/************************************************************************** * * Copyright (C) 2002-2010, International Business Machines * Corporation and others. All Rights Reserved. * *************************************************************************** */ // // ugrep - an ICU sample program illustrating the use of ICU Regular Expressions. // // The use of the ICU Regex API all occurs within the main() // function. The rest of the code deals with with opening files, // encoding conversions, printing results, etc. // // This is not a full-featured grep program. The command line options // have been kept to a minimum to avoid complicating the sample code. // #include #include #include #include "unicode/utypes.h" #include "unicode/ustring.h" #include "unicode/regex.h" #include "unicode/ucnv.h" #include "unicode/uclean.h" // // The following variables contain paramters that may be set from the command line. // const char *pattern = NULL; // The regular expression int firstFileNum; // argv index of the first file name UBool displayFileName = FALSE; UBool displayLineNum = FALSE; // // Info regarding the file currently being processed // const char *fileName; int fileLen; // Length, in UTF-16 Code Units. UChar *ucharBuf = 0; // Buffer, holds converted file. (Simple minded program, always reads // the whole file at once. char *charBuf = 0; // Buffer, for original, unconverted file data. // // Info regarding the line currently being processed // int lineStart; // Index of first char of the current line in the file buffer int lineEnd; // Index of char following the new line sequence for the current line int lineNum; // // Converter, used on output to convert Unicode data back to char * // so that it will display in non-Unicode terminal windows. // UConverter *outConverter = 0; // // Function forward declarations // void processOptions(int argc, const char **argv); void nextLine(int start); void printMatch(); void printUsage(); void readFile(const char *name); //------------------------------------------------------------------------------------------ // // main for ugrep // // Structurally, all use of the ICU Regular Expression API is in main(), // and all of the supporting stuff necessary to make a running program, but // not directly related to regular expressions, is factored out into these other // functions. // //------------------------------------------------------------------------------------------ int main(int argc, const char** argv) { UBool matchFound = FALSE; // // Process the commmand line options. // processOptions(argc, argv); // // Create a RegexPattern object from the user supplied pattern string. // UErrorCode status = U_ZERO_ERROR; // All ICU operations report success or failure // in a status variable. UParseError parseErr; // In the event of a syntax error in the regex pattern, // this struct will contain the position of the // error. RegexPattern *rePat = RegexPattern::compile(pattern, parseErr, status); // Note that C++ is doing an automatic conversion // of the (char *) pattern to a temporary // UnicodeString object. if (U_FAILURE(status)) { fprintf(stderr, "ugrep: error in pattern: \"%s\" at position %d\n", u_errorName(status), parseErr.offset); exit(-1); } // // Create a RegexMatcher from the newly created pattern. // UnicodeString empty; RegexMatcher *matcher = rePat->matcher(empty, status); if (U_FAILURE(status)) { fprintf(stderr, "ugrep: error in creating RegexMatcher: \"%s\"\n", u_errorName(status)); exit(-1); } // // Loop, processing each of the input files. // for (int fileNum=firstFileNum; fileNum < argc; fileNum++) { readFile(argv[fileNum]); // // Loop through the lines of a file, trying to match the regex pattern on each. // for (nextLine(0); lineStartreset(s); if (matcher->find()) { matchFound = TRUE; printMatch(); } } } // // Clean up // delete matcher; delete rePat; free(ucharBuf); free(charBuf); ucnv_close(outConverter); u_cleanup(); // shut down ICU, release any cached data it owns. return matchFound? 0: 1; } //------------------------------------------------------------------------------------------ // // doOptions Run through the command line options, and set // the global variables accordingly. // // exit without returning if an error occured and // ugrep should not proceed further. // //------------------------------------------------------------------------------------------ void processOptions(int argc, const char **argv) { int optInd; UBool doUsage = FALSE; UBool doVersion = FALSE; const char *arg; for(optInd = 1; optInd < argc; ++optInd) { arg = argv[optInd]; /* version info */ if(strcmp(arg, "-V") == 0 || strcmp(arg, "--version") == 0) { doVersion = TRUE; } /* usage info */ else if(strcmp(arg, "--help") == 0) { doUsage = TRUE; } else if(strcmp(arg, "-n") == 0 || strcmp(arg, "--line-number") == 0) { displayLineNum = TRUE; } /* POSIX.1 says all arguments after -- are not options */ else if(strcmp(arg, "--") == 0) { /* skip the -- */ ++optInd; break; } /* unrecognized option */ else if(strncmp(arg, "-", strlen("-")) == 0) { printf("ugrep: invalid option -- %s\n", arg+1); doUsage = TRUE; } /* done with options */ else { break; } } if (doUsage) { printUsage(); exit(0); } if (doVersion) { printf("ugrep version 0.01\n"); if (optInd == argc) { exit(0); } } int remainingArgs = argc-optInd; // pattern file ... if (remainingArgs < 2) { fprintf(stderr, "ugrep: files or pattern are missing.\n"); printUsage(); exit(1); } if (remainingArgs > 2) { // More than one file to be processed. Display file names with match output. displayFileName = TRUE; } pattern = argv[optInd]; firstFileNum = optInd+1; } //------------------------------------------------------------------------------------------ // // printUsage // //------------------------------------------------------------------------------------------ void printUsage() { printf("ugrep [options] pattern file...\n" " -V or --version display version information\n" " --help display this help and exit\n" " -- stop further option processing\n" "-n, --line-number Prefix each line of output with the line number within its input file.\n" ); exit(0); } //------------------------------------------------------------------------------------------ // // readFile Read a file into memory, and convert it to Unicode. // // Since this is just a demo program, take the simple minded approach // of always reading the whole file at once. No intelligent buffering // is done. // //------------------------------------------------------------------------------------------ void readFile(const char *name) { // // Initialize global file variables // fileName = name; fileLen = 0; // zero length prevents processing in case of errors. // // Open the file and determine its size. // FILE *file = fopen(name, "rb"); if (file == 0 ) { fprintf(stderr, "ugrep: Could not open file \"%s\"\n", fileName); return; } fseek(file, 0, SEEK_END); int rawFileLen = ftell(file); fseek(file, 0, SEEK_SET); // // Read in the file // charBuf = (char *)realloc(charBuf, rawFileLen+1); // Need error checking... int t = fread(charBuf, 1, rawFileLen, file); if (t != rawFileLen) { fprintf(stderr, "Error reading file \"%s\"\n", fileName); fclose(file); return; } charBuf[rawFileLen]=0; fclose(file); // // Look for a Unicode Signature (BOM) in the data // int32_t signatureLength; const char * charDataStart = charBuf; UErrorCode status = U_ZERO_ERROR; const char* encoding = ucnv_detectUnicodeSignature( charDataStart, rawFileLen, &signatureLength, &status); if (U_FAILURE(status)) { fprintf(stderr, "ugrep: ICU Error \"%s\" from ucnv_detectUnicodeSignature()\n", u_errorName(status)); return; } if(encoding!=NULL ){ charDataStart += signatureLength; rawFileLen -= signatureLength; } // // Open a converter to take the file to UTF-16 // UConverter* conv; conv = ucnv_open(encoding, &status); if (U_FAILURE(status)) { fprintf(stderr, "ugrep: ICU Error \"%s\" from ucnv_open()\n", u_errorName(status)); return; } // // Convert the file data to UChar. // Preflight first to determine required buffer size. // uint32_t destCap = ucnv_toUChars(conv, NULL, // dest, 0, // destCapacity, charDataStart, rawFileLen, &status); if (status != U_BUFFER_OVERFLOW_ERROR) { fprintf(stderr, "ugrep: ucnv_toUChars: ICU Error \"%s\"\n", u_errorName(status)); return; }; status = U_ZERO_ERROR; ucharBuf = (UChar *)realloc(ucharBuf, (destCap+1) * sizeof(UChar)); ucnv_toUChars(conv, ucharBuf, // dest, destCap+1, charDataStart, rawFileLen, &status); if (U_FAILURE(status)) { fprintf(stderr, "ugrep: ucnv_toUChars: ICU Error \"%s\"\n", u_errorName(status)); return; }; ucnv_close(conv); // // Successful conversion. Set the global size variables so that // the rest of the processing will proceed for this file. // fileLen = destCap; } //------------------------------------------------------------------------------------------ // // nextLine Advance the line index variables, starting at the // specified position in the input file buffer, by // scanning forwrd until the next end-of-line. // // Need to take into account all of the possible Unicode // line ending sequences. // //------------------------------------------------------------------------------------------ void nextLine(int startPos) { if (startPos == 0) { lineNum = 0; } else { lineNum++; } lineStart = lineEnd = startPos; for (;;) { if (lineEnd >= fileLen) { return; } UChar c = ucharBuf[lineEnd]; lineEnd++; if (c == 0x0a || // Line Feed c == 0x0c || // Form Feed c == 0x0d || // Carriage Return c == 0x85 || // Next Line c == 0x2028 || // Line Separator c == 0x2029) // Paragraph separator { break; } } // Check for CR/LF sequence, and advance over the LF if we're in the middle of one. if (lineEnd < fileLen && ucharBuf[lineEnd-1] == 0x0d && ucharBuf[lineEnd] == 0x0a) { lineEnd++; } } //------------------------------------------------------------------------------------------ // // printMatch Called when a matching line has been located. // Print out the line from the file with the match, after // converting it back to the default code page. // //------------------------------------------------------------------------------------------ void printMatch() { char buf[2000]; UErrorCode status = U_ZERO_ERROR; // If we haven't already created a converter for output, do it now. if (outConverter == 0) { outConverter = ucnv_open(NULL, &status); if (U_FAILURE(status)) { fprintf(stderr, "ugrep: Error opening default converter: \"%s\"\n", u_errorName(status)); exit(-1); } }; // Convert the line to be printed back to the default 8 bit code page. // If the line is too long for our buffer, just truncate it. ucnv_fromUChars(outConverter, buf, // destination buffer for conversion sizeof(buf), // capacity of destination buffer &ucharBuf[lineStart], // Input to conversion lineEnd-lineStart, // number of UChars to convert &status); buf[sizeof(buf)-1] = 0; // Add null for use in case of too long lines. // The converter null-terminates its output unless // the buffer completely fills. if (displayFileName) { printf("%s:", fileName); } if (displayLineNum) { printf("%d:", lineNum); } printf("%s", buf); }