| /* | |
| * Copyright (c) 2004-2005 Sergey Lyubka <valenok@gmail.com> | |
| * All rights reserved | |
| * | |
| * "THE BEER-WARE LICENSE" (Revision 42): | |
| * Sergey Lyubka wrote this file. As long as you retain this notice you | |
| * can do whatever you want with this stuff. If we meet some day, and you think | |
| * this stuff is worth it, you can buy me a beer in return. | |
| */ | |
| /* | |
| * Downloaded Sat Nov 5 17:42:08 CET 2011 at | |
| * http://slre.sourceforge.net/1.0/slre.h | |
| */ | |
| /* | |
| * This is a regular expression library that implements a subset of Perl RE. | |
| * Please refer to http://slre.sourceforge.net for detailed description. | |
| * | |
| * Usage example (parsing HTTP request): | |
| * | |
| * struct slre slre; | |
| * struct cap captures[4 + 1]; // Number of braket pairs + 1 | |
| * ... | |
| * | |
| * slre_compile(&slre,"^(GET|POST) (\S+) HTTP/(\S+?)\r\n"); | |
| * | |
| * if (slre_match(&slre, buf, len, captures)) { | |
| * printf("Request line length: %d\n", captures[0].len); | |
| * printf("Method: %.*s\n", captures[1].len, captures[1].ptr); | |
| * printf("URI: %.*s\n", captures[2].len, captures[2].ptr); | |
| * } | |
| * | |
| * Supported syntax: | |
| * ^ Match beginning of a buffer | |
| * $ Match end of a buffer | |
| * () Grouping and substring capturing | |
| * [...] Match any character from set | |
| * [^...] Match any character but ones from set | |
| * \s Match whitespace | |
| * \S Match non-whitespace | |
| * \d Match decimal digit | |
| * \r Match carriage return | |
| * \n Match newline | |
| * + Match one or more times (greedy) | |
| * +? Match one or more times (non-greedy) | |
| * * Match zero or more times (greedy) | |
| * *? Match zero or more times (non-greedy) | |
| * ? Match zero or once | |
| * \xDD Match byte with hex value 0xDD | |
| * \meta Match one of the meta character: ^$().[*+?\ | |
| */ | |
| #ifndef SLRE_HEADER_DEFINED | |
| #define SLRE_HEADER_DEFINED | |
| /* | |
| * Compiled regular expression | |
| */ | |
| struct slre { | |
| unsigned char code[256]; | |
| unsigned char data[256]; | |
| int code_size; | |
| int data_size; | |
| int num_caps; /* Number of bracket pairs */ | |
| const char *err_str; /* Error string */ | |
| }; | |
| /* | |
| * Captured substring | |
| */ | |
| struct cap { | |
| const char *ptr; /* Pointer to the substring */ | |
| int len; /* Substring length */ | |
| }; | |
| /* | |
| * Compile regular expression. If success, 1 is returned. | |
| * If error, 0 is returned and slre.err_str points to the error message. | |
| */ | |
| int slre_compile(struct slre *, const char *re); | |
| /* | |
| * Return 1 if match, 0 if no match. | |
| * If `captured_substrings' array is not NULL, then it is filled with the | |
| * values of captured substrings. captured_substrings[0] element is always | |
| * a full matched substring. The round bracket captures start from | |
| * captured_substrings[1]. | |
| * It is assumed that the size of captured_substrings array is enough to | |
| * hold all captures. The caller function must make sure it is! So, the | |
| * array_size = number_of_round_bracket_pairs + 1 | |
| */ | |
| int slre_match(const struct slre *, const char *buf, int buf_len, | |
| struct cap *captured_substrings); | |
| #ifdef SLRE_TEST | |
| void slre_dump(const struct slre *r, FILE *fp); | |
| #endif /* SLRE_TEST */ | |
| #endif /* SLRE_HEADER_DEFINED */ |