/* * ***** BEGIN GPL LICENSE BLOCK ***** * * This program is free software; you can 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 * of the License, or (at your option) any later version. * * This program 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 this program; if not, write to the Free Software Foundation, * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. * * The Original Code is Copyright (C) 2001-2002 by NaN Holding BV. * All rights reserved. * * The Original Code is: all of this file. * * Contributor(s): none yet. * * ***** END GPL LICENSE BLOCK ***** * */ /** \file blender/blenlib/intern/string.c * \ingroup bli */ #include #include #include #include #include "MEM_guardedalloc.h" #include "BLI_dynstr.h" #include "BLI_string.h" #include "BLI_utildefines.h" #ifdef __GNUC__ # pragma GCC diagnostic error "-Wsign-conversion" #endif // #define DEBUG_STRSIZE /** * Duplicates the first \a len bytes of cstring \a str * into a newly mallocN'd string and returns it. \a str * is assumed to be at least len bytes long. * * \param str The string to be duplicated * \param len The number of bytes to duplicate * \retval Returns the duplicated string */ char *BLI_strdupn(const char *str, const size_t len) { char *n = MEM_mallocN(len + 1, "strdup"); memcpy(n, str, len); n[len] = '\0'; return n; } /** * Duplicates the cstring \a str into a newly mallocN'd * string and returns it. * * \param str The string to be duplicated * \retval Returns the duplicated string */ char *BLI_strdup(const char *str) { return BLI_strdupn(str, strlen(str)); } /** * Appends the two strings, and returns new mallocN'ed string * \param str1 first string for copy * \param str2 second string for append * \retval Returns dst */ char *BLI_strdupcat(const char *__restrict str1, const char *__restrict str2) { /* include the NULL terminator of str2 only */ const size_t str1_len = strlen(str1); const size_t str2_len = strlen(str2) + 1; char *str, *s; str = MEM_mallocN(str1_len + str2_len, "strdupcat"); s = str; memcpy(s, str1, str1_len); s += str1_len; memcpy(s, str2, str2_len); return str; } /** * Like strncpy but ensures dst is always * '\0' terminated. * * \param dst Destination for copy * \param src Source string to copy * \param maxncpy Maximum number of characters to copy (generally * the size of dst) * \retval Returns dst */ char *BLI_strncpy(char *__restrict dst, const char *__restrict src, const size_t maxncpy) { size_t srclen = BLI_strnlen(src, maxncpy - 1); BLI_assert(maxncpy != 0); #ifdef DEBUG_STRSIZE memset(dst, 0xff, sizeof(*dst) * maxncpy); #endif memcpy(dst, src, srclen); dst[srclen] = '\0'; return dst; } /** * Like BLI_strncpy but ensures dst is always padded by given char, on both sides (unless src is empty). * * \param dst Destination for copy * \param src Source string to copy * \param pad the char to use for padding * \param maxncpy Maximum number of characters to copy (generally the size of dst) * \retval Returns dst */ char *BLI_strncpy_ensure_pad(char *__restrict dst, const char *__restrict src, const char pad, size_t maxncpy) { BLI_assert(maxncpy != 0); #ifdef DEBUG_STRSIZE memset(dst, 0xff, sizeof(*dst) * maxncpy); #endif if (src[0] == '\0') { dst[0] = '\0'; } else { /* Add heading/trailing wildcards if needed. */ size_t idx = 0; size_t srclen; if (src[idx] != pad) { dst[idx++] = pad; maxncpy--; } maxncpy--; /* trailing '\0' */ srclen = BLI_strnlen(src, maxncpy); if ((src[srclen - 1] != pad) && (srclen == maxncpy)) { srclen--; } memcpy(&dst[idx], src, srclen); idx += srclen; if (dst[idx - 1] != pad) { dst[idx++] = pad; } dst[idx] = '\0'; } return dst; } /** * Like strncpy but ensures dst is always * '\0' terminated. * * \note This is a duplicate of #BLI_strncpy that returns bytes copied. * And is a drop in replacement for 'snprintf(str, sizeof(str), "%s", arg);' * * \param dst Destination for copy * \param src Source string to copy * \param maxncpy Maximum number of characters to copy (generally * the size of dst) * \retval The number of bytes copied (The only difference from BLI_strncpy). */ size_t BLI_strncpy_rlen(char *__restrict dst, const char *__restrict src, const size_t maxncpy) { size_t srclen = BLI_strnlen(src, maxncpy - 1); BLI_assert(maxncpy != 0); #ifdef DEBUG_STRSIZE memset(dst, 0xff, sizeof(*dst) * maxncpy); #endif memcpy(dst, src, srclen); dst[srclen] = '\0'; return srclen; } size_t BLI_strcpy_rlen(char *__restrict dst, const char *__restrict src) { size_t srclen = strlen(src); memcpy(dst, src, srclen + 1); return srclen; } /** * Portable replacement for #vsnprintf */ size_t BLI_vsnprintf(char *__restrict buffer, size_t maxncpy, const char *__restrict format, va_list arg) { size_t n; BLI_assert(buffer != NULL); BLI_assert(maxncpy > 0); BLI_assert(format != NULL); n = (size_t)vsnprintf(buffer, maxncpy, format, arg); if (n != -1 && n < maxncpy) { buffer[n] = '\0'; } else { buffer[maxncpy - 1] = '\0'; } return n; } /** * A version of #BLI_vsnprintf that returns ``strlen(buffer)`` */ size_t BLI_vsnprintf_rlen(char *__restrict buffer, size_t maxncpy, const char *__restrict format, va_list arg) { size_t n; BLI_assert(buffer != NULL); BLI_assert(maxncpy > 0); BLI_assert(format != NULL); n = (size_t)vsnprintf(buffer, maxncpy, format, arg); if (n != -1 && n < maxncpy) { /* pass */ } else { n = maxncpy - 1; } buffer[n] = '\0'; return n; } /** * Portable replacement for #snprintf */ size_t BLI_snprintf(char *__restrict dst, size_t maxncpy, const char *__restrict format, ...) { size_t n; va_list arg; #ifdef DEBUG_STRSIZE memset(dst, 0xff, sizeof(*dst) * maxncpy); #endif va_start(arg, format); n = BLI_vsnprintf(dst, maxncpy, format, arg); va_end(arg); return n; } /** * A version of #BLI_snprintf that returns ``strlen(dst)`` */ size_t BLI_snprintf_rlen(char *__restrict dst, size_t maxncpy, const char *__restrict format, ...) { size_t n; va_list arg; #ifdef DEBUG_STRSIZE memset(dst, 0xff, sizeof(*dst) * maxncpy); #endif va_start(arg, format); n = BLI_vsnprintf_rlen(dst, maxncpy, format, arg); va_end(arg); return n; } /** * Print formatted string into a newly #MEM_mallocN'd string * and return it. */ char *BLI_sprintfN(const char *__restrict format, ...) { DynStr *ds; va_list arg; char *n; va_start(arg, format); ds = BLI_dynstr_new(); BLI_dynstr_vappendf(ds, format, arg); n = BLI_dynstr_get_cstring(ds); BLI_dynstr_free(ds); va_end(arg); return n; } /* match pythons string escaping, assume double quotes - (") * TODO: should be used to create RNA animation paths. * TODO: support more fancy string escaping. current code is primitive * this basically is an ascii version of PyUnicode_EncodeUnicodeEscape() * which is a useful reference. */ size_t BLI_strescape(char *__restrict dst, const char *__restrict src, const size_t maxncpy) { size_t len = 0; BLI_assert(maxncpy != 0); while (len < maxncpy) { switch (*src) { case '\0': goto escape_finish; case '\\': case '"': ATTR_FALLTHROUGH; /* less common but should also be support */ case '\t': case '\n': case '\r': if (len + 1 < maxncpy) { *dst++ = '\\'; len++; } else { /* not enough space to escape */ break; } ATTR_FALLTHROUGH; default: *dst = *src; break; } dst++; src++; len++; } escape_finish: *dst = '\0'; return len; } /** * Makes a copy of the text within the "" that appear after some text 'blahblah' * i.e. for string 'pose["apples"]' with prefix 'pose[', it should grab "apples" * * - str: is the entire string to chop * - prefix: is the part of the string to leave out * * Assume that the strings returned must be freed afterwards, and that the inputs will contain * data we want... * * \return the offset and a length so as to avoid doing an allocation. */ char *BLI_str_quoted_substrN(const char *__restrict str, const char *__restrict prefix) { const char *startMatch, *endMatch; /* get the starting point (i.e. where prefix starts, and add prefixLen+1 to it to get be after the first " */ startMatch = strstr(str, prefix); if (startMatch) { const size_t prefixLen = strlen(prefix); startMatch += prefixLen + 1; /* get the end point (i.e. where the next occurance of " is after the starting point) */ endMatch = startMatch; while ((endMatch = strchr(endMatch, '"'))) { if (LIKELY(*(endMatch - 1) != '\\')) { break; } else { endMatch++; } } if (endMatch) { /* return the slice indicated */ return BLI_strdupn(startMatch, (size_t)(endMatch - startMatch)); } } return BLI_strdupn("", 0); } /** * string with all instances of substr_old replaced with substr_new, * Returns a copy of the cstring \a str into a newly mallocN'd * and returns it. * * \note A rather wasteful string-replacement utility, though this shall do for now... * Feel free to replace this with an even safe + nicer alternative * * \param str The string to replace occurrences of substr_old in * \param substr_old The text in the string to find and replace * \param substr_new The text in the string to find and replace * \retval Returns the duplicated string */ char *BLI_str_replaceN(const char *__restrict str, const char *__restrict substr_old, const char *__restrict substr_new) { DynStr *ds = NULL; size_t len_old = strlen(substr_old); const char *match; BLI_assert(substr_old[0] != '\0'); /* while we can still find a match for the old substring that we're searching for, * keep dicing and replacing */ while ((match = strstr(str, substr_old))) { /* the assembly buffer only gets created when we actually need to rebuild the string */ if (ds == NULL) ds = BLI_dynstr_new(); /* if the match position does not match the current position in the string, * copy the text up to this position and advance the current position in the string */ if (str != match) { /* add the segment of the string from str to match to the buffer, then restore the value at match */ BLI_dynstr_nappend(ds, str, (match - str)); /* now our current position should be set on the start of the match */ str = match; } /* add the replacement text to the accumulation buffer */ BLI_dynstr_append(ds, substr_new); /* advance the current position of the string up to the end of the replaced segment */ str += len_old; } /* finish off and return a new string that has had all occurrences of */ if (ds) { char *str_new; /* add what's left of the string to the assembly buffer * - we've been adjusting str to point at the end of the replaced segments */ BLI_dynstr_append(ds, str); /* convert to new c-string (MEM_malloc'd), and free the buffer */ str_new = BLI_dynstr_get_cstring(ds); BLI_dynstr_free(ds); return str_new; } else { /* just create a new copy of the entire string - we avoid going through the assembly buffer * for what should be a bit more efficiency... */ return BLI_strdup(str); } } /** * In-place replace every \a src to \a dst in \a str. * * \param str: The string to operate on. * \param src: The character to replace. * \param dst: The character to replace with. */ void BLI_str_replace_char(char *str, char src, char dst) { while (*str) { if (*str == src) { *str = dst; } str++; } } /** * Compare two strings without regard to case. * * \retval True if the strings are equal, false otherwise. */ int BLI_strcaseeq(const char *a, const char *b) { return (BLI_strcasecmp(a, b) == 0); } /** * Portable replacement for #strcasestr (not available in MSVC) */ char *BLI_strcasestr(const char *s, const char *find) { register char c, sc; register size_t len; if ((c = *find++) != 0) { c = tolower(c); len = strlen(find); do { do { if ((sc = *s++) == 0) return (NULL); sc = tolower(sc); } while (sc != c); } while (BLI_strncasecmp(s, find, len) != 0); s--; } return ((char *) s); } /** * Variation of #BLI_strcasestr with string length limited to \a len */ char *BLI_strncasestr(const char *s, const char *find, size_t len) { register char c, sc; if ((c = *find++) != 0) { c = tolower(c); if (len > 1) { do { do { if ((sc = *s++) == 0) return NULL; sc = tolower(sc); } while (sc != c); } while (BLI_strncasecmp(s, find, len - 1) != 0); } else { { do { if ((sc = *s++) == 0) return NULL; sc = tolower(sc); } while (sc != c); } } s--; } return ((char *)s); } int BLI_strcasecmp(const char *s1, const char *s2) { register int i; register char c1, c2; for (i = 0;; i++) { c1 = tolower(s1[i]); c2 = tolower(s2[i]); if (c1 < c2) { return -1; } else if (c1 > c2) { return 1; } else if (c1 == 0) { break; } } return 0; } int BLI_strncasecmp(const char *s1, const char *s2, size_t len) { register size_t i; register char c1, c2; for (i = 0; i < len; i++) { c1 = tolower(s1[i]); c2 = tolower(s2[i]); if (c1 < c2) { return -1; } else if (c1 > c2) { return 1; } else if (c1 == 0) { break; } } return 0; } /* compare number on the left size of the string */ static int left_number_strcmp(const char *s1, const char *s2, int *tiebreaker) { const char *p1 = s1, *p2 = s2; int numdigit, numzero1, numzero2; /* count and skip leading zeros */ for (numzero1 = 0; *p1 && (*p1 == '0'); numzero1++) p1++; for (numzero2 = 0; *p2 && (*p2 == '0'); numzero2++) p2++; /* find number of consecutive digits */ for (numdigit = 0; ; numdigit++) { if (isdigit(*(p1 + numdigit)) && isdigit(*(p2 + numdigit))) continue; else if (isdigit(*(p1 + numdigit))) return 1; /* s2 is bigger */ else if (isdigit(*(p2 + numdigit))) return -1; /* s1 is bigger */ else break; } /* same number of digits, compare size of number */ if (numdigit > 0) { int compare = (int)strncmp(p1, p2, (size_t)numdigit); if (compare != 0) return compare; } /* use number of leading zeros as tie breaker if still equal */ if (*tiebreaker == 0) { if (numzero1 > numzero2) *tiebreaker = 1; else if (numzero1 < numzero2) *tiebreaker = -1; } return 0; } /* natural string compare, keeping numbers in order */ int BLI_natstrcmp(const char *s1, const char *s2) { register int d1 = 0, d2 = 0; register char c1, c2; int tiebreaker = 0; /* if both chars are numeric, to a left_number_strcmp(). * then increase string deltas as long they are * numeric, else do a tolower and char compare */ while (1) { c1 = tolower(s1[d1]); c2 = tolower(s2[d2]); if (isdigit(c1) && isdigit(c2)) { int numcompare = left_number_strcmp(s1 + d1, s2 + d2, &tiebreaker); if (numcompare != 0) return numcompare; d1++; while (isdigit(s1[d1])) d1++; d2++; while (isdigit(s2[d2])) d2++; c1 = tolower(s1[d1]); c2 = tolower(s2[d2]); } /* first check for '.' so "foo.bar" comes before "foo 1.bar" */ if (c1 == '.' && c2 != '.') return -1; if (c1 != '.' && c2 == '.') return 1; else if (c1 < c2) { return -1; } else if (c1 > c2) { return 1; } else if (c1 == 0) { break; } d1++; d2++; } if (tiebreaker) return tiebreaker; /* we might still have a different string because of lower/upper case, in * that case fall back to regular string comparison */ return strcmp(s1, s2); } /** * Like strcmp, but will ignore any heading/trailing pad char for comparison. * So e.g. if pad is '*', '*world' and 'world*' will compare equal. */ int BLI_strcmp_ignore_pad(const char *str1, const char *str2, const char pad) { size_t str1_len, str2_len; while (*str1 == pad) { str1++; } while (*str2 == pad) { str2++; } str1_len = strlen(str1); str2_len = strlen(str2); while (str1_len && (str1[str1_len - 1] == pad)) { str1_len--; } while (str2_len && (str2[str2_len - 1] == pad)) { str2_len--; } if (str1_len == str2_len) { return strncmp(str1, str2, str2_len); } else if (str1_len > str2_len) { int ret = strncmp(str1, str2, str2_len); if (ret == 0) { ret = 1; } return ret; } else { int ret = strncmp(str1, str2, str1_len); if (ret == 0) { ret = -1; } return ret; } } /* determine the length of a fixed-size string */ size_t BLI_strnlen(const char *s, const size_t maxlen) { size_t len; for (len = 0; len < maxlen; len++, s++) { if (!*s) break; } return len; } void BLI_str_tolower_ascii(char *str, const size_t len) { size_t i; for (i = 0; (i < len) && str[i]; i++) if (str[i] >= 'A' && str[i] <= 'Z') str[i] += 'a' - 'A'; } void BLI_str_toupper_ascii(char *str, const size_t len) { size_t i; for (i = 0; (i < len) && str[i]; i++) if (str[i] >= 'a' && str[i] <= 'z') str[i] -= 'a' - 'A'; } /** * Strip trailing zeros from a float, eg: * 0.0000 -> 0.0 * 2.0010 -> 2.001 * * \param str * \param pad * \return The number of zeros stripped. */ int BLI_str_rstrip_float_zero(char *str, const char pad) { char *p = strchr(str, '.'); int totstrip = 0; if (p) { char *end_p; p++; /* position at first decimal place */ end_p = p + (strlen(p) - 1); /* position at last character */ if (end_p > p) { while (end_p != p && *end_p == '0') { *end_p = pad; end_p--; } } } return totstrip; } /** * Return index of a string in a string array. * * \param str The string to find. * \param str_array Array of strings. * \param str_array_len The length of the array, or -1 for a NULL-terminated array. * \return The index of str in str_array or -1. */ int BLI_str_index_in_array_n(const char *__restrict str, const char **__restrict str_array, const int str_array_len) { int index; const char **str_iter = str_array; for (index = 0; index < str_array_len; str_iter++, index++) { if (STREQ(str, *str_iter)) { return index; } } return -1; } /** * Return index of a string in a string array. * * \param str The string to find. * \param str_array Array of strings, (must be NULL-terminated). * \return The index of str in str_array or -1. */ int BLI_str_index_in_array(const char *__restrict str, const char **__restrict str_array) { int index; const char **str_iter = str_array; for (index = 0; *str_iter; str_iter++, index++) { if (STREQ(str, *str_iter)) { return index; } } return -1; } bool BLI_strn_endswith(const char *__restrict str, const char *__restrict end, size_t slength) { size_t elength = strlen(end); if (elength < slength) { const char *iter = &str[slength - elength]; while (*iter) { if (*iter++ != *end++) { return false; } } return true; } return false; } /** * Find if a string ends with another string. * * \param str The string to search within. * \param end The string we look for at the end. * \return If str ends with end. */ bool BLI_str_endswith(const char *__restrict str, const char * __restrict end) { const size_t slength = strlen(str); return BLI_strn_endswith(str, end, slength); } /** * Find the first char matching one of the chars in \a delim, from left. * * \param str The string to search within. * \param delim The set of delimiters to search for, as unicode values. * \param sep Return value, set to the first delimiter found (or NULL if none found). * \param suf Return value, set to next char after the first delimiter found (or NULL if none found). * \return The length of the prefix (i.e. *sep - str). */ size_t BLI_str_partition(const char *str, const char delim[], const char **sep, const char **suf) { return BLI_str_partition_ex(str, NULL, delim, sep, suf, false); } /** * Find the first char matching one of the chars in \a delim, from right. * * \param str The string to search within. * \param delim The set of delimiters to search for, as unicode values. * \param sep Return value, set to the first delimiter found (or NULL if none found). * \param suf Return value, set to next char after the first delimiter found (or NULL if none found). * \return The length of the prefix (i.e. *sep - str). */ size_t BLI_str_rpartition(const char *str, const char delim[], const char **sep, const char **suf) { return BLI_str_partition_ex(str, NULL, delim, sep, suf, true); } /** * Find the first char matching one of the chars in \a delim, either from left or right. * * \param str The string to search within. * \param end If non-NULL, the right delimiter of the string. * \param delim The set of delimiters to search for, as unicode values. * \param sep Return value, set to the first delimiter found (or NULL if none found). * \param suf Return value, set to next char after the first delimiter found (or NULL if none found). * \param from_right If %true, search from the right of \a str, else, search from its left. * \return The length of the prefix (i.e. *sep - str). */ size_t BLI_str_partition_ex( const char *str, const char *end, const char delim[], const char **sep, const char **suf, const bool from_right) { const char *d; char *(*func)(const char *str, int c) = from_right ? strrchr : strchr; BLI_assert(end == NULL || end > str); *sep = *suf = NULL; for (d = delim; *d != '\0'; ++d) { const char *tmp; if (end) { if (from_right) { for (tmp = end - 1; (tmp >= str) && (*tmp != *d); tmp--); if (tmp < str) { tmp = NULL; } } else { tmp = func(str, *d); if (tmp >= end) { tmp = NULL; } } } else { tmp = func(str, *d); } if (tmp && (from_right ? (*sep < tmp) : (!*sep || *sep > tmp))) { *sep = tmp; } } if (*sep) { *suf = *sep + 1; return (size_t)(*sep - str); } return end ? (size_t)(end - str) : strlen(str); } /** * Format ints with decimal grouping. * 1000 -> 1,000 * * \param dst The resulting string * \param num Number to format * \return The length of \a dst */ size_t BLI_str_format_int_grouped(char dst[16], int num) { char src[16]; char *p_src = src; char *p_dst = dst; const char separator = ','; int num_len, commas; num_len = sprintf(src, "%d", num); if (*p_src == '-') { *p_dst++ = *p_src++; num_len--; } for (commas = 2 - num_len % 3; *p_src; commas = (commas + 1) % 3) { *p_dst++ = *p_src++; if (commas == 1) { *p_dst++ = separator; } } *--p_dst = '\0'; return (size_t)(p_dst - dst); } /** * Find the ranges needed to split \a str into its individual words. * * \param str: The string to search for words. * \param len: Size of the string to search. * \param delim: Character to use as a delimiter. * \param r_words: Info about the words found. Set to [index, len] pairs. * \param words_max: Max number of words to find * \return The number of words found in \a str */ int BLI_string_find_split_words( const char *str, const size_t len, const char delim, int r_words[][2], int words_max) { int n = 0, i; bool charsearch = true; /* Skip leading spaces */ for (i = 0; (i < len) && (str[i] != '\0'); i++) { if (str[i] != delim) { break; } } for (; (i < len) && (str[i] != '\0') && (n < words_max); i++) { if ((str[i] != delim) && (charsearch == true)) { r_words[n][0] = i; charsearch = false; } else { if ((str[i] == delim) && (charsearch == false)) { r_words[n][1] = i - r_words[n][0]; n++; charsearch = true; } } } if (charsearch == false) { r_words[n][1] = i - r_words[n][0]; n++; } return n; }