⭐ 欢迎来到虫虫下载站! | 📦 资源下载 📁 资源专辑 ℹ️ 关于我们
⭐ 虫虫下载站

📄 cook.c

📁 Linux下利用NTP协议的网络时间同步客户端
💻 C
字号:
/* *  $Id: cook.c,v 4.10 2007/02/04 17:44:12 bkorb Exp $ *  Time-stamp:      "2006-09-24 15:21:02 bkorb" * *  This file contains the routines that deal with processing quoted strings *  into an internal format. *//* *  Automated Options copyright 1992-2007 Bruce Korb * *  Automated Options is free software. *  You may redistribute it and/or modify it under the terms of the *  GNU General Public License, as published by the Free Software *  Foundation; either version 2, or (at your option) any later version. * *  Automated Options is distributed in the hope that it will be useful, *  but WITHOUT ANY WARRANTY; without even the implied warranty of *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the *  GNU General Public License for more details. * *  You should have received a copy of the GNU General Public License *  along with Automated Options.  See the file "COPYING".  If not, *  write to:  The Free Software Foundation, Inc., *             51 Franklin Street, Fifth Floor, *             Boston, MA  02110-1301, USA. * * As a special exception, Bruce Korb gives permission for additional * uses of the text contained in his release of AutoOpts. * * The exception is that, if you link the AutoOpts library with other * files to produce an executable, this does not by itself cause the * resulting executable to be covered by the GNU General Public License. * Your use of that executable is in no way restricted on account of * linking the AutoOpts library code into it. * * This exception does not however invalidate any other reasons why * the executable file might be covered by the GNU General Public License. * * This exception applies only to the code released by Bruce Korb under * the name AutoOpts.  If you copy code from other sources under the * General Public License into a copy of AutoOpts, as the General Public * License permits, the exception does not apply to the code that you add * in this way.  To avoid misleading anyone as to the status of such * modified files, you must delete this exception notice from them. * * If you write modifications of your own for AutoOpts, it is your choice * whether to permit this exception to apply to your modifications. * If you do not wish that, delete this exception notice. *//* = = = START-STATIC-FORWARD = = = *//* static forward declarations maintained by :mkfwd *//* = = = END-STATIC-FORWARD = = = *//*=export_func  ao_string_cook_escape_char * private: * * what:  escape-process a string fragment * arg:   + char const*  + pzScan  + points to character after the escape + * arg:   + char*        + pRes    + Where to put the result byte + * arg:   + unsigned int + nl_ch   + replacement char if scanned char is \n + * * ret-type: unsigned int * ret-desc: The number of bytes consumed processing the escaped character. * * doc: * *  This function converts "t" into "\t" and all your other favorite *  escapes, including numeric ones:  hex and ocatal, too. *  The returned result tells the caller how far to advance the *  scan pointer (passed in).  The default is to just pass through the *  escaped character and advance the scan by one. * *  Some applications need to keep an escaped newline, others need to *  suppress it.  This is accomplished by supplying a '\n' replacement *  character that is different from \n, if need be.  For example, use *  0x7F and never emit a 0x7F. * * err:  @code{NULL} is returned if the string is mal-formed.=*/unsigned intao_string_cook_escape_char( char const* pzIn, char* pRes, u_int nl ){    unsigned int  res = 1;    switch (*pRes = *pzIn++) {    case NUL:         /* NUL - end of input string */        return 0;    case '\r':        if (*pzIn != '\n')            return 1;        res++;        /* FALLTHROUGH */    case '\n':        /* NL  - emit newline        */        *pRes = (char)nl;        return res;    case 'a': *pRes = '\a'; break;    case 'b': *pRes = '\b'; break;    case 'f': *pRes = '\f'; break;    case 'n': *pRes = '\n'; break;    case 'r': *pRes = '\r'; break;    case 't': *pRes = '\t'; break;    case 'v': *pRes = '\v'; break;    case 'x':         /* HEX Escape       */        if (isxdigit( (int)*pzIn ))  {            unsigned int  val;            unsigned char ch = *pzIn++;            if ((ch >= 'A') && (ch <= 'F'))                val = 10 + (ch - 'A');            else if ((ch >= 'a') && (ch <= 'f'))                val = 10 + (ch - 'a');            else val = ch - '0';            ch = *pzIn;            if (! isxdigit( ch )) {                *pRes = val;                res   = 2;                break;            }            val <<= 4;            if ((ch >= 'A') && (ch <= 'F'))                val += 10 + (ch - 'A');            else if ((ch >= 'a') && (ch <= 'f'))                val += 10 + (ch - 'a');            else val += ch - '0';            res = 3;            *pRes = val;        }        break;    default:        /*         *  IF the character copied was an octal digit,         *  THEN set the output character to an octal value         */        if (isdigit( (int)*pRes ) && (*pRes < '8'))  {            unsigned int  val = *pRes - '0';            unsigned char ch  = *pzIn++;            /*             *  IF the second character is *not* an octal digit,             *  THEN save the value and bail             */            if ((ch < '0') || (ch > '7')) {                *pRes = val;                break;            }            val = (val<<3) + (ch - '0');            ch  = *pzIn;            res = 2;            /*             *  IF the THIRD character is *not* an octal digit,             *  THEN save the value and bail             */            if ((ch < '0') || (ch > '7')) {                *pRes = val;                break;            }            /*             *  IF the new value would not be too large,             *  THEN add on the third and last character value             */            if ((val<<3) < 0xFF) {                val = (val<<3) + (ch - '0');                res = 3;            }            *pRes = val;            break;        }    }    return res;}/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *  A quoted string has been found. *  Find the end of it and compress any escape sequences. *//*=export_func  ao_string_cook * private: * * what:  concatenate and escape-process strings * arg:   + char* + pzScan     + The *MODIFIABLE* input buffer + * arg:   + int*  + pLineCt    + The (possibly NULL) pointer to a line count + * * ret-type: char* * ret-desc: The address of the text following the processed strings. *           The return value is NULL if the strings are ill-formed. * * doc: * *  A series of one or more quoted strings are concatenated together. *  If they are quoted with double quotes (@code{"}), then backslash *  escapes are processed per the C programming language.  If they are *  single quote strings, then the backslashes are honored only when they *  precede another backslash or a single quote character. * * err:  @code{NULL} is returned if the string(s) is/are mal-formed.=*/char*ao_string_cook( char* pzScan, int* pLineCt ){    int   l = 0;    char  q = *pzScan;    /*     *  It is a quoted string.  Process the escape sequence characters     *  (in the set "abfnrtv") and make sure we find a closing quote.     */    char* pzD = pzScan++;    char* pzS = pzScan;    if (pLineCt == NULL)        pLineCt = &l;    for (;;) {        /*         *  IF the next character is the quote character, THEN we may end the         *  string.  We end it unless the next non-blank character *after* the         *  string happens to also be a quote.  If it is, then we will change         *  our quote character to the new quote character and continue         *  condensing text.         */        while (*pzS == q) {            *pzD = NUL; /* This is probably the end of the line */            pzS++;        scan_for_quote:            while (isspace((int)*pzS))                if (*(pzS++) == '\n')                    (*pLineCt)++;            /*             *  IF the next character is a quote character,             *  THEN we will concatenate the strings.             */            switch (*pzS) {            case '"':            case '\'':                break;            case '/':                /*                 *  Allow for a comment embedded in the concatenated string.                 */                switch (pzS[1]) {                default:  return NULL;                case '/':                    /*                     *  Skip to end of line                     */                    pzS = strchr( pzS, '\n' );                    if (pzS == NULL)                        return NULL;                    (*pLineCt)++;                    break;                case '*':                {                    char* p = strstr( pzS+2, "*/" );                    /*                     *  Skip to terminating star slash                     */                    if (p == NULL)                        return NULL;                    while (pzS < p) {                        if (*(pzS++) == '\n')                            (*pLineCt)++;                    }                    pzS = p + 2;                }                }                goto scan_for_quote;            default:                /*                 *  The next non-whitespace character is not a quote.                 *  The series of quoted strings has come to an end.                 */                return pzS;            }            q = *(pzS++);  /* assign new quote character and advance scan */        }        /*         *  We are inside a quoted string.  Copy text.         */        switch (*(pzD++) = *(pzS++)) {        case NUL:            return NULL;        case '\n':            (*pLineCt)++;            break;        case '\\':            /*             *  IF we are escaping a new line,             *  THEN drop both the escape and the newline from             *       the result string.             */            if (*pzS == '\n') {                pzS++;                pzD--;                (*pLineCt)++;            }            /*             *  ELSE IF the quote character is '"' or '`',             *  THEN we do the full escape character processing             */            else if (q != '\'') {                int ct = ao_string_cook_escape_char( pzS, pzD-1, (u_int)'\n' );                if (ct == 0)                    return NULL;                pzS += ct;            }     /* if (q != '\'')                  */            /*             *  OTHERWISE, we only process "\\", "\'" and "\#" sequences.             *  The latter only to easily hide preprocessing directives.             */            else switch (*pzS) {            case '\\':            case '\'':            case '#':                pzD[-1] = *pzS++;            }        }     /* switch (*(pzD++) = *(pzS++))    */    }         /* for (;;)                        */}/* * Local Variables: * mode: C * c-file-style: "stroustrup" * indent-tabs-mode: nil * End: * end of autoopts/cook.c */

⌨️ 快捷键说明

复制代码 Ctrl + C
搜索代码 Ctrl + F
全屏模式 F11
切换主题 Ctrl + Shift + D
显示快捷键 ?
增大字号 Ctrl + =
减小字号 Ctrl + -