summaryrefslogtreecommitdiffstats
path: root/libkcal/versit/vcc.y
diff options
context:
space:
mode:
Diffstat (limited to 'libkcal/versit/vcc.y')
-rw-r--r--libkcal/versit/vcc.y1232
1 files changed, 1232 insertions, 0 deletions
diff --git a/libkcal/versit/vcc.y b/libkcal/versit/vcc.y
new file mode 100644
index 000000000..fbfee0f6e
--- /dev/null
+++ b/libkcal/versit/vcc.y
@@ -0,0 +1,1232 @@
+%{
+
+/***************************************************************************
+(C) Copyright 1996 Apple Computer, Inc., AT&T Corp., International
+Business Machines Corporation and Siemens Rolm Communications Inc.
+
+For purposes of this license notice, the term Licensors shall mean,
+collectively, Apple Computer, Inc., AT&T Corp., International
+Business Machines Corporation and Siemens Rolm Communications Inc.
+The term Licensor shall mean any of the Licensors.
+
+Subject to acceptance of the following conditions, permission is hereby
+granted by Licensors without the need for written agreement and without
+license or royalty fees, to use, copy, modify and distribute this
+software for any purpose.
+
+The above copyright notice and the following four paragraphs must be
+reproduced in all copies of this software and any software including
+this software.
+
+THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS AND NO LICENSOR SHALL HAVE
+ANY OBLIGATION TO PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS OR
+MODIFICATIONS.
+
+IN NO EVENT SHALL ANY LICENSOR BE LIABLE TO ANY PARTY FOR DIRECT,
+INDIRECT, SPECIAL OR CONSEQUENTIAL DAMAGES OR LOST PROFITS ARISING OUT
+OF THE USE OF THIS SOFTWARE EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+DAMAGE.
+
+EACH LICENSOR SPECIFICALLY DISCLAIMS ANY WARRANTIES, EXPRESS OR IMPLIED,
+INCLUDING BUT NOT LIMITED TO ANY WARRANTY OF NONINFRINGEMENT OR THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.
+
+The software is provided with RESTRICTED RIGHTS. Use, duplication, or
+disclosure by the government are subject to restrictions set forth in
+DFARS 252.227-7013 or 48 CFR 52.227-19, as applicable.
+
+***************************************************************************/
+
+/*
+ * src: vcc.c
+ * doc: Parser for vCard and vCalendar. Note that this code is
+ * generated by a yacc parser generator. Generally it should not
+ * be edited by hand. The real source is vcc.y. The #line directives
+ * can be commented out here to make it easier to trace through
+ * in a debugger. However, if a bug is found it should
+ * be fixed in vcc.y and this file regenerated.
+ */
+
+
+/* debugging utilities */
+#if __DEBUG
+#define DBG_(x) printf x
+#else
+#define DBG_(x)
+#endif
+
+/**** External Functions ****/
+
+/* assign local name to parser variables and functions so that
+ we can use more than one yacc based parser.
+*/
+
+#define yyparse mime_parse
+#define yylex mime_lex
+#define yyerror mime_error
+#define yychar mime_char
+/* #define p_yyval p_mime_val */
+#undef yyval
+#define yyval mime_yyval
+/* #define p_yylval p_mime_lval */
+#undef yylval
+#define yylval mime_yylval
+#define yydebug mime_debug
+#define yynerrs mime_nerrs
+#define yyerrflag mime_errflag
+#define yyss mime_ss
+#define yyssp mime_ssp
+#define yyvs mime_vs
+#define yyvsp mime_vsp
+#define yylhs mime_lhs
+#define yylen mime_len
+#define yydefred mime_defred
+#define yydgoto mime_dgoto
+#define yysindex mime_sindex
+#define yyrindex mime_rindex
+#define yygindex mime_gindex
+#define yytable mime_table
+#define yycheck mime_check
+#define yyname mime_name
+#define yyrule mime_rule
+#undef YYPREFIX
+#define YYPREFIX "mime_"
+
+
+#ifndef _NO_LINE_FOLDING
+#define _SUPPORT_LINE_FOLDING 1
+#endif
+
+#include <string.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <ctype.h>
+#include "vcc.h"
+
+/* The following is a hack that I hope will get things compiling
+ * on SunOS 4.1.x systems
+ */
+#ifndef SEEK_SET
+#define SEEK_SET 0 /* Seek from beginning of file. */
+#define SEEK_CUR 1 /* Seek from current position. */
+#define SEEK_END 2 /* Seek from end of file. */
+#endif
+
+/**** Types, Constants ****/
+
+#define YYDEBUG 0 /* 1 to compile in some debugging code */
+#define MAXTOKEN 256 /* maximum token (line) length */
+#define YYSTACKSIZE 1000 /* ~unref ? */
+#define MAXLEVEL 10 /* max # of nested objects parseable */
+ /* (includes outermost) */
+
+
+/**** Global Variables ****/
+int mime_lineNum, mime_numErrors; /* yyerror() can use these */
+static VObject* vObjList;
+static VObject *curProp;
+static VObject *curObj;
+static VObject* ObjStack[MAXLEVEL];
+static int ObjStackTop;
+
+
+/* A helpful utility for the rest of the app. */
+#if __CPLUSPLUS__
+extern "C" {
+#endif
+
+ /* static void Parse_Debug(const char *s);*/
+ static void yyerror(char *s);
+
+#if __CPLUSPLUS__
+ };
+#endif
+
+int yyparse();
+static int yylex();
+
+static VObject* popVObject();
+
+
+enum LexMode {
+ L_NORMAL,
+ L_VCARD,
+ L_VCAL,
+ L_VEVENT,
+ L_VTODO,
+ L_VALUES,
+ L_BASE64,
+ L_QUOTED_PRINTABLE
+ };
+
+/**** Private Forward Declarations ****/
+static int pushVObject(const char *prop);
+static VObject* popVObject();
+char* lexDataFromBase64();
+static void lexPopMode(int top);
+static int lexWithinMode(enum LexMode mode);
+static void lexPushMode(enum LexMode mode);
+static void enterProps(const char *s);
+static void enterAttr(const char *s1, const char *s2);
+/* static void enterValues(const char *value); */
+static void appendValue(const char *value);
+static void mime_error_(char *s);
+
+%}
+
+/***************************************************************************/
+/*** The grammar ****/
+/***************************************************************************/
+
+%union {
+ char *str;
+ VObject *vobj;
+ }
+
+%token
+ EQ COLON DOT SEMICOLON SPACE HTAB LINESEP NEWLINE
+ BEGIN_VCARD END_VCARD BEGIN_VCAL END_VCAL
+ BEGIN_VEVENT END_VEVENT BEGIN_VTODO END_VTODO
+ ID
+
+/*
+ * NEWLINE is the token that would occur outside a vCard,
+ * while LINESEP is the token that would occur inside a vCard.
+ */
+
+%token <str>
+ STRING ID
+
+%type <str> name value
+
+%type <vobj> vcard vcal vobject
+
+%start mime
+
+%%
+
+
+mime: vobjects
+ ;
+
+vobjects: vobject
+ { addList(&vObjList, $1); curObj = 0; }
+ vobjects
+ | vobject
+ { addList(&vObjList, $1); curObj = 0; }
+ ;
+
+vobject: vcard
+ | vcal
+ ;
+
+vcard:
+ BEGIN_VCARD
+ {
+ lexPushMode(L_VCARD);
+ if (!pushVObject(VCCardProp)) YYERROR;
+ }
+ items END_VCARD
+ {
+ lexPopMode(0);
+ $$ = popVObject();
+ }
+ | BEGIN_VCARD
+ {
+ lexPushMode(L_VCARD);
+ if (!pushVObject(VCCardProp)) YYERROR;
+ }
+ END_VCARD
+ {
+ lexPopMode(0);
+ $$ = popVObject();
+ }
+ ;
+
+items: item items
+ | item
+ ;
+
+item: prop COLON
+ {
+ lexPushMode(L_VALUES);
+ }
+ values LINESEP
+ {
+ if (lexWithinMode(L_BASE64) || lexWithinMode(L_QUOTED_PRINTABLE))
+ lexPopMode(0);
+ lexPopMode(0);
+ }
+ | error
+ ;
+
+prop: name
+ {
+ enterProps($1);
+ }
+ attr_params
+ | name
+ {
+ enterProps($1);
+ }
+ ;
+
+attr_params: attr_param attr_params
+ | attr_param
+ ;
+
+attr_param: SEMICOLON attr
+ ;
+
+attr: name
+ {
+ enterAttr($1,0);
+ }
+ | name EQ name
+ {
+ enterAttr($1,$3);
+
+ }
+ ;
+
+name: ID
+ ;
+
+values: value SEMICOLON { appendValue($1); } values
+ | value
+ { appendValue($1); }
+ ;
+
+value: STRING
+ |
+ { $$ = 0; }
+ ;
+
+vcal:
+ BEGIN_VCAL
+ { if (!pushVObject(VCCalProp)) YYERROR; }
+ calitems
+ END_VCAL
+ { $$ = popVObject(); }
+ | BEGIN_VCAL
+ { if (!pushVObject(VCCalProp)) YYERROR; }
+ END_VCAL
+ { $$ = popVObject(); }
+ ;
+
+calitems: calitem calitems
+ | calitem
+ ;
+
+calitem:
+ eventitem
+ | todoitem
+ | items
+ ;
+
+eventitem:
+ BEGIN_VEVENT
+ {
+ lexPushMode(L_VEVENT);
+ if (!pushVObject(VCEventProp)) YYERROR;
+ }
+ items
+ END_VEVENT
+ {
+ lexPopMode(0);
+ popVObject();
+ }
+ | BEGIN_VEVENT
+ {
+ lexPushMode(L_VEVENT);
+ if (!pushVObject(VCEventProp)) YYERROR;
+ }
+ END_VEVENT
+ {
+ lexPopMode(0);
+ popVObject();
+ }
+ ;
+
+todoitem:
+ BEGIN_VTODO
+ {
+ lexPushMode(L_VTODO);
+ if (!pushVObject(VCTodoProp)) YYERROR;
+ }
+ items
+ END_VTODO
+ {
+ lexPopMode(0);
+ popVObject();
+ }
+ | BEGIN_VTODO
+ {
+ lexPushMode(L_VTODO);
+ if (!pushVObject(VCTodoProp)) YYERROR;
+ }
+ END_VTODO
+ {
+ lexPopMode(0);
+ popVObject();
+ }
+ ;
+
+%%
+/****************************************************************************/
+static int pushVObject(const char *prop)
+ {
+ VObject *newObj;
+ if (ObjStackTop == MAXLEVEL)
+ return FALSE;
+
+ ObjStack[++ObjStackTop] = curObj;
+
+ if (curObj) {
+ newObj = addProp(curObj,prop);
+ curObj = newObj;
+ }
+ else
+ curObj = newVObject(prop);
+
+ return TRUE;
+ }
+
+
+/****************************************************************************/
+/* This pops the recently built vCard off the stack and returns it. */
+static VObject* popVObject()
+ {
+ VObject *oldObj;
+ if (ObjStackTop < 0) {
+ yyerror("pop on empty Object Stack\n");
+ return 0;
+ }
+ oldObj = curObj;
+ curObj = ObjStack[ObjStackTop--];
+
+ return oldObj;
+ }
+
+
+/* static void enterValues(const char *value) */
+/* { */
+/* if (fieldedProp && *fieldedProp) { */
+/* if (value) { */
+/* addPropValue(curProp,*fieldedProp,value); */
+/* } */
+ /* else this field is empty, advance to next field */
+/* fieldedProp++; */
+/* } */
+/* else { */
+/* if (value) { */
+/* setVObjectUStringZValue_(curProp,fakeUnicode(value,0)); */
+/* } */
+/* } */
+/* deleteStr(value); */
+/* } */
+
+static void appendValue(const char *value)
+{
+ char *p1, *p2;
+ wchar_t *p3;
+ int i;
+
+ if (fieldedProp && *fieldedProp) {
+ if (value) {
+ addPropValue(curProp, *fieldedProp, value);
+ }
+ /* else this field is empty, advance to next field */
+ fieldedProp++;
+ } else {
+ if (value) {
+ if (vObjectUStringZValue(curProp)) {
+ p1 = fakeCString(vObjectUStringZValue(curProp));
+ p2 = malloc(sizeof(char *) * (strlen(p1)+strlen(value)+1));
+ strcpy(p2, p1);
+ deleteStr(p1);
+
+ i = strlen(p2);
+ p2[i] = ',';
+ p2[i+1] = '\0';
+ p2 = strcat(p2, value);
+ p3 = (wchar_t *) vObjectUStringZValue(curProp);
+ free(p3);
+ setVObjectUStringZValue_(curProp,fakeUnicode(p2,0));
+ deleteStr(p2);
+ } else {
+ setVObjectUStringZValue_(curProp,fakeUnicode(value,0));
+ }
+ }
+ }
+ deleteStr(value);
+}
+
+
+static void enterProps(const char *s)
+ {
+ curProp = addGroup(curObj,s);
+ deleteStr(s);
+ }
+
+static void enterAttr(const char *s1, const char *s2)
+ {
+ const char *p1=0L, *p2=0L;
+ p1 = lookupProp_(s1);
+ if (s2) {
+ VObject *a;
+ p2 = lookupProp_(s2);
+ a = addProp(curProp,p1);
+ setVObjectStringZValue(a,p2);
+ }
+ else
+ addProp(curProp,p1);
+ if (strcasecmp(p1,VCBase64Prop) == 0 || (s2 && strcasecmp(p2,VCBase64Prop)==0))
+ lexPushMode(L_BASE64);
+ else if (strcasecmp(p1,VCQuotedPrintableProp) == 0
+ || (s2 && strcasecmp(p2,VCQuotedPrintableProp)==0))
+ lexPushMode(L_QUOTED_PRINTABLE);
+ deleteStr(s1); deleteStr(s2);
+ }
+
+
+#define MAX_LEX_LOOKAHEAD_0 32
+#define MAX_LEX_LOOKAHEAD 64
+#define MAX_LEX_MODE_STACK_SIZE 10
+#define LEXMODE() (lexBuf.lexModeStack[lexBuf.lexModeStackTop])
+
+struct LexBuf {
+ /* input */
+ FILE *inputFile;
+ char *inputString;
+ unsigned long curPos;
+ unsigned long inputLen;
+ /* lookahead buffer */
+ /* -- lookahead buffer is short instead of char so that EOF
+ / can be represented correctly.
+ */
+ unsigned long len;
+ short buf[MAX_LEX_LOOKAHEAD];
+ unsigned long getPtr;
+ /* context stack */
+ unsigned long lexModeStackTop;
+ enum LexMode lexModeStack[MAX_LEX_MODE_STACK_SIZE];
+ /* token buffer */
+ unsigned long maxToken;
+ char *strs;
+ unsigned long strsLen;
+ } lexBuf;
+
+static void lexPushMode(enum LexMode mode)
+ {
+ if (lexBuf.lexModeStackTop == (MAX_LEX_MODE_STACK_SIZE-1))
+ yyerror("lexical context stack overflow");
+ else {
+ lexBuf.lexModeStack[++lexBuf.lexModeStackTop] = mode;
+ }
+ }
+
+static void lexPopMode(int top)
+ {
+ /* special case of pop for ease of error recovery -- this
+ version will never underflow */
+ if (top)
+ lexBuf.lexModeStackTop = 0;
+ else
+ if (lexBuf.lexModeStackTop > 0) lexBuf.lexModeStackTop--;
+ }
+
+static int lexWithinMode(enum LexMode mode) {
+ unsigned long i;
+ for (i=0;i<lexBuf.lexModeStackTop;i++)
+ if (mode == lexBuf.lexModeStack[i]) return 1;
+ return 0;
+ }
+
+static int lexGetc_()
+ {
+ /* get next char from input, no buffering. */
+ if (lexBuf.curPos == lexBuf.inputLen)
+ return EOF;
+ else if (lexBuf.inputString)
+ return *(lexBuf.inputString + lexBuf.curPos++);
+ else {
+ if (!feof(lexBuf.inputFile))
+ return fgetc(lexBuf.inputFile);
+ else
+ return EOF;
+ }
+ }
+
+static int lexGeta()
+ {
+ ++lexBuf.len;
+ return (lexBuf.buf[lexBuf.getPtr] = lexGetc_());
+ }
+
+static int lexGeta_(int i)
+ {
+ ++lexBuf.len;
+ return (lexBuf.buf[(lexBuf.getPtr+i)%MAX_LEX_LOOKAHEAD] = lexGetc_());
+ }
+
+static void lexSkipLookahead() {
+ if (lexBuf.len > 0 && lexBuf.buf[lexBuf.getPtr]!=EOF) {
+ /* don't skip EOF. */
+ lexBuf.getPtr = (lexBuf.getPtr + 1) % MAX_LEX_LOOKAHEAD;
+ lexBuf.len--;
+ }
+ }
+
+static int lexLookahead() {
+ int c = (lexBuf.len)?
+ lexBuf.buf[lexBuf.getPtr]:
+ lexGeta();
+ /* do the \r\n -> \n or \r -> \n translation here */
+ if (c == '\r') {
+ int a = (lexBuf.len>1)?
+ lexBuf.buf[(lexBuf.getPtr+1)%MAX_LEX_LOOKAHEAD]:
+ lexGeta_(1);
+ if (a == '\n') {
+ lexSkipLookahead();
+ }
+ lexBuf.buf[lexBuf.getPtr] = c = '\n';
+ }
+ else if (c == '\n') {
+ int a;
+ if (lexBuf.len > 1)
+ a = lexBuf.buf[lexBuf.getPtr];
+ else
+ a = lexGeta_(1);
+ if (a == '\r') {
+ lexSkipLookahead();
+ }
+ lexBuf.buf[lexBuf.getPtr] = '\n';
+ }
+ return c;
+ }
+
+static int lexGetc() {
+ int c = lexLookahead();
+ if (lexBuf.len > 0 && lexBuf.buf[lexBuf.getPtr]!=EOF) {
+ /* EOF will remain in lookahead buffer */
+ lexBuf.getPtr = (lexBuf.getPtr + 1) % MAX_LEX_LOOKAHEAD;
+ lexBuf.len--;
+ }
+ return c;
+ }
+
+static void lexSkipLookaheadWord() {
+ if (lexBuf.strsLen <= lexBuf.len) {
+ lexBuf.len -= lexBuf.strsLen;
+ lexBuf.getPtr = (lexBuf.getPtr + lexBuf.strsLen) % MAX_LEX_LOOKAHEAD;
+ }
+ }
+
+static void lexClearToken()
+ {
+ lexBuf.strsLen = 0;
+ }
+
+static void lexAppendc(int c)
+ {
+ /* not sure if I am doing this right to fix purify report -- PGB */
+ lexBuf.strs = (char *) realloc(lexBuf.strs, (size_t) lexBuf.strsLen + 1);
+ lexBuf.strs[lexBuf.strsLen] = c;
+ /* append up to zero termination */
+ if (c == 0) return;
+ lexBuf.strsLen++;
+ if (lexBuf.strsLen > lexBuf.maxToken) {
+ /* double the token string size */
+ lexBuf.maxToken <<= 1;
+ lexBuf.strs = (char*) realloc(lexBuf.strs,(size_t)lexBuf.maxToken);
+ }
+ }
+
+static char* lexStr() {
+ return dupStr(lexBuf.strs,(size_t)lexBuf.strsLen+1);
+ }
+
+static void lexSkipWhite() {
+ int c = lexLookahead();
+ while (c == ' ' || c == '\t') {
+ lexSkipLookahead();
+ c = lexLookahead();
+ }
+ }
+
+static char* lexGetWord() {
+ int c;
+ lexSkipWhite();
+ lexClearToken();
+ c = lexLookahead();
+ /* some "words" have a space in them, like "NEEDS ACTION".
+ this may be an oversight of the spec, but it is true nevertheless.
+ while (c != EOF && !strchr("\t\n ;:=",c)) { */
+ while (c != EOF && !strchr("\n;:=",c)) {
+ lexAppendc(c);
+ lexSkipLookahead();
+ c = lexLookahead();
+ }
+ lexAppendc(0);
+ return lexStr();
+ }
+
+void lexPushLookahead(char *s, int len) {
+ int putptr;
+ if (len == 0) len = strlen(s);
+ putptr = (int)lexBuf.getPtr - len;
+ /* this function assumes that length of word to push back
+ / is not greater than MAX_LEX_LOOKAHEAD.
+ */
+ if (putptr < 0) putptr += MAX_LEX_LOOKAHEAD;
+ lexBuf.getPtr = putptr;
+ while (*s) {
+ lexBuf.buf[putptr] = *s++;
+ putptr = (putptr + 1) % MAX_LEX_LOOKAHEAD;
+ }
+ lexBuf.len += len;
+ }
+
+static void lexPushLookaheadc(int c) {
+ int putptr;
+ /* can't putback EOF, because it never leaves lookahead buffer */
+ if (c == EOF) return;
+ putptr = (int)lexBuf.getPtr - 1;
+ if (putptr < 0) putptr += MAX_LEX_LOOKAHEAD;
+ lexBuf.getPtr = putptr;
+ lexBuf.buf[putptr] = c;
+ lexBuf.len += 1;
+ }
+
+static char* lexLookaheadWord() {
+ /* this function can lookahead word with max size of MAX_LEX_LOOKAHEAD_0
+ / and thing bigger than that will stop the lookahead and return 0;
+ / leading white spaces are not recoverable.
+ */
+ int c;
+ int len = 0;
+ int curgetptr = 0;
+ lexSkipWhite();
+ lexClearToken();
+ curgetptr = (int)lexBuf.getPtr; /* remember! */
+ while (len < (MAX_LEX_LOOKAHEAD_0)) {
+ c = lexGetc();
+ len++;
+ if (c == EOF || strchr("\t\n ;:=", c)) {
+ lexAppendc(0);
+ /* restore lookahead buf. */
+ lexBuf.len += len;
+ lexBuf.getPtr = curgetptr;
+ return lexStr();
+ }
+ else
+ lexAppendc(c);
+ }
+ lexBuf.len += len; /* char that has been moved to lookahead buffer */
+ lexBuf.getPtr = curgetptr;
+ return 0;
+ }
+
+#ifdef _SUPPORT_LINE_FOLDING
+static void handleMoreRFC822LineBreak(int c) {
+ /* suport RFC 822 line break in cases like
+ * ADR: foo;
+ * morefoo;
+ * more foo;
+ */
+ if (c == ';') {
+ int a;
+ lexSkipLookahead();
+ /* skip white spaces */
+ a = lexLookahead();
+ while (a == ' ' || a == '\t') {
+ lexSkipLookahead();
+ a = lexLookahead();
+ }
+ if (a == '\n') {
+ lexSkipLookahead();
+ a = lexLookahead();
+ if (a == ' ' || a == '\t') {
+ /* continuation, throw away all the \n and spaces read so
+ * far
+ */
+ lexSkipWhite();
+ lexPushLookaheadc(';');
+ }
+ else {
+ lexPushLookaheadc('\n');
+ lexPushLookaheadc(';');
+ }
+ }
+ else {
+ lexPushLookaheadc(';');
+ }
+ }
+ }
+
+static char* lexGet1Value() {
+ int c;
+ lexSkipWhite();
+ c = lexLookahead();
+ lexClearToken();
+ while (c != EOF && c != ';') {
+ if (c == '\n') {
+ int a;
+ lexSkipLookahead();
+ a = lexLookahead();
+ if (a == ' ' || a == '\t') {
+ lexAppendc(' ');
+ lexSkipLookahead();
+ }
+ else {
+ lexPushLookaheadc('\n');
+ break;
+ }
+ }
+ else {
+ lexAppendc(c);
+ lexSkipLookahead();
+ }
+ c = lexLookahead();
+ }
+ lexAppendc(0);
+ handleMoreRFC822LineBreak(c);
+ return c==EOF?0:lexStr();
+ }
+#endif
+
+char* lexGetStrUntil(char *termset) {
+ int c = lexLookahead();
+ lexClearToken();
+ while (c != EOF && !strchr(termset,c)) {
+ lexAppendc(c);
+ lexSkipLookahead();
+ c = lexLookahead();
+ }
+ lexAppendc(0);
+ return c==EOF?0:lexStr();
+ }
+
+static int match_begin_name(int end) {
+ char *n = lexLookaheadWord();
+ int token = ID;
+ if (n) {
+ if (!strcasecmp(n,"vcard")) token = end?END_VCARD:BEGIN_VCARD;
+ else if (!strcasecmp(n,"vcalendar")) token = end?END_VCAL:BEGIN_VCAL;
+ else if (!strcasecmp(n,"vevent")) token = end?END_VEVENT:BEGIN_VEVENT;
+ else if (!strcasecmp(n,"vtodo")) token = end?END_VTODO:BEGIN_VTODO;
+ deleteStr(n);
+ return token;
+ }
+ return 0;
+ }
+
+
+void initLex(const char *inputstring, unsigned long inputlen, FILE *inputfile)
+ {
+ /* initialize lex mode stack */
+ lexBuf.lexModeStack[lexBuf.lexModeStackTop=0] = L_NORMAL;
+
+ /* iniatialize lex buffer. */
+ lexBuf.inputString = (char*) inputstring;
+ lexBuf.inputLen = inputlen;
+ lexBuf.curPos = 0;
+ lexBuf.inputFile = inputfile;
+
+ lexBuf.len = 0;
+ lexBuf.getPtr = 0;
+
+ lexBuf.maxToken = MAXTOKEN;
+ lexBuf.strs = (char*)malloc(MAXTOKEN);
+ lexBuf.strsLen = 0;
+
+ }
+
+static void finiLex() {
+ free(lexBuf.strs);
+ }
+
+
+/****************************************************************************/
+/* This parses and converts the base64 format for binary encoding into
+ * a decoded buffer (allocated with new). See RFC 1521.
+ */
+static char * lexGetDataFromBase64()
+ {
+ unsigned long bytesLen = 0, bytesMax = 0;
+ int quadIx = 0, pad = 0;
+ unsigned long trip = 0;
+ unsigned char b;
+ int c;
+ unsigned char *bytes = NULL;
+ unsigned char *oldBytes = NULL;
+
+ DBG_(("db: lexGetDataFromBase64\n"));
+ while (1) {
+ c = lexGetc();
+ if (c == '\n') {
+ ++mime_lineNum;
+ if (lexLookahead() == '\n') {
+ /* a '\n' character by itself means end of data */
+ break;
+ }
+ else continue; /* ignore '\n' */
+ }
+ else {
+ if ((c >= 'A') && (c <= 'Z'))
+ b = (unsigned char)(c - 'A');
+ else if ((c >= 'a') && (c <= 'z'))
+ b = (unsigned char)(c - 'a') + 26;
+ else if ((c >= '0') && (c <= '9'))
+ b = (unsigned char)(c - '0') + 52;
+ else if (c == '+')
+ b = 62;
+ else if (c == '/')
+ b = 63;
+ else if (c == '=') {
+ b = 0;
+ pad++;
+ } else if ((c == ' ') || (c == '\t')) {
+ continue;
+ } else { /* error condition */
+ if (bytes) free(bytes);
+ else if (oldBytes) free(oldBytes);
+ /* error recovery: skip until 2 adjacent newlines. */
+ DBG_(("db: invalid character 0x%x '%c'\n", c,c));
+ if (c != EOF) {
+ c = lexGetc();
+ while (c != EOF) {
+ if (c == '\n' && lexLookahead() == '\n') {
+ ++mime_lineNum;
+ break;
+ }
+ c = lexGetc();
+ }
+ }
+ return NULL;
+ }
+ trip = (trip << 6) | b;
+ if (++quadIx == 4) {
+ unsigned char outBytes[3];
+ int numOut;
+ int i;
+ for (i = 0; i < 3; i++) {
+ outBytes[2-i] = (unsigned char)(trip & 0xFF);
+ trip >>= 8;
+ }
+ numOut = 3 - pad;
+ if (bytesLen + numOut > bytesMax) {
+ if (!bytes) {
+ bytesMax = 1024;
+ bytes = (unsigned char*)malloc((size_t)bytesMax);
+ }
+ else {
+ bytesMax <<= 2;
+ oldBytes = bytes;
+ bytes = (unsigned char*)realloc(bytes,(size_t)bytesMax);
+ }
+ if (bytes == 0) {
+ mime_error("out of memory while processing BASE64 data\n");
+ }
+ }
+ if (bytes) {
+ memcpy(bytes + bytesLen, outBytes, numOut);
+ bytesLen += numOut;
+ }
+ trip = 0;
+ quadIx = 0;
+ }
+ }
+ } /* while */
+ DBG_(("db: bytesLen = %d\n", bytesLen));
+ /* kludge: all this won't be necessary if we have tree form
+ representation */
+ if (bytes) {
+ setValueWithSize(curProp,bytes,(unsigned int)bytesLen);
+ free(bytes);
+ }
+ else if (oldBytes) {
+ setValueWithSize(curProp,oldBytes,(unsigned int)bytesLen);
+ free(oldBytes);
+ }
+ return 0;
+ }
+
+static int match_begin_end_name(int end) {
+ int token;
+ lexSkipWhite();
+ if (lexLookahead() != ':') return ID;
+ lexSkipLookahead();
+ lexSkipWhite();
+ token = match_begin_name(end);
+ if (token == ID) {
+ lexPushLookaheadc(':');
+ DBG_(("db: ID '%s'\n", yylval.str));
+ return ID;
+ }
+ else if (token != 0) {
+ lexSkipLookaheadWord();
+ deleteStr(yylval.str);
+ DBG_(("db: begin/end %d\n", token));
+ return token;
+ }
+ return 0;
+ }
+
+static char* lexGetQuotedPrintable()
+ {
+ char cur;
+
+ lexClearToken();
+ do {
+ cur = lexGetc();
+ switch (cur) {
+ case '=': {
+ int c = 0;
+ int next[2];
+ int i;
+ for (i = 0; i < 2; i++) {
+ next[i] = lexGetc();
+ if (next[i] >= '0' && next[i] <= '9')
+ c = c * 16 + next[i] - '0';
+ else if (next[i] >= 'A' && next[i] <= 'F')
+ c = c * 16 + next[i] - 'A' + 10;
+ else
+ break;
+ }
+ if (i == 0) {
+ /* single '=' follow by LINESEP is continuation sign? */
+ if (next[0] == '\n') {
+ ++mime_lineNum;
+ }
+ else {
+ lexPushLookaheadc('=');
+ goto EndString;
+ }
+ }
+ else if (i == 1) {
+ lexPushLookaheadc(next[1]);
+ lexPushLookaheadc(next[0]);
+ lexAppendc('=');
+ } else {
+ lexAppendc(c);
+ }
+ break;
+ } /* '=' */
+ case '\n': {
+ lexPushLookaheadc('\n');
+ goto EndString;
+ }
+ case (char)EOF:
+ break;
+ default:
+ lexAppendc(cur);
+ break;
+ } /* switch */
+ } while (cur != (char)EOF);
+
+EndString:
+ lexAppendc(0);
+ return lexStr();
+ } /* LexQuotedPrintable */
+
+static int yylex() {
+
+ int lexmode = LEXMODE();
+ if (lexmode == L_VALUES) {
+ int c = lexGetc();
+ if (c == ';') {
+ DBG_(("db: SEMICOLON\n"));
+ lexPushLookaheadc(c);
+ handleMoreRFC822LineBreak(c);
+ lexSkipLookahead();
+ return SEMICOLON;
+ }
+ else if (strchr("\n",c)) {
+ ++mime_lineNum;
+ /* consume all line separator(s) adjacent to each other */
+ c = lexLookahead();
+ while (strchr("\n",c)) {
+ lexSkipLookahead();
+ c = lexLookahead();
+ ++mime_lineNum;
+ }
+ DBG_(("db: LINESEP\n"));
+ return LINESEP;
+ }
+ else {
+ char *p = 0;
+ lexPushLookaheadc(c);
+ if (lexWithinMode(L_BASE64)) {
+ /* get each char and convert to bin on the fly... */
+ p = lexGetDataFromBase64();
+ yylval.str = p;
+ return STRING;
+ }
+ else if (lexWithinMode(L_QUOTED_PRINTABLE)) {
+ p = lexGetQuotedPrintable();
+ }
+ else {
+#ifdef _SUPPORT_LINE_FOLDING
+ p = lexGet1Value();
+#else
+ p = lexGetStrUntil(";\n");
+#endif
+ }
+ if (p) {
+ DBG_(("db: STRING: '%s'\n", p));
+ yylval.str = p;
+ return STRING;
+ }
+ else return 0;
+ }
+ }
+ else {
+ /* normal mode */
+ while (1) {
+ int c = lexGetc();
+ switch(c) {
+ case ':': {
+ /* consume all line separator(s) adjacent to each other */
+ /* ignoring linesep immediately after colon. */
+ c = lexLookahead();
+ while (strchr("\n",c)) {
+ lexSkipLookahead();
+ c = lexLookahead();
+ ++mime_lineNum;
+ }
+ DBG_(("db: COLON\n"));
+ return COLON;
+ }
+ case ';':
+ DBG_(("db: SEMICOLON\n"));
+ return SEMICOLON;
+ case '=':
+ DBG_(("db: EQ\n"));
+ return EQ;
+ /* ignore tabs/newlines in this mode. We can't ignore
+ * spaces, because values like NEEDS ACTION have a space. */
+ case '\t': continue;
+ case '\n': {
+ ++mime_lineNum;
+ continue;
+ }
+ case EOF: return 0;
+ break;
+ default: {
+ lexPushLookaheadc(c);
+ if (isalpha(c) || c == ' ') {
+ char *t = lexGetWord();
+ yylval.str = t;
+ if (!strcasecmp(t, "begin")) {
+ return match_begin_end_name(0);
+ }
+ else if (!strcasecmp(t,"end")) {
+ return match_begin_end_name(1);
+ }
+ else {
+ DBG_(("db: ID '%s'\n", t));
+ return ID;
+ }
+ }
+ else {
+ /* unknown token */
+ return 0;
+ }
+ break;
+ }
+ }
+ }
+ }
+ return 0;
+ }
+
+
+/***************************************************************************/
+/*** Public Functions ****/
+/***************************************************************************/
+
+static VObject* Parse_MIMEHelper()
+ {
+ ObjStackTop = -1;
+ mime_numErrors = 0;
+ mime_lineNum = 1;
+ vObjList = 0;
+ curObj = 0;
+
+ if (yyparse() != 0)
+ return 0;
+
+ finiLex();
+ return vObjList;
+ }
+
+/****************************************************************************/
+VObject* Parse_MIME(const char *input, unsigned long len)
+ {
+ initLex(input, len, 0);
+ return Parse_MIMEHelper();
+ }
+
+
+VObject* Parse_MIME_FromFile(FILE *file)
+ {
+ VObject *result;
+ long startPos;
+
+ initLex(0,(unsigned long)-1,file);
+ startPos = ftell(file);
+ if (!(result = Parse_MIMEHelper())) {
+ fseek(file,startPos,SEEK_SET);
+ }
+ return result;
+ }
+
+VObject* Parse_MIME_FromFileName(const char *fname)
+ {
+ FILE *fp = fopen(fname,"r");
+ if (fp) {
+ VObject* o = Parse_MIME_FromFile(fp);
+ fclose(fp);
+ return o;
+ }
+ else {
+ char msg[255];
+ sprintf(msg, "can't open file '%s' for reading\n", fname);
+ mime_error_(msg);
+ return 0;
+ }
+ }
+
+/****************************************************************************/
+void YYDebug(const char *s)
+{
+ Parse_Debug(s);
+}
+
+
+static MimeErrorHandler mimeErrorHandler;
+
+void registerMimeErrorHandler(MimeErrorHandler me)
+ {
+ mimeErrorHandler = me;
+ }
+
+static void mime_error(char *s)
+ {
+ char msg[256];
+ if (mimeErrorHandler) {
+ sprintf(msg,"%s at line %d", s, mime_lineNum);
+ mimeErrorHandler(msg);
+ }
+ }
+
+static void mime_error_(char *s)
+ {
+ if (mimeErrorHandler) {
+ mimeErrorHandler(s);
+ }
+ }
+