1 /* Various utility functions.
2 Copyright (C) 2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
36 #ifdef HAVE_SYS_TIME_H
37 # include <sys/time.h>
43 # include <sys/mman.h>
51 #ifdef HAVE_SYS_UTIME_H
52 # include <sys/utime.h>
56 # include <libc.h> /* for access() */
62 /* For TIOCGWINSZ and friends: */
63 #ifdef HAVE_SYS_IOCTL_H
64 # include <sys/ioctl.h>
70 /* Needed for run_with_timeout. */
76 #ifndef HAVE_SIGSETJMP
77 /* If sigsetjmp is a macro, configure won't pick it up. */
79 # define HAVE_SIGSETJMP
83 #undef USE_SIGNAL_TIMEOUT
85 # if defined(HAVE_SIGSETJMP) || defined(HAVE_SIGBLOCK)
86 # define USE_SIGNAL_TIMEOUT
94 /* Utility function: like xstrdup(), but also lowercases S. */
97 xstrdup_lower (const char *s)
99 char *copy = xstrdup (s);
106 /* Copy the string formed by two pointers (one on the beginning, other
107 on the char after the last char) to a new, malloc-ed location.
110 strdupdelim (const char *beg, const char *end)
112 char *res = xmalloc (end - beg + 1);
113 memcpy (res, beg, end - beg);
114 res[end - beg] = '\0';
118 /* Parse a string containing comma-separated elements, and return a
119 vector of char pointers with the elements. Spaces following the
120 commas are ignored. */
122 sepstring (const char *s)
136 res = xrealloc (res, (i + 2) * sizeof (char *));
137 res[i] = strdupdelim (p, s);
140 /* Skip the blanks following the ','. */
148 res = xrealloc (res, (i + 2) * sizeof (char *));
149 res[i] = strdupdelim (p, s);
154 /* Like sprintf, but allocates a string of sufficient size with malloc
155 and returns it. GNU libc has a similar function named asprintf,
156 which requires the pointer to the string to be passed. */
159 aprintf (const char *fmt, ...)
161 /* This function is implemented using vsnprintf, which we provide
162 for the systems that don't have it. Therefore, it should be 100%
166 char *str = xmalloc (size);
173 /* See log_vprintf_internal for explanation why it's OK to rely
174 on the return value of vsnprintf. */
176 va_start (args, fmt);
177 n = vsnprintf (str, size, fmt, args);
180 /* If the printing worked, return the string. */
181 if (n > -1 && n < size)
184 /* Else try again with a larger buffer. */
185 if (n > -1) /* C99 */
186 size = n + 1; /* precisely what is needed */
188 size <<= 1; /* twice the old size */
189 str = xrealloc (str, size);
193 /* Concatenate the NULL-terminated list of string arguments into
194 freshly allocated space. */
197 concat_strings (const char *str0, ...)
200 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
203 const char *next_str;
204 int total_length = 0;
207 /* Calculate the length of and allocate the resulting string. */
210 va_start (args, str0);
211 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
213 int len = strlen (next_str);
214 if (argcount < countof (saved_lengths))
215 saved_lengths[argcount++] = len;
219 p = ret = xmalloc (total_length + 1);
221 /* Copy the strings into the allocated space. */
224 va_start (args, str0);
225 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
228 if (argcount < countof (saved_lengths))
229 len = saved_lengths[argcount++];
231 len = strlen (next_str);
232 memcpy (p, next_str, len);
241 /* Return pointer to a static char[] buffer in which zero-terminated
242 string-representation of TM (in form hh:mm:ss) is printed.
244 If TM is NULL, the current time will be used. */
247 time_str (time_t *tm)
249 static char output[15];
251 time_t secs = tm ? *tm : time (NULL);
255 /* In case of error, return the empty string. Maybe we should
256 just abort if this happens? */
260 ptm = localtime (&secs);
261 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
265 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
268 datetime_str (time_t *tm)
270 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
272 time_t secs = tm ? *tm : time (NULL);
276 /* In case of error, return the empty string. Maybe we should
277 just abort if this happens? */
281 ptm = localtime (&secs);
282 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
283 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
284 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
288 /* The Windows versions of the following two functions are defined in
293 fork_to_background (void)
296 /* Whether we arrange our own version of opt.lfilename here. */
297 int logfile_changed = 0;
301 /* We must create the file immediately to avoid either a race
302 condition (which arises from using unique_name and failing to
303 use fopen_excl) or lying to the user about the log file name
304 (which arises from using unique_name, printing the name, and
305 using fopen_excl later on.) */
306 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, 0, &opt.lfilename);
322 /* parent, no error */
323 printf (_("Continuing in background, pid %d.\n"), (int)pid);
325 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
326 exit (0); /* #### should we use _exit()? */
329 /* child: give up the privileges and keep running. */
331 freopen ("/dev/null", "r", stdin);
332 freopen ("/dev/null", "w", stdout);
333 freopen ("/dev/null", "w", stderr);
335 #endif /* not WINDOWS */
337 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
338 specified with TM. The atime ("access time") is set to the current
342 touch (const char *file, time_t tm)
344 #ifdef HAVE_STRUCT_UTIMBUF
345 struct utimbuf times;
353 times.actime = time (NULL);
354 if (utime (file, ×) == -1)
355 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
358 /* Checks if FILE is a symbolic link, and removes it if it is. Does
359 nothing under MS-Windows. */
361 remove_link (const char *file)
366 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
368 DEBUGP (("Unlinking %s (symlink).\n", file));
371 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
372 file, strerror (errno));
377 /* Does FILENAME exist? This is quite a lousy implementation, since
378 it supplies no error codes -- only a yes-or-no answer. Thus it
379 will return that a file does not exist if, e.g., the directory is
380 unreadable. I don't mind it too much currently, though. The
381 proper way should, of course, be to have a third, error state,
382 other than true/false, but that would introduce uncalled-for
383 additional complexity to the callers. */
385 file_exists_p (const char *filename)
388 return access (filename, F_OK) >= 0;
391 return stat (filename, &buf) >= 0;
395 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
396 Returns 0 on error. */
398 file_non_directory_p (const char *path)
401 /* Use lstat() rather than stat() so that symbolic links pointing to
402 directories can be identified correctly. */
403 if (lstat (path, &buf) != 0)
405 return S_ISDIR (buf.st_mode) ? 0 : 1;
408 /* Return the size of file named by FILENAME, or -1 if it cannot be
409 opened or seeked into. */
411 file_size (const char *filename)
413 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
415 /* We use fseek rather than stat to determine the file size because
416 that way we can also verify that the file is readable without
417 explicitly checking for permissions. Inspired by the POST patch
419 FILE *fp = fopen (filename, "rb");
422 fseeko (fp, 0, SEEK_END);
428 if (stat (filename, &st) < 0)
434 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
435 doesn't exist is found. Return a freshly allocated copy of the
439 unique_name_1 (const char *prefix)
442 int plen = strlen (prefix);
443 char *template = (char *)alloca (plen + 1 + 24);
444 char *template_tail = template + plen;
446 memcpy (template, prefix, plen);
447 *template_tail++ = '.';
450 number_to_string (template_tail, count++);
451 while (file_exists_p (template));
453 return xstrdup (template);
456 /* Return a unique file name, based on FILE.
458 More precisely, if FILE doesn't exist, it is returned unmodified.
459 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
460 file name that doesn't exist is returned.
462 The resulting file is not created, only verified that it didn't
463 exist at the point in time when the function was called.
464 Therefore, where security matters, don't rely that the file created
465 by this function exists until you open it with O_EXCL or
468 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
469 string. Otherwise, it may return FILE if the file doesn't exist
470 (and therefore doesn't need changing). */
473 unique_name (const char *file, int allow_passthrough)
475 /* If the FILE itself doesn't exist, return it without
477 if (!file_exists_p (file))
478 return allow_passthrough ? (char *)file : xstrdup (file);
480 /* Otherwise, find a numeric suffix that results in unused file name
482 return unique_name_1 (file);
485 /* Create a file based on NAME, except without overwriting an existing
486 file with that name. Providing O_EXCL is correctly implemented,
487 this function does not have the race condition associated with
488 opening the file returned by unique_name. */
491 unique_create (const char *name, int binary, char **opened_name)
493 /* unique file name, based on NAME */
494 char *uname = unique_name (name, 0);
496 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
499 uname = unique_name (name, 0);
501 if (opened_name && fp != NULL)
504 *opened_name = uname;
516 /* Open the file for writing, with the addition that the file is
517 opened "exclusively". This means that, if the file already exists,
518 this function will *fail* and errno will be set to EEXIST. If
519 BINARY is set, the file will be opened in binary mode, equivalent
522 If opening the file fails for any reason, including the file having
523 previously existed, this function returns NULL and sets errno
527 fopen_excl (const char *fname, int binary)
531 int flags = O_WRONLY | O_CREAT | O_EXCL;
536 fd = open (fname, flags, 0666);
539 return fdopen (fd, binary ? "wb" : "w");
540 #else /* not O_EXCL */
541 /* Manually check whether the file exists. This is prone to race
542 conditions, but systems without O_EXCL haven't deserved
544 if (file_exists_p (fname))
549 return fopen (fname, binary ? "wb" : "w");
550 #endif /* not O_EXCL */
553 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
554 are missing, create them first. In case any mkdir() call fails,
555 return its error status. Returns 0 on successful completion.
557 The behaviour of this function should be identical to the behaviour
558 of `mkdir -p' on systems where mkdir supports the `-p' option. */
560 make_directory (const char *directory)
562 int i, ret, quit = 0;
565 /* Make a copy of dir, to be able to write to it. Otherwise, the
566 function is unsafe if called with a read-only char *argument. */
567 STRDUP_ALLOCA (dir, directory);
569 /* If the first character of dir is '/', skip it (and thus enable
570 creation of absolute-pathname directories. */
571 for (i = (*dir == '/'); 1; ++i)
573 for (; dir[i] && dir[i] != '/'; i++)
578 /* Check whether the directory already exists. Allow creation of
579 of intermediate directories to fail, as the initial path components
580 are not necessarily directories! */
581 if (!file_exists_p (dir))
582 ret = mkdir (dir, 0777);
593 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
594 should be a file name.
596 file_merge("/foo/bar", "baz") => "/foo/baz"
597 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
598 file_merge("foo", "bar") => "bar"
600 In other words, it's a simpler and gentler version of uri_merge_1. */
603 file_merge (const char *base, const char *file)
606 const char *cut = (const char *)strrchr (base, '/');
609 return xstrdup (file);
611 result = xmalloc (cut - base + 1 + strlen (file) + 1);
612 memcpy (result, base, cut - base);
613 result[cut - base] = '/';
614 strcpy (result + (cut - base) + 1, file);
619 static int in_acclist (const char *const *, const char *, int);
621 /* Determine whether a file is acceptable to be followed, according to
622 lists of patterns to accept/reject. */
624 acceptable (const char *s)
628 while (l && s[l] != '/')
635 return (in_acclist ((const char *const *)opt.accepts, s, 1)
636 && !in_acclist ((const char *const *)opt.rejects, s, 1));
638 return in_acclist ((const char *const *)opt.accepts, s, 1);
640 else if (opt.rejects)
641 return !in_acclist ((const char *const *)opt.rejects, s, 1);
645 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
646 `/something', frontcmp() will return 1 only if S2 begins with
647 `/something'. Otherwise, 0 is returned. */
649 frontcmp (const char *s1, const char *s2)
651 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
655 /* Iterate through STRLIST, and return the first element that matches
656 S, through wildcards or front comparison (as appropriate). */
658 proclist (char **strlist, const char *s, enum accd flags)
661 for (x = strlist; *x; x++)
663 /* Remove leading '/' if ALLABS */
664 char *p = *x + ((flags & ALLABS) && (**x == '/'));
665 if (has_wildcards_p (p))
667 if (fnmatch (p, s, FNM_PATHNAME) == 0)
679 /* Returns whether DIRECTORY is acceptable for download, wrt the
680 include/exclude lists.
682 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
683 and absolute paths may be freely intermixed. */
685 accdir (const char *directory, enum accd flags)
687 /* Remove starting '/'. */
688 if (flags & ALLABS && *directory == '/')
692 if (!proclist (opt.includes, directory, flags))
697 if (proclist (opt.excludes, directory, flags))
703 /* Return non-zero if STRING ends with TAIL. For instance:
705 match_tail ("abc", "bc", 0) -> 1
706 match_tail ("abc", "ab", 0) -> 0
707 match_tail ("abc", "abc", 0) -> 1
709 If FOLD_CASE_P is non-zero, the comparison will be
713 match_tail (const char *string, const char *tail, int fold_case_p)
717 /* We want this to be fast, so we code two loops, one with
718 case-folding, one without. */
722 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
723 if (string[i] != tail[j])
728 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
729 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
733 /* If the tail was exhausted, the match was succesful. */
740 /* Checks whether string S matches each element of ACCEPTS. A list
741 element are matched either with fnmatch() or match_tail(),
742 according to whether the element contains wildcards or not.
744 If the BACKWARD is 0, don't do backward comparison -- just compare
747 in_acclist (const char *const *accepts, const char *s, int backward)
749 for (; *accepts; accepts++)
751 if (has_wildcards_p (*accepts))
753 /* fnmatch returns 0 if the pattern *does* match the
755 if (fnmatch (*accepts, s, 0) == 0)
762 if (match_tail (s, *accepts, 0))
767 if (!strcmp (s, *accepts))
775 /* Return the location of STR's suffix (file extension). Examples:
776 suffix ("foo.bar") -> "bar"
777 suffix ("foo.bar.baz") -> "baz"
778 suffix ("/foo/bar") -> NULL
779 suffix ("/foo.bar/baz") -> NULL */
781 suffix (const char *str)
785 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
789 return (char *)str + i;
794 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
798 has_wildcards_p (const char *s)
801 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
806 /* Return non-zero if FNAME ends with a typical HTML suffix. The
807 following (case-insensitive) suffixes are presumed to be HTML files:
811 ?html (`?' matches one character)
813 #### CAVEAT. This is not necessarily a good indication that FNAME
814 refers to a file that contains HTML! */
816 has_html_suffix_p (const char *fname)
820 if ((suf = suffix (fname)) == NULL)
822 if (!strcasecmp (suf, "html"))
824 if (!strcasecmp (suf, "htm"))
826 if (suf[0] && !strcasecmp (suf + 1, "html"))
831 /* Read a line from FP and return the pointer to freshly allocated
832 storage. The storage space is obtained through malloc() and should
833 be freed with free() when it is no longer needed.
835 The length of the line is not limited, except by available memory.
836 The newline character at the end of line is retained. The line is
837 terminated with a zero character.
839 After end-of-file is encountered without anything being read, NULL
840 is returned. NULL is also returned on error. To distinguish
841 between these two cases, use the stdio function ferror(). */
844 read_whole_line (FILE *fp)
848 char *line = xmalloc (bufsize);
850 while (fgets (line + length, bufsize - length, fp))
852 length += strlen (line + length);
854 /* Possible for example when reading from a binary file where
855 a line begins with \0. */
858 if (line[length - 1] == '\n')
861 /* fgets() guarantees to read the whole line, or to use up the
862 space we've given it. We can double the buffer
865 line = xrealloc (line, bufsize);
867 if (length == 0 || ferror (fp))
872 if (length + 1 < bufsize)
873 /* Relieve the memory from our exponential greediness. We say
874 `length + 1' because the terminating \0 is not included in
875 LENGTH. We don't need to zero-terminate the string ourselves,
876 though, because fgets() does that. */
877 line = xrealloc (line, length + 1);
881 /* Read FILE into memory. A pointer to `struct file_memory' are
882 returned; use struct element `content' to access file contents, and
883 the element `length' to know the file length. `content' is *not*
884 zero-terminated, and you should *not* read or write beyond the [0,
885 length) range of characters.
887 After you are done with the file contents, call read_file_free to
890 Depending on the operating system and the type of file that is
891 being read, read_file() either mmap's the file into memory, or
892 reads the file into the core using read().
894 If file is named "-", fileno(stdin) is used for reading instead.
895 If you want to read from a real file named "-", use "./-" instead. */
898 read_file (const char *file)
901 struct file_memory *fm;
903 int inhibit_close = 0;
905 /* Some magic in the finest tradition of Perl and its kin: if FILE
906 is "-", just use stdin. */
911 /* Note that we don't inhibit mmap() in this case. If stdin is
912 redirected from a regular file, mmap() will still work. */
915 fd = open (file, O_RDONLY);
918 fm = xnew (struct file_memory);
923 if (fstat (fd, &buf) < 0)
925 fm->length = buf.st_size;
926 /* NOTE: As far as I know, the callers of this function never
927 modify the file text. Relying on this would enable us to
928 specify PROT_READ and MAP_SHARED for a marginal gain in
929 efficiency, but at some cost to generality. */
930 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
932 if (fm->content == (char *)MAP_FAILED)
942 /* The most common reason why mmap() fails is that FD does not point
943 to a plain file. However, it's also possible that mmap() doesn't
944 work for a particular type of file. Therefore, whenever mmap()
945 fails, we just fall back to the regular method. */
946 #endif /* HAVE_MMAP */
949 size = 512; /* number of bytes fm->contents can
950 hold at any given time. */
951 fm->content = xmalloc (size);
955 if (fm->length > size / 2)
957 /* #### I'm not sure whether the whole exponential-growth
958 thing makes sense with kernel read. On Linux at least,
959 read() refuses to read more than 4K from a file at a
960 single chunk anyway. But other Unixes might optimize it
961 better, and it doesn't *hurt* anything, so I'm leaving
964 /* Normally, we grow SIZE exponentially to make the number
965 of calls to read() and realloc() logarithmic in relation
966 to file size. However, read() can read an amount of data
967 smaller than requested, and it would be unreasonable to
968 double SIZE every time *something* was read. Therefore,
969 we double SIZE only when the length exceeds half of the
970 entire allocated size. */
972 fm->content = xrealloc (fm->content, size);
974 nread = read (fd, fm->content + fm->length, size - fm->length);
976 /* Successful read. */
987 if (size > fm->length && fm->length != 0)
988 /* Due to exponential growth of fm->content, the allocated region
989 might be much larger than what is actually needed. */
990 fm->content = xrealloc (fm->content, fm->length);
1002 /* Release the resources held by FM. Specifically, this calls
1003 munmap() or xfree() on fm->content, depending whether mmap or
1004 malloc/read were used to read in the file. It also frees the
1005 memory needed to hold the FM structure itself. */
1008 read_file_free (struct file_memory *fm)
1013 munmap (fm->content, fm->length);
1018 xfree (fm->content);
1023 /* Free the pointers in a NULL-terminated vector of pointers, then
1024 free the pointer itself. */
1026 free_vec (char **vec)
1037 /* Append vector V2 to vector V1. The function frees V2 and
1038 reallocates V1 (thus you may not use the contents of neither
1039 pointer after the call). If V1 is NULL, V2 is returned. */
1041 merge_vecs (char **v1, char **v2)
1051 /* To avoid j == 0 */
1056 for (i = 0; v1[i]; i++);
1058 for (j = 0; v2[j]; j++);
1059 /* Reallocate v1. */
1060 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1061 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1066 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1067 is allocated as needed. Return the new value of the vector. */
1070 vec_append (char **vec, const char *str)
1072 int cnt; /* count of vector elements, including
1073 the one we're about to append */
1076 for (cnt = 0; vec[cnt]; cnt++)
1082 /* Reallocate the array to fit the new element and the NULL. */
1083 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1084 /* Append a copy of STR to the vector. */
1085 vec[cnt - 1] = xstrdup (str);
1090 /* Sometimes it's useful to create "sets" of strings, i.e. special
1091 hash tables where you want to store strings as keys and merely
1092 query for their existence. Here is a set of utility routines that
1093 makes that transparent. */
1096 string_set_add (struct hash_table *ht, const char *s)
1098 /* First check whether the set element already exists. If it does,
1099 do nothing so that we don't have to free() the old element and
1100 then strdup() a new one. */
1101 if (hash_table_contains (ht, s))
1104 /* We use "1" as value. It provides us a useful and clear arbitrary
1105 value, and it consumes no memory -- the pointers to the same
1106 string "1" will be shared by all the key-value pairs in all `set'
1108 hash_table_put (ht, xstrdup (s), "1");
1111 /* Synonym for hash_table_contains... */
1114 string_set_contains (struct hash_table *ht, const char *s)
1116 return hash_table_contains (ht, s);
1120 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1122 char ***arrayptr = (char ***) arg;
1123 *(*arrayptr)++ = (char *) key;
1127 /* Convert the specified string set to array. ARRAY should be large
1128 enough to hold hash_table_count(ht) char pointers. */
1130 void string_set_to_array (struct hash_table *ht, char **array)
1132 hash_table_map (ht, string_set_to_array_mapper, &array);
1136 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1143 string_set_free (struct hash_table *ht)
1145 hash_table_map (ht, string_set_free_mapper, NULL);
1146 hash_table_destroy (ht);
1150 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1157 /* Another utility function: call free() on all keys and values of HT. */
1160 free_keys_and_values (struct hash_table *ht)
1162 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1166 /* Add thousand separators to a number already in string form. Used
1167 by with_thousand_seps and with_thousand_seps_large. */
1170 add_thousand_seps (const char *repr)
1172 static char outbuf[48];
1177 /* Reset the pointers. */
1181 /* Ignore the sign for the purpose of adding thousand
1188 /* How many digits before the first separator? */
1189 mod = strlen (inptr) % 3;
1191 for (i = 0; i < mod; i++)
1192 *outptr++ = inptr[i];
1193 /* Now insert the rest of them, putting separator before every
1195 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1197 if (i % 3 == 0 && i1 != 0)
1199 *outptr++ = inptr[i1];
1201 /* Zero-terminate the string. */
1206 /* Return a static pointer to the number printed with thousand
1207 separators inserted at the right places. */
1210 with_thousand_seps (wgint l)
1213 /* Print the number into the buffer. */
1214 number_to_string (inbuf, l);
1215 return add_thousand_seps (inbuf);
1218 /* Write a string representation of LARGE_INT NUMBER into the provided
1221 It would be dangerous to use sprintf, because the code wouldn't
1222 work on a machine with gcc-provided long long support, but without
1223 libc support for "%lld". However, such old systems platforms
1224 typically lack snprintf and will end up using our version, which
1225 does support "%lld" whereever long longs are available. */
1228 large_int_to_string (char *buffer, int bufsize, LARGE_INT number)
1230 snprintf (buffer, bufsize, LARGE_INT_FMT, number);
1233 /* The same as with_thousand_seps, but works on LARGE_INT. */
1236 with_thousand_seps_large (LARGE_INT l)
1239 large_int_to_string (inbuf, sizeof (inbuf), l);
1240 return add_thousand_seps (inbuf);
1243 /* N, a byte quantity, is converted to a human-readable abberviated
1244 form a la sizes printed by `ls -lh'. The result is written to a
1245 static buffer, a pointer to which is returned.
1247 Unlike `with_thousand_seps', this approximates to the nearest unit.
1248 Quoting GNU libit: "Most people visually process strings of 3-4
1249 digits effectively, but longer strings of digits are more prone to
1250 misinterpretation. Hence, converting to an abbreviated form
1251 usually improves readability."
1253 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1254 original computer science meaning of "powers of 1024". Powers of
1255 1000 would be useless since Wget already displays sizes with
1256 thousand separators. We don't use the "*bibyte" names invented in
1257 1998, and seldom used in practice. Wikipedia's entry on kilobyte
1258 discusses this in some detail. */
1261 human_readable (wgint n)
1263 /* These suffixes are compatible with those of GNU `ls -lh'. */
1264 static char powers[] =
1266 'K', /* kilobyte, 2^10 bytes */
1267 'M', /* megabyte, 2^20 bytes */
1268 'G', /* gigabyte, 2^30 bytes */
1269 'T', /* terabyte, 2^40 bytes */
1270 'P', /* petabyte, 2^50 bytes */
1271 'E', /* exabyte, 2^60 bytes */
1276 /* If the quantity is smaller than 1K, just print it. */
1279 snprintf (buf, sizeof (buf), "%d", (int) n);
1283 /* Loop over powers, dividing N with 1024 in each iteration. This
1284 works unchanged for all sizes of wgint, while still avoiding
1285 non-portable `long double' arithmetic. */
1286 for (i = 0; i < countof (powers); i++)
1288 /* At each iteration N is greater than the *subsequent* power.
1289 That way N/1024.0 produces a decimal number in the units of
1291 if ((n >> 10) < 1024 || i == countof (powers) - 1)
1293 /* Must cast to long first because MS VC can't directly cast
1294 __int64 to double. (This is safe because N is known to
1296 double val = (double) (long) n / 1024.0;
1297 /* Print values smaller than 10 with one decimal digits, and
1298 others without any decimals. */
1299 snprintf (buf, sizeof (buf), "%.*f%c",
1300 val < 10 ? 1 : 0, val, powers[i]);
1305 return NULL; /* unreached */
1308 /* Count the digits in the provided number. Used to allocate space
1309 when printing numbers. */
1312 numdigit (wgint number)
1316 ++cnt; /* accomodate '-' */
1317 while ((number /= 10) != 0)
1322 #define PR(mask) *p++ = n / (mask) + '0'
1324 /* DIGITS_<D> is used to print a D-digit number and should be called
1325 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1326 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1327 Recursively this continues until DIGITS_1 is invoked. */
1329 #define DIGITS_1(mask) PR (mask)
1330 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1331 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1332 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1333 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1334 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1335 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1336 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1337 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1338 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1340 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1342 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1343 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1344 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1345 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1346 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1347 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1348 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1349 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1350 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1352 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1353 cases and to portably support strange sizes of wgint. Ideally this
1354 would just use "%j" and intmax_t, but many systems don't support
1355 it, so it's used only if nothing else works. */
1356 #if SIZEOF_LONG >= SIZEOF_WGINT
1357 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1358 #elif SIZEOF_LONG_LONG >= SIZEOF_WGINT
1359 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1360 #elif defined(WINDOWS)
1361 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64", (__int64) (n))
1363 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
1366 /* Shorthand for casting to wgint. */
1369 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1370 `sprintf(buffer, "%lld", (long long) number)', only typically much
1371 faster and portable to machines without long long.
1373 The speedup may make a difference in programs that frequently
1374 convert numbers to strings. Some implementations of sprintf,
1375 particularly the one in GNU libc, have been known to be extremely
1376 slow when converting integers to strings.
1378 Return the pointer to the location where the terminating zero was
1379 printed. (Equivalent to calling buffer+strlen(buffer) after the
1382 BUFFER should be big enough to accept as many bytes as you expect
1383 the number to take up. On machines with 64-bit longs the maximum
1384 needed size is 24 bytes. That includes the digits needed for the
1385 largest 64-bit number, the `-' sign in case it's negative, and the
1386 terminating '\0'. */
1389 number_to_string (char *buffer, wgint number)
1394 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1395 /* We are running in a strange or misconfigured environment. Let
1396 sprintf cope with it. */
1397 SPRINTF_WGINT (buffer, n);
1398 p += strlen (buffer);
1399 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1405 /* -n would overflow. Have sprintf deal with this. */
1406 SPRINTF_WGINT (buffer, n);
1407 p += strlen (buffer);
1415 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1416 way printing any N is fully open-coded without a loop or jump.
1417 (Also see description of DIGITS_*.) */
1419 if (n < 10) DIGITS_1 (1);
1420 else if (n < 100) DIGITS_2 (10);
1421 else if (n < 1000) DIGITS_3 (100);
1422 else if (n < 10000) DIGITS_4 (1000);
1423 else if (n < 100000) DIGITS_5 (10000);
1424 else if (n < 1000000) DIGITS_6 (100000);
1425 else if (n < 10000000) DIGITS_7 (1000000);
1426 else if (n < 100000000) DIGITS_8 (10000000);
1427 else if (n < 1000000000) DIGITS_9 (100000000);
1428 #if SIZEOF_WGINT == 4
1429 /* wgint is 32 bits wide: no number has more than 10 digits. */
1430 else DIGITS_10 (1000000000);
1432 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1433 Constants are constructed by compile-time multiplication to avoid
1434 dealing with different notations for 64-bit constants
1435 (nL/nLL/nI64, depending on the compiler and architecture). */
1436 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1437 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1438 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1439 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1440 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1441 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1442 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1443 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1444 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1445 else DIGITS_19 (1000000000*(W)1000000000);
1449 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1478 /* Print NUMBER to a statically allocated string and return a pointer
1479 to the printed representation.
1481 This function is intended to be used in conjunction with printf.
1482 It is hard to portably print wgint values:
1483 a) you cannot use printf("%ld", number) because wgint can be long
1484 long on 32-bit machines with LFS.
1485 b) you cannot use printf("%lld", number) because NUMBER could be
1486 long on 32-bit machines without LFS, or on 64-bit machines,
1487 which do not require LFS. Also, Windows doesn't support %lld.
1488 c) you cannot use printf("%j", (int_max_t) number) because not all
1489 versions of printf support "%j", the most notable being the one
1491 d) you cannot #define WGINT_FMT to the appropriate format and use
1492 printf(WGINT_FMT, number) because that would break translations
1493 for user-visible messages, such as printf("Downloaded: %d
1496 What you should use instead is printf("%s", number_to_static_string
1499 CAVEAT: since the function returns pointers to static data, you
1500 must be careful to copy its result before calling it again.
1501 However, to make it more useful with printf, the function maintains
1502 an internal ring of static buffers to return. That way things like
1503 printf("%s %s", number_to_static_string (num1),
1504 number_to_static_string (num2)) work as expected. Three buffers
1505 are currently used, which means that "%s %s %s" will work, but "%s
1506 %s %s %s" won't. If you need to print more than three wgints,
1507 bump the RING_SIZE (or rethink your message.) */
1510 number_to_static_string (wgint number)
1512 static char ring[RING_SIZE][24];
1514 char *buf = ring[ringpos];
1515 number_to_string (buf, number);
1516 ringpos = (ringpos + 1) % RING_SIZE;
1520 /* Determine the width of the terminal we're running on. If that's
1521 not possible, return 0. */
1524 determine_screen_width (void)
1526 /* If there's a way to get the terminal size using POSIX
1527 tcgetattr(), somebody please tell me. */
1532 if (opt.lfilename != NULL)
1535 fd = fileno (stderr);
1536 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1537 return 0; /* most likely ENOTTY */
1540 #elif defined(WINDOWS)
1541 CONSOLE_SCREEN_BUFFER_INFO csbi;
1542 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1544 return csbi.dwSize.X;
1545 #else /* neither TIOCGWINSZ nor WINDOWS */
1547 #endif /* neither TIOCGWINSZ nor WINDOWS */
1550 /* Return a random number between 0 and MAX-1, inclusive.
1552 If MAX is greater than the value of RAND_MAX+1 on the system, the
1553 returned value will be in the range [0, RAND_MAX]. This may be
1554 fixed in a future release.
1556 The random number generator is seeded automatically the first time
1559 This uses rand() for portability. It has been suggested that
1560 random() offers better randomness, but this is not required for
1561 Wget, so I chose to go for simplicity and use rand
1564 DO NOT use this for cryptographic purposes. It is only meant to be
1565 used in situations where quality of the random numbers returned
1566 doesn't really matter. */
1569 random_number (int max)
1577 srand (time (NULL));
1582 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1583 and enforce that assumption by masking other bits. */
1585 # define RAND_MAX 32767
1589 /* This is equivalent to rand() % max, but uses the high-order bits
1590 for better randomness on architecture where rand() is implemented
1591 using a simple congruential generator. */
1593 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1594 return (int)bounded;
1597 /* Return a random uniformly distributed floating point number in the
1598 [0, 1) range. The precision of returned numbers is 9 digits.
1600 Modify this to use erand48() where available! */
1605 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1606 sure it's greater than 1000. */
1607 int rnd1 = random_number (1000);
1608 int rnd2 = random_number (1000);
1609 int rnd3 = random_number (1000);
1610 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1613 /* Implementation of run_with_timeout, a generic timeout-forcing
1614 routine for systems with Unix-like signal handling. */
1616 #ifdef USE_SIGNAL_TIMEOUT
1617 # ifdef HAVE_SIGSETJMP
1618 # define SETJMP(env) sigsetjmp (env, 1)
1620 static sigjmp_buf run_with_timeout_env;
1623 abort_run_with_timeout (int sig)
1625 assert (sig == SIGALRM);
1626 siglongjmp (run_with_timeout_env, -1);
1628 # else /* not HAVE_SIGSETJMP */
1629 # define SETJMP(env) setjmp (env)
1631 static jmp_buf run_with_timeout_env;
1634 abort_run_with_timeout (int sig)
1636 assert (sig == SIGALRM);
1637 /* We don't have siglongjmp to preserve the set of blocked signals;
1638 if we longjumped out of the handler at this point, SIGALRM would
1639 remain blocked. We must unblock it manually. */
1640 int mask = siggetmask ();
1641 mask &= ~sigmask (SIGALRM);
1644 /* Now it's safe to longjump. */
1645 longjmp (run_with_timeout_env, -1);
1647 # endif /* not HAVE_SIGSETJMP */
1649 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1650 setitimer where available, alarm otherwise.
1652 TIMEOUT should be non-zero. If the timeout value is so small that
1653 it would be rounded to zero, it is rounded to the least legal value
1654 instead (1us for setitimer, 1s for alarm). That ensures that
1655 SIGALRM will be delivered in all cases. */
1658 alarm_set (double timeout)
1661 /* Use the modern itimer interface. */
1662 struct itimerval itv;
1664 itv.it_value.tv_sec = (long) timeout;
1665 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1666 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1667 /* Ensure that we wait for at least the minimum interval.
1668 Specifying zero would mean "wait forever". */
1669 itv.it_value.tv_usec = 1;
1670 setitimer (ITIMER_REAL, &itv, NULL);
1671 #else /* not ITIMER_REAL */
1672 /* Use the old alarm() interface. */
1673 int secs = (int) timeout;
1675 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1676 because alarm(0) means "never deliver the alarm", i.e. "wait
1677 forever", which is not what someone who specifies a 0.5s
1678 timeout would expect. */
1681 #endif /* not ITIMER_REAL */
1684 /* Cancel the alarm set with alarm_set. */
1690 struct itimerval disable;
1692 setitimer (ITIMER_REAL, &disable, NULL);
1693 #else /* not ITIMER_REAL */
1695 #endif /* not ITIMER_REAL */
1698 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1699 seconds. Returns non-zero if the function was interrupted with a
1700 timeout, zero otherwise.
1702 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1703 using setitimer() or alarm(). The timeout is enforced by
1704 longjumping out of the SIGALRM handler. This has several
1705 advantages compared to the traditional approach of relying on
1706 signals causing system calls to exit with EINTR:
1708 * The callback function is *forcibly* interrupted after the
1709 timeout expires, (almost) regardless of what it was doing and
1710 whether it was in a syscall. For example, a calculation that
1711 takes a long time is interrupted as reliably as an IO
1714 * It works with both SYSV and BSD signals because it doesn't
1715 depend on the default setting of SA_RESTART.
1717 * It doesn't require special handler setup beyond a simple call
1718 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1721 The only downside is that, if FUN allocates internal resources that
1722 are normally freed prior to exit from the functions, they will be
1723 lost in case of timeout. */
1726 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1736 signal (SIGALRM, abort_run_with_timeout);
1737 if (SETJMP (run_with_timeout_env) != 0)
1739 /* Longjumped out of FUN with a timeout. */
1740 signal (SIGALRM, SIG_DFL);
1743 alarm_set (timeout);
1746 /* Preserve errno in case alarm() or signal() modifies it. */
1747 saved_errno = errno;
1749 signal (SIGALRM, SIG_DFL);
1750 errno = saved_errno;
1755 #else /* not USE_SIGNAL_TIMEOUT */
1758 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1759 define it under Windows, because Windows has its own version of
1760 run_with_timeout that uses threads. */
1763 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1768 #endif /* not WINDOWS */
1769 #endif /* not USE_SIGNAL_TIMEOUT */
1773 /* Sleep the specified amount of seconds. On machines without
1774 nanosleep(), this may sleep shorter if interrupted by signals. */
1777 xsleep (double seconds)
1779 #ifdef HAVE_NANOSLEEP
1780 /* nanosleep is the preferred interface because it offers high
1781 accuracy and, more importantly, because it allows us to reliably
1782 restart receiving a signal such as SIGWINCH. (There was an
1783 actual Debian bug report about --limit-rate malfunctioning while
1784 the terminal was being resized.) */
1785 struct timespec sleep, remaining;
1786 sleep.tv_sec = (long) seconds;
1787 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1788 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1789 /* If nanosleep has been interrupted by a signal, adjust the
1790 sleeping period and return to sleep. */
1792 #elif defined(HAVE_USLEEP)
1793 /* If usleep is available, use it in preference to select. */
1796 /* On some systems, usleep cannot handle values larger than
1797 1,000,000. If the period is larger than that, use sleep
1798 first, then add usleep for subsecond accuracy. */
1800 seconds -= (long) seconds;
1802 usleep (seconds * 1000000);
1803 #elif defined(HAVE_SELECT)
1804 /* Note that, although Windows supports select, this sleeping
1805 strategy doesn't work there because Winsock's select doesn't
1806 implement timeout when it is passed NULL pointers for all fd
1807 sets. (But it does work under Cygwin, which implements its own
1809 struct timeval sleep;
1810 sleep.tv_sec = (long) seconds;
1811 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1812 select (0, NULL, NULL, NULL, &sleep);
1813 /* If select returns -1 and errno is EINTR, it means we were
1814 interrupted by a signal. But without knowing how long we've
1815 actually slept, we can't return to sleep. Using gettimeofday to
1816 track sleeps is slow and unreliable due to clock skew. */
1817 #else /* none of the above */
1822 #endif /* not WINDOWS */
1824 /* Encode the string STR of length LENGTH to base64 format and place it
1825 to B64STORE. The output will be \0-terminated, and must point to a
1826 writable buffer of at least 1+BASE64_LENGTH(length) bytes. It
1827 returns the length of the resulting base64 data, not counting the
1830 This implementation will not emit newlines after 76 characters of
1834 base64_encode (const char *str, int length, char *b64store)
1836 /* Conversion table. */
1837 static char tbl[64] = {
1838 'A','B','C','D','E','F','G','H',
1839 'I','J','K','L','M','N','O','P',
1840 'Q','R','S','T','U','V','W','X',
1841 'Y','Z','a','b','c','d','e','f',
1842 'g','h','i','j','k','l','m','n',
1843 'o','p','q','r','s','t','u','v',
1844 'w','x','y','z','0','1','2','3',
1845 '4','5','6','7','8','9','+','/'
1848 const unsigned char *s = (const unsigned char *) str;
1851 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1852 for (i = 0; i < length; i += 3)
1854 *p++ = tbl[s[0] >> 2];
1855 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1856 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1857 *p++ = tbl[s[2] & 0x3f];
1861 /* Pad the result if necessary... */
1862 if (i == length + 1)
1864 else if (i == length + 2)
1865 *(p - 1) = *(p - 2) = '=';
1867 /* ...and zero-terminate it. */
1870 return p - b64store;
1873 #define IS_ASCII(c) (((c) & 0x80) == 0)
1874 #define IS_BASE64(c) ((IS_ASCII (c) && base64_char_to_value[c] >= 0) || c == '=')
1876 /* Get next character from the string, except that non-base64
1877 characters are ignored, as mandated by rfc2045. */
1878 #define NEXT_BASE64_CHAR(c, p) do { \
1880 } while (c != '\0' && !IS_BASE64 (c))
1882 /* Decode data from BASE64 (assumed to be encoded as base64) into
1883 memory pointed to by TO. TO should be large enough to accomodate
1884 the decoded data, which is guaranteed to be less than
1887 Since TO is assumed to contain binary data, it is not
1888 NUL-terminated. The function returns the length of the data
1889 written to TO. -1 is returned in case of error caused by malformed
1893 base64_decode (const char *base64, char *to)
1895 /* Table of base64 values for first 128 characters. Note that this
1896 assumes ASCII (but so does Wget in other places). */
1897 static short base64_char_to_value[128] =
1899 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1900 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1901 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1902 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1903 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1904 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1905 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1906 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1907 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1908 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1909 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1910 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1911 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1914 const char *p = base64;
1920 unsigned long value;
1922 /* Process first byte of a quadruplet. */
1923 NEXT_BASE64_CHAR (c, p);
1927 return -1; /* illegal '=' while decoding base64 */
1928 value = base64_char_to_value[c] << 18;
1930 /* Process scond byte of a quadruplet. */
1931 NEXT_BASE64_CHAR (c, p);
1933 return -1; /* premature EOF while decoding base64 */
1935 return -1; /* illegal `=' while decoding base64 */
1936 value |= base64_char_to_value[c] << 12;
1939 /* Process third byte of a quadruplet. */
1940 NEXT_BASE64_CHAR (c, p);
1942 return -1; /* premature EOF while decoding base64 */
1946 NEXT_BASE64_CHAR (c, p);
1948 return -1; /* premature EOF while decoding base64 */
1950 return -1; /* padding `=' expected but not found */
1954 value |= base64_char_to_value[c] << 6;
1955 *q++ = 0xff & value >> 8;
1957 /* Process fourth byte of a quadruplet. */
1958 NEXT_BASE64_CHAR (c, p);
1960 return -1; /* premature EOF while decoding base64 */
1964 value |= base64_char_to_value[c];
1965 *q++ = 0xff & value;
1973 #undef NEXT_BASE64_CHAR
1975 /* Simple merge sort for use by stable_sort. Implementation courtesy
1976 Zeljko Vrba with additional debugging by Nenad Barbutov. */
1979 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
1980 int (*cmpfun) (const void *, const void *))
1982 #define ELT(array, pos) ((char *)(array) + (pos) * size)
1986 size_t mid = (to + from) / 2;
1987 mergesort_internal (base, temp, size, from, mid, cmpfun);
1988 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
1991 for (k = from; (i <= mid) && (j <= to); k++)
1992 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
1993 memcpy (ELT (temp, k), ELT (base, i++), size);
1995 memcpy (ELT (temp, k), ELT (base, j++), size);
1997 memcpy (ELT (temp, k++), ELT (base, i++), size);
1999 memcpy (ELT (temp, k++), ELT (base, j++), size);
2000 for (k = from; k <= to; k++)
2001 memcpy (ELT (base, k), ELT (temp, k), size);
2006 /* Stable sort with interface exactly like standard library's qsort.
2007 Uses mergesort internally, allocating temporary storage with
2011 stable_sort (void *base, size_t nmemb, size_t size,
2012 int (*cmpfun) (const void *, const void *))
2016 void *temp = alloca (nmemb * size * sizeof (void *));
2017 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);