diff options
Diffstat (limited to 'WWW/Library/Implementation/HTUU.c')
-rw-r--r-- | WWW/Library/Implementation/HTUU.c | 210 |
1 files changed, 210 insertions, 0 deletions
diff --git a/WWW/Library/Implementation/HTUU.c b/WWW/Library/Implementation/HTUU.c new file mode 100644 index 00000000..4848cce5 --- /dev/null +++ b/WWW/Library/Implementation/HTUU.c @@ -0,0 +1,210 @@ +/* + * $LynxId: HTUU.c,v 1.11 2010/09/21 23:55:12 tom Exp $ + * + * MODULE HTUU.c + * UUENCODE AND UUDECODE + * + * ACKNOWLEDGEMENT: + * This code is taken from rpem distribution, and was originally + * written by Mark Riordan. + * + * AUTHORS: + * MR Mark Riordan riordanmr@clvax1.cl.msu.edu + * AL Ari Luotonen luotonen@dxcern.cern.ch + * + * HISTORY: + * Added as part of the WWW library and edited to conform + * with the WWW project coding standards by: AL 5 Aug 1993 + * Originally written by: MR 12 Aug 1990 + * Original header text: + * ------------------------------------------------------------- + * File containing routines to convert a buffer + * of bytes to/from RFC 1113 printable encoding format. + * + * This technique is similar to the familiar Unix uuencode + * format in that it maps 6 binary bits to one ASCII + * character (or more aptly, 3 binary bytes to 4 ASCII + * characters). However, RFC 1113 does not use the same + * mapping to printable characters as uuencode. + * + * Mark Riordan 12 August 1990 and 17 Feb 1991. + * This code is hereby placed in the public domain. + * ------------------------------------------------------------- + */ + +#include <HTUtils.h> +#include <HTUU.h> + +#include <LYLeaks.h> + +static char six2pr[64] = +{ + 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', + 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', + 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', + 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z', + '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', '+', '/' +}; + +static unsigned char pr2six[256]; + +/*--- function HTUU_encode ----------------------------------------------- + * + * Encode a single line of binary data to a standard format that + * uses only printing ASCII characters (but takes up 33% more bytes). + * + * Entry bufin points to a buffer of bytes. If nbytes is not + * a multiple of three, then the byte just beyond + * the last byte in the buffer must be 0. + * nbytes is the number of bytes in that buffer. + * This cannot be more than 48. + * bufcoded points to an output buffer. Be sure that this + * can hold at least 1 + (4*nbytes)/3 characters. + * + * Exit bufcoded contains the coded line. The first 4*nbytes/3 bytes + * contain printing ASCII characters representing + * those binary bytes. This may include one or + * two '=' characters used as padding at the end. + * The last byte is a zero byte. + * Returns the number of ASCII characters in "bufcoded". + */ +int HTUU_encode(unsigned char *bufin, + size_t nbytes, + char *bufcoded) +{ +/* ENC is the basic 1 character encoding function to make a char printing */ +#define ENC(c) six2pr[c] + + register char *outptr = bufcoded; + size_t i; + + /* This doesn't seem to be needed (AL): register unsigned char *inptr = bufin; */ + + for (i = 0; i < nbytes; i += 3) { + *(outptr++) = ENC(*bufin >> 2); /* c1 */ + *(outptr++) = ENC(((*bufin << 4) & 060) | ((bufin[1] >> 4) & 017)); /*c2 */ + *(outptr++) = ENC(((bufin[1] << 2) & 074) | ((bufin[2] >> 6) & 03)); /*c3 */ + *(outptr++) = ENC(bufin[2] & 077); /* c4 */ + + bufin += 3; + } + + /* If nbytes was not a multiple of 3, then we have encoded too + * many characters. Adjust appropriately. + */ + if (i == nbytes + 1) { + /* There were only 2 bytes in that last group */ + outptr[-1] = '='; + } else if (i == nbytes + 2) { + /* There was only 1 byte in that last group */ + outptr[-1] = '='; + outptr[-2] = '='; + } + *outptr = '\0'; + return (int) (outptr - bufcoded); +} + +/*--- function HTUU_decode ------------------------------------------------ + * + * Decode an ASCII-encoded buffer back to its original binary form. + * + * Entry bufcoded points to a uuencoded string. It is + * terminated by any character not in + * the printable character table six2pr, but + * leading whitespace is stripped. + * bufplain points to the output buffer; must be big + * enough to hold the decoded string (generally + * shorter than the encoded string) plus + * as many as two extra bytes used during + * the decoding process. + * outbufsize is the maximum number of bytes that + * can fit in bufplain. + * + * Exit Returns the number of binary bytes decoded. + * bufplain contains these bytes. + */ +int HTUU_decode(char *bufcoded, + unsigned char *bufplain, + int outbufsize) +{ +/* single character decode */ +#define DEC(c) pr2six[(int)c] +#define MAXVAL 63 + + static int first = 1; + + int nbytesdecoded, j; + register char *bufin; + register unsigned char *bufout = bufplain; + register int nprbytes; + + /* If this is the first call, initialize the mapping table. + * This code should work even on non-ASCII machines. + */ + if (first) { + first = 0; + for (j = 0; j < 256; j++) + pr2six[j] = MAXVAL + 1; + + for (j = 0; j < 64; j++) + pr2six[UCH(six2pr[j])] = UCH(j); +#if 0 + /* *INDENT-OFF* */ + pr2six['A']= 0; pr2six['B']= 1; pr2six['C']= 2; pr2six['D']= 3; + pr2six['E']= 4; pr2six['F']= 5; pr2six['G']= 6; pr2six['H']= 7; + pr2six['I']= 8; pr2six['J']= 9; pr2six['K']=10; pr2six['L']=11; + pr2six['M']=12; pr2six['N']=13; pr2six['O']=14; pr2six['P']=15; + pr2six['Q']=16; pr2six['R']=17; pr2six['S']=18; pr2six['T']=19; + pr2six['U']=20; pr2six['V']=21; pr2six['W']=22; pr2six['X']=23; + pr2six['Y']=24; pr2six['Z']=25; pr2six['a']=26; pr2six['b']=27; + pr2six['c']=28; pr2six['d']=29; pr2six['e']=30; pr2six['f']=31; + pr2six['g']=32; pr2six['h']=33; pr2six['i']=34; pr2six['j']=35; + pr2six['k']=36; pr2six['l']=37; pr2six['m']=38; pr2six['n']=39; + pr2six['o']=40; pr2six['p']=41; pr2six['q']=42; pr2six['r']=43; + pr2six['s']=44; pr2six['t']=45; pr2six['u']=46; pr2six['v']=47; + pr2six['w']=48; pr2six['x']=49; pr2six['y']=50; pr2six['z']=51; + pr2six['0']=52; pr2six['1']=53; pr2six['2']=54; pr2six['3']=55; + pr2six['4']=56; pr2six['5']=57; pr2six['6']=58; pr2six['7']=59; + pr2six['8']=60; pr2six['9']=61; pr2six['+']=62; pr2six['/']=63; + /* *INDENT-ON* */ + +#endif + } + + /* Strip leading whitespace. */ + + while (*bufcoded == ' ' || *bufcoded == '\t') + bufcoded++; + + /* Figure out how many characters are in the input buffer. + * If this would decode into more bytes than would fit into + * the output buffer, adjust the number of input bytes downwards. + */ + bufin = bufcoded; + while (pr2six[UCH(*(bufin++))] <= MAXVAL) ; + nprbytes = (int) (bufin - bufcoded - 1); + nbytesdecoded = ((nprbytes + 3) / 4) * 3; + if (nbytesdecoded > outbufsize) { + nprbytes = (outbufsize * 4) / 3; + } + + bufin = bufcoded; + + while (nprbytes > 0) { + *(bufout++) = UCH((DEC(bufin[0]) << 2) | (DEC(bufin[1]) >> 4)); + *(bufout++) = UCH((DEC(bufin[1]) << 4) | (DEC(bufin[2]) >> 2)); + *(bufout++) = UCH((DEC(bufin[2]) << 6) | (DEC(bufin[3]))); + bufin += 4; + nprbytes -= 4; + } + + if (nprbytes & 03) { + if (pr2six[(int) bufin[-2]] > MAXVAL) { + nbytesdecoded -= 2; + } else { + nbytesdecoded -= 1; + } + } + + return (nbytesdecoded); +} |