(szLine+1) - 2; + //int stylech = styler.StyleAt(nsPos); + while (nsPos < nePos) + { + //stylech = styler.StyleAt(nePos); + int stylech = styler.StyleAt(nsPos); + if (!(stylech == SCE_AU3_COMMENT)) { + char ch = styler.SafeGetCharAt(nePos); + if (!isspacechar(ch)) { + if (ch == '_') + return true; + else + return false; + } + } + nePos--; // skip to next char + } // End While + return false; +} // IsContinuationLine() + +// +// syntax highlighting logic +static void ColouriseAU3Doc(unsigned int startPos, + int length, int initStyle, + WordList *keywordlists[], + Accessor &styler) { + + WordList &keywords = *keywordlists[0]; + WordList &keywords2 = *keywordlists[1]; + WordList &keywords3 = *keywordlists[2]; + WordList &keywords4 = *keywordlists[3]; + WordList &keywords5 = *keywordlists[4]; + WordList &keywords6 = *keywordlists[5]; + WordList &keywords7 = *keywordlists[6]; + WordList &keywords8 = *keywordlists[7]; + // find the first previous line without continuation character at the end + int lineCurrent = styler.GetLine(startPos); + int s_startPos = startPos; + // When not inside a Block comment: find First line without _ + if (!(initStyle==SCE_AU3_COMMENTBLOCK)) { + while ((lineCurrent > 0 && IsContinuationLine(lineCurrent,styler)) || + (lineCurrent > 1 && IsContinuationLine(lineCurrent-1,styler))) { + lineCurrent--; + startPos = styler.LineStart(lineCurrent); // get start position + initStyle = 0; // reset the start style to 0 + } + } + // Set the new length to include it from the start and set the start position + length = length + s_startPos - startPos; // correct the total length to process + styler.StartAt(startPos); + + StyleContext sc(startPos, length, initStyle, styler); + char si; // string indicator "=1 '=2 + char ni; // Numeric indicator error=9 normal=0 normal+dec=1 hex=2 Enot=3 + char ci; // comment indicator 0=not linecomment(;) + char s_save[100]; + si=0; + ni=0; + ci=0; + //$$$ + for (; sc.More(); sc.Forward()) { + char s[100]; + sc.GetCurrentLowered(s, sizeof(s)); + // ********************************************** + // save the total current word for eof processing + if (IsAWordChar(sc.ch) || sc.ch == '}') + { + strcpy(s_save,s); + int tp = strlen(s_save); + if (tp < 99) { + s_save[tp] = static_cast<char>(tolower(sc.ch)); + s_save[tp+1] = '\0'; + } + } + // ********************************************** + // + switch (sc.state) + { + case SCE_AU3_COMMENTBLOCK: + { + //Reset at line end + if (sc.atLineEnd) { + ci=0; + sc.SetState(SCE_AU3_COMMENTBLOCK); + } + //skip rest of line when a ; is encountered + if (sc.chPrev == ';') { + ci=2; + sc.SetState(SCE_AU3_COMMENTBLOCK); + } + // skip rest of the line + if (ci==2) + break; + // check when first character is detected on the line + if (ci==0) { + if (IsAWordStart(static_cast<char>(sc.ch)) || IsAOperator(static_cast<char>(sc.ch))) { + ci=1; + sc.SetState(SCE_AU3_COMMENTBLOCK); + } + break; + } + if (!(IsAWordChar(sc.ch) || (sc.ch == '-' && strcmp(s, "#comments") == 0))) { + if ((strcmp(s, "#ce")== 0 || strcmp(s, "#comments-end")== 0)) + sc.SetState(SCE_AU3_COMMENT); // set to comment line for the rest of the line + else + ci=2; // line doesn't begin with #CE so skip the rest of the line + } + break; + } + case SCE_AU3_COMMENT: + { + if (sc.atLineEnd) {sc.SetState(SCE_AU3_DEFAULT);} + break; + } + case SCE_AU3_OPERATOR: + { + // check if its a COMobject + if (sc.chPrev == '.' && IsAWordChar(sc.ch)) { + sc.SetState(SCE_AU3_COMOBJ); + } + else { + sc.SetState(SCE_AU3_DEFAULT); + } + break; + } + case SCE_AU3_SPECIAL: + { + if (sc.ch == ';') {sc.SetState(SCE_AU3_COMMENT);} + if (sc.atLineEnd) {sc.SetState(SCE_AU3_DEFAULT);} + break; + } + case SCE_AU3_KEYWORD: + { + if (!(IsAWordChar(sc.ch) || (sc.ch == '-' && (strcmp(s, "#comments") == 0 || strcmp(s, "#include") == 0)))) + { + if (!IsTypeCharacter(sc.ch)) + { + if (strcmp(s, "#cs")== 0 || strcmp(s, "#comments-start")== 0 ) + { + sc.ChangeState(SCE_AU3_COMMENTBLOCK); + sc.SetState(SCE_AU3_COMMENTBLOCK); + } + else if (keywords.InList(s)) { + sc.ChangeState(SCE_AU3_KEYWORD); + sc.SetState(SCE_AU3_DEFAULT); + } + else if (keywords2.InList(s)) { + sc.ChangeState(SCE_AU3_FUNCTION); + sc.SetState(SCE_AU3_DEFAULT); + } + else if (keywords3.InList(s)) { + sc.ChangeState(SCE_AU3_MACRO); + sc.SetState(SCE_AU3_DEFAULT); + } + else if (keywords5.InList(s)) { + sc.ChangeState(SCE_AU3_PREPROCESSOR); + sc.SetState(SCE_AU3_DEFAULT); + if (strcmp(s, "#include")== 0) + { + si = 3; // use to determine string start for #inlude <> + } + } + else if (keywords6.InList(s)) { + sc.ChangeState(SCE_AU3_SPECIAL); + sc.SetState(SCE_AU3_SPECIAL); + } + else if ((keywords7.InList(s)) && (!IsAOperator(static_cast<char>(sc.ch)))) { + sc.ChangeState(SCE_AU3_EXPAND); + sc.SetState(SCE_AU3_DEFAULT); + } + else if (keywords8.InList(s)) { + sc.ChangeState(SCE_AU3_UDF); + sc.SetState(SCE_AU3_DEFAULT); + } + else if (strcmp(s, "_") == 0) { + sc.ChangeState(SCE_AU3_OPERATOR); + sc.SetState(SCE_AU3_DEFAULT); + } + else if (!IsAWordChar(sc.ch)) { + sc.ChangeState(SCE_AU3_DEFAULT); + sc.SetState(SCE_AU3_DEFAULT); + } + } + } + if (sc.atLineEnd) { + sc.SetState(SCE_AU3_DEFAULT);} + break; + } + case SCE_AU3_NUMBER: + { + // Numeric indicator error=9 normal=0 normal+dec=1 hex=2 E-not=3 + // + // test for Hex notation + if (strcmp(s, "0") == 0 && (sc.ch == 'x' || sc.ch == 'X') && ni == 0) + { + ni = 2; + break; + } + // test for E notation + if (IsADigit(sc.chPrev) && (sc.ch == 'e' || sc.ch == 'E') && ni <= 1) + { + ni = 3; + break; + } + // Allow Hex characters inside hex numeric strings + if ((ni == 2) && + (sc.ch == 'a' || sc.ch == 'b' || sc.ch == 'c' || sc.ch == 'd' || sc.ch == 'e' || sc.ch == 'f' || + sc.ch == 'A' || sc.ch == 'B' || sc.ch == 'C' || sc.ch == 'D' || sc.ch == 'E' || sc.ch == 'F' )) + { + break; + } + // test for 1 dec point only + if (sc.ch == '.') + { + if (ni==0) + { + ni=1; + } + else + { + ni=9; + } + break; + } + // end of numeric string ? + if (!(IsADigit(sc.ch))) + { + if (ni==9) + { + sc.ChangeState(SCE_AU3_DEFAULT); + } + sc.SetState(SCE_AU3_DEFAULT); + } + break; + } + case SCE_AU3_VARIABLE: + { + // Check if its a COMObject + if (sc.ch == '.' && !IsADigit(sc.chNext)) { + sc.SetState(SCE_AU3_OPERATOR); + } + else if (!IsAWordChar(sc.ch)) { + sc.SetState(SCE_AU3_DEFAULT); + } + break; + } + case SCE_AU3_COMOBJ: + { + if (!(IsAWordChar(sc.ch))) { + sc.SetState(SCE_AU3_DEFAULT); + } + break; + } + case SCE_AU3_STRING: + { + // check for " to end a double qouted string or + // check for ' to end a single qouted string + if ((si == 1 && sc.ch == '\"') || (si == 2 && sc.ch == '\'') || (si == 3 && sc.ch == '>')) + { + sc.ForwardSetState(SCE_AU3_DEFAULT); + si=0; + } + if (sc.atLineEnd) + { + si=0; + // at line end and not found a continuation char then reset to default + int lineCurrent = styler.GetLine(sc.currentPos); + if (!IsContinuationLine(lineCurrent,styler)) + { + sc.SetState(SCE_AU3_DEFAULT); + } + } + // find Sendkeys in a STRING + if (sc.ch == '{' || sc.ch == '+' || sc.ch == '!' || sc.ch == '^' || sc.ch == '#' ) { + sc.SetState(SCE_AU3_SENT);} + break; + } + + case SCE_AU3_SENT: + { + // Send key string ended + if (sc.chPrev == '}' && sc.ch != '}') + { + // set color to SENDKEY when valid sendkey .. else set back to regular string + char sk[100]; + // split {111 222} and return {111} and check if 222 is valid. + // if return code = 1 then invalid 222 so must be string + if (GetSendKey(s,sk)) + { + sc.ChangeState(SCE_AU3_STRING); + } + // if single char between {?} then its ok as sendkey for a single character + else if (strlen(sk) == 3) + { + sc.ChangeState(SCE_AU3_SENT); + } + // if sendkey {111} is in table then ok as sendkey + else if (keywords4.InList(sk)) + { + sc.ChangeState(SCE_AU3_SENT); + } + else + { + sc.ChangeState(SCE_AU3_STRING); + } + sc.SetState(SCE_AU3_STRING); + } + else + { + // check if the start is a valid SendKey start + int nPos = 0; + int nState = 1; + char cTemp; + while (!(nState == 2) && ((cTemp = s[nPos]) != '\0')) + { + if (cTemp == '{' && nState == 1) + { + nState = 2; + } + if (nState == 1 && !(cTemp == '+' || cTemp == '!' || cTemp == '^' || cTemp == '#' )) + { + nState = 0; + } + nPos++; + } + //Verify characters infront of { ... if not assume regular string + if (nState == 1 && (!(sc.ch == '{' || sc.ch == '+' || sc.ch == '!' || sc.ch == '^' || sc.ch == '#' ))) { + sc.ChangeState(SCE_AU3_STRING); + sc.SetState(SCE_AU3_STRING); + } + // If invalid character found then assume its a regular string + if (nState == 0) { + sc.ChangeState(SCE_AU3_STRING); + sc.SetState(SCE_AU3_STRING); + } + } + // check if next portion is again a sendkey + if (sc.atLineEnd) + { + sc.ChangeState(SCE_AU3_STRING); + sc.SetState(SCE_AU3_DEFAULT); + si = 0; // reset string indicator + } + //* check in next characters following a sentkey are again a sent key + // Need this test incase of 2 sentkeys like {F1}{ENTER} but not detect {{} + if (sc.state == SCE_AU3_STRING && (sc.ch == '{' || sc.ch == '+' || sc.ch == '!' || sc.ch == '^' || sc.ch == '#' )) { + sc.SetState(SCE_AU3_SENT);} + // check to see if the string ended... + // Sendkey string isn't complete but the string ended.... + if ((si == 1 && sc.ch == '\"') || (si == 2 && sc.ch == '\'')) + { + sc.ChangeState(SCE_AU3_STRING); + sc.ForwardSetState(SCE_AU3_DEFAULT); + } + break; + } + } //switch (sc.state) + + // Determine if a new state should be entered: + + if (sc.state == SCE_AU3_DEFAULT) + { + if (sc.ch == ';') {sc.SetState(SCE_AU3_COMMENT);} + else if (sc.ch == '#') {sc.SetState(SCE_AU3_KEYWORD);} + else if (sc.ch == '$') {sc.SetState(SCE_AU3_VARIABLE);} + else if (sc.ch == '.' && !IsADigit(sc.chNext)) {sc.SetState(SCE_AU3_OPERATOR);} + else if (sc.ch == '@') {sc.SetState(SCE_AU3_KEYWORD);} + //else if (sc.ch == '_') {sc.SetState(SCE_AU3_KEYWORD);} + else if (sc.ch == '<' && si==3) {sc.SetState(SCE_AU3_STRING);} // string after #include + else if (sc.ch == '\"') { + sc.SetState(SCE_AU3_STRING); + si = 1; } + else if (sc.ch == '\'') { + sc.SetState(SCE_AU3_STRING); + si = 2; } + else if (IsADigit(sc.ch) || (sc.ch == '.' && IsADigit(sc.chNext))) + { + sc.SetState(SCE_AU3_NUMBER); + ni = 0; + } + else if (IsAWordStart(sc.ch)) {sc.SetState(SCE_AU3_KEYWORD);} + else if (IsAOperator(static_cast<char>(sc.ch))) {sc.SetState(SCE_AU3_OPERATOR);} + else if (sc.atLineEnd) {sc.SetState(SCE_AU3_DEFAULT);} + } + } //for (; sc.More(); sc.Forward()) + + //************************************* + // Colourize the last word correctly + //************************************* + if (sc.state == SCE_AU3_KEYWORD) + { + if (strcmp(s_save, "#cs")== 0 || strcmp(s_save, "#comments-start")== 0 ) + { + sc.ChangeState(SCE_AU3_COMMENTBLOCK); + sc.SetState(SCE_AU3_COMMENTBLOCK); + } + else if (keywords.InList(s_save)) { + sc.ChangeState(SCE_AU3_KEYWORD); + sc.SetState(SCE_AU3_KEYWORD); + } + else if (keywords2.InList(s_save)) { + sc.ChangeState(SCE_AU3_FUNCTION); + sc.SetState(SCE_AU3_FUNCTION); + } + else if (keywords3.InList(s_save)) { + sc.ChangeState(SCE_AU3_MACRO); + sc.SetState(SCE_AU3_MACRO); + } + else if (keywords5.InList(s_save)) { + sc.ChangeState(SCE_AU3_PREPROCESSOR); + sc.SetState(SCE_AU3_PREPROCESSOR); + } + else if (keywords6.InList(s_save)) { + sc.ChangeState(SCE_AU3_SPECIAL); + sc.SetState(SCE_AU3_SPECIAL); + } + else if (keywords7.InList(s_save) && sc.atLineEnd) { + sc.ChangeState(SCE_AU3_EXPAND); + sc.SetState(SCE_AU3_EXPAND); + } + else if (keywords8.InList(s_save)) { + sc.ChangeState(SCE_AU3_UDF); + sc.SetState(SCE_AU3_UDF); + } + else { + sc.ChangeState(SCE_AU3_DEFAULT); + sc.SetState(SCE_AU3_DEFAULT); + } + } + if (sc.state == SCE_AU3_SENT) + { + // Send key string ended + if (sc.chPrev == '}' && sc.ch != '}') + { + // set color to SENDKEY when valid sendkey .. else set back to regular string + char sk[100]; + // split {111 222} and return {111} and check if 222 is valid. + // if return code = 1 then invalid 222 so must be string + if (GetSendKey(s_save,sk)) + { + sc.ChangeState(SCE_AU3_STRING); + } + // if single char between {?} then its ok as sendkey for a single character + else if (strlen(sk) == 3) + { + sc.ChangeState(SCE_AU3_SENT); + } + // if sendkey {111} is in table then ok as sendkey + else if (keywords4.InList(sk)) + { + sc.ChangeState(SCE_AU3_SENT); + } + else + { + sc.ChangeState(SCE_AU3_STRING); + } + sc.SetState(SCE_AU3_STRING); + } + // check if next portion is again a sendkey + if (sc.atLineEnd) + { + sc.ChangeState(SCE_AU3_STRING); + sc.SetState(SCE_AU3_DEFAULT); + } + } + //************************************* + sc.Complete(); +} + +// +static bool IsStreamCommentStyle(int style) { + return style == SCE_AU3_COMMENT || style == SCE_AU3_COMMENTBLOCK; +} + +// +// Routine to find first none space on the current line and return its Style +// needed for comment lines not starting on pos 1 +static int GetStyleFirstWord(unsigned int szLine, Accessor &styler) +{ + int nsPos = styler.LineStart(szLine); + int nePos = styler.LineStart(szLine+1) - 1; + while (isspacechar(styler.SafeGetCharAt(nsPos)) && nsPos < nePos) + { + nsPos++; // skip to next char + + } // End While + return styler.StyleAt(nsPos); + +} // GetStyleFirstWord() + + +// +static void FoldAU3Doc(unsigned int startPos, int length, int, WordList *[], Accessor &styler) +{ + int endPos = startPos + length; + // get settings from the config files for folding comments and preprocessor lines + bool foldComment = styler.GetPropertyInt("fold.comment") != 0; + bool foldInComment = styler.GetPropertyInt("fold.comment") == 2; + bool foldCompact = styler.GetPropertyInt("fold.compact", 1) != 0; + bool foldpreprocessor = styler.GetPropertyInt("fold.preprocessor") != 0; + // Backtrack to previous line in case need to fix its fold status + int lineCurrent = styler.GetLine(startPos); + if (startPos > 0) { + if (lineCurrent > 0) { + lineCurrent--; + startPos = styler.LineStart(lineCurrent); + } + } + // vars for style of previous/current/next lines + int style = GetStyleFirstWord(lineCurrent,styler); + int stylePrev = 0; + // find the first previous line without continuation character at the end + while ((lineCurrent > 0 && IsContinuationLine(lineCurrent,styler)) || + (lineCurrent > 1 && IsContinuationLine(lineCurrent-1,styler))) { + lineCurrent--; + startPos = styler.LineStart(lineCurrent); + } + if (lineCurrent > 0) { + stylePrev = GetStyleFirstWord(lineCurrent-1,styler); + } + // vars for getting first word to check for keywords + bool FirstWordStart = false; + bool FirstWordEnd = false; + char szKeyword[10]=""; + int szKeywordlen = 0; + char szThen[5]=""; + int szThenlen = 0; + bool ThenFoundLast = false; + // var for indentlevel + int levelCurrent = SC_FOLDLEVELBASE; + if (lineCurrent > 0) + levelCurrent = styler.LevelAt(lineCurrent-1) >> 16; + int levelNext = levelCurrent; + // + int visibleChars = 0; + char chNext = styler.SafeGetCharAt(startPos); + char chPrev = ' '; + // + for (int i = startPos; i < endPos; i++) { + char ch = chNext; + chNext = styler.SafeGetCharAt(i + 1); + if (IsAWordChar(ch)) { + visibleChars++; + } + // get the syle for the current character neede to check in comment + int stylech = styler.StyleAt(i); + // get first word for the line for indent check max 9 characters + if (FirstWordStart && (!(FirstWordEnd))) { + if (!IsAWordChar(ch)) { + FirstWordEnd = true; + szKeyword[szKeywordlen] = '\0'; + } + else { + if (szKeywordlen < 10) { + szKeyword[szKeywordlen++] = static_cast<char>(tolower(ch)); + } + } + } + // start the capture of the first word + if (!(FirstWordStart)) { + if (IsAWordChar(ch) || IsAWordStart(ch) || ch == ';') { + FirstWordStart = true; + szKeyword[szKeywordlen++] = static_cast<char>(tolower(ch)); + } + } + // only process this logic when not in comment section + if (!(stylech == SCE_AU3_COMMENT)) { + if (ThenFoundLast) { + if (IsAWordChar(ch)) { + ThenFoundLast = false; + } + } + // find out if the word "then" is the last on a "if" line + if (FirstWordEnd && strcmp(szKeyword,"if") == 0) { + if (szThenlen == 4) { + szThen[0] = szThen[1]; + szThen[1] = szThen[2]; + szThen[2] = szThen[3]; + szThen[3] = static_cast<char>(tolower(ch)); + if (strcmp(szThen,"then") == 0 ) { + ThenFoundLast = true; + } + } + else { + szThen[szThenlen++] = static_cast<char>(tolower(ch)); + if (szThenlen == 5) { + szThen[4] = '\0'; + } + } + } + } + // End of Line found so process the information + if ((ch == '\r' && chNext != '\n') || (ch == '\n') || (i == endPos)) { + // ************************** + // Folding logic for Keywords + // ************************** + // if a keyword is found on the current line and the line doesn't end with _ (continuation) + // and we are not inside a commentblock. + if (szKeywordlen > 0 && (!(chPrev == '_')) && + ((!(IsStreamCommentStyle(style)) || foldInComment)) ) { + szKeyword[szKeywordlen] = '\0'; + // only fold "if" last keyword is "then" (else its a one line if) + if (strcmp(szKeyword,"if") == 0 && ThenFoundLast) { + levelNext++; + } + // create new fold for these words + if (strcmp(szKeyword,"do") == 0 || strcmp(szKeyword,"for") == 0 || + strcmp(szKeyword,"func") == 0 || strcmp(szKeyword,"while") == 0|| + strcmp(szKeyword,"with") == 0 || strcmp(szKeyword,"#region") == 0 ) { + levelNext++; + } + // create double Fold for select&switch because Case will subtract one of the current level + if (strcmp(szKeyword,"select") == 0 || strcmp(szKeyword,"switch") == 0) { + levelNext++; + levelNext++; + } + // end the fold for these words before the current line + if (strcmp(szKeyword,"endfunc") == 0 || strcmp(szKeyword,"endif") == 0 || + strcmp(szKeyword,"next") == 0 || strcmp(szKeyword,"until") == 0 || + strcmp(szKeyword,"endwith") == 0 ||strcmp(szKeyword,"wend") == 0){ + levelNext--; + levelCurrent--; + } + // end the fold for these words before the current line and Start new fold + if (strcmp(szKeyword,"case") == 0 || strcmp(szKeyword,"else") == 0 || + strcmp(szKeyword,"elseif") == 0 ) { + levelCurrent--; + } + // end the double fold for this word before the current line + if (strcmp(szKeyword,"endselect") == 0 || strcmp(szKeyword,"endswitch") == 0 ) { + levelNext--; + levelNext--; + levelCurrent--; + levelCurrent--; + } + // end the fold for these words on the current line + if (strcmp(szKeyword,"#endregion") == 0 ) { + levelNext--; + } + } + // Preprocessor and Comment folding + int styleNext = GetStyleFirstWord(lineCurrent + 1,styler); + // ************************************* + // Folding logic for preprocessor blocks + // ************************************* + // process preprosessor line + if (foldpreprocessor && style == SCE_AU3_PREPROCESSOR) { + if (!(stylePrev == SCE_AU3_PREPROCESSOR) && (styleNext == SCE_AU3_PREPROCESSOR)) { + levelNext++; + } + // fold till the last line for normal comment lines + else if (stylePrev == SCE_AU3_PREPROCESSOR && !(styleNext == SCE_AU3_PREPROCESSOR)) { + levelNext--; + } + } + // ********************************* + // Folding logic for Comment blocks + // ********************************* + if (foldComment && IsStreamCommentStyle(style)) { + // Start of a comment block + if (!(stylePrev==style) && IsStreamCommentStyle(styleNext) && styleNext==style) { + levelNext++; + } + // fold till the last line for normal comment lines + else if (IsStreamCommentStyle(stylePrev) + && !(styleNext == SCE_AU3_COMMENT) + && stylePrev == SCE_AU3_COMMENT + && style == SCE_AU3_COMMENT) { + levelNext--; + } + // fold till the one but last line for Blockcomment lines + else if (IsStreamCommentStyle(stylePrev) + && !(styleNext == SCE_AU3_COMMENTBLOCK) + && style == SCE_AU3_COMMENTBLOCK) { + levelNext--; + levelCurrent--; + } + } + int levelUse = levelCurrent; + 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); + } + // reset values for the next line + lineCurrent++; + stylePrev = style; + style = styleNext; + levelCurrent = levelNext; + visibleChars = 0; + // if the last character is an Underscore then don't reset since the line continues on the next line. + if (!(chPrev == '_')) { + szKeywordlen = 0; + szThenlen = 0; + FirstWordStart = false; + FirstWordEnd = false; + ThenFoundLast = false; + } + } + // save the last processed character + if (!isspacechar(ch)) { + chPrev = ch; + visibleChars++; + } + } +} + + +// + +static const char * const AU3WordLists[] = { + "#autoit keywords", + "#autoit functions", + "#autoit macros", + "#autoit Sent keys", + "#autoit Pre-processors", + "#autoit Special", + "#autoit Expand", + "#autoit UDF", + 0 +}; +LexerModule lmAU3(SCLEX_AU3, ColouriseAU3Doc, "au3", FoldAU3Doc , AU3WordLists); |