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
84 #if defined(HAVE_SIGSETJMP) || defined(HAVE_SIGBLOCK)
85 # define USE_SIGNAL_TIMEOUT
92 /* Utility function: like xstrdup(), but also lowercases S. */
95 xstrdup_lower (const char *s)
97 char *copy = xstrdup (s);
104 /* Copy the string formed by two pointers (one on the beginning, other
105 on the char after the last char) to a new, malloc-ed location.
108 strdupdelim (const char *beg, const char *end)
110 char *res = xmalloc (end - beg + 1);
111 memcpy (res, beg, end - beg);
112 res[end - beg] = '\0';
116 /* Parse a string containing comma-separated elements, and return a
117 vector of char pointers with the elements. Spaces following the
118 commas are ignored. */
120 sepstring (const char *s)
134 res = xrealloc (res, (i + 2) * sizeof (char *));
135 res[i] = strdupdelim (p, s);
138 /* Skip the blanks following the ','. */
146 res = xrealloc (res, (i + 2) * sizeof (char *));
147 res[i] = strdupdelim (p, s);
152 /* Like sprintf, but allocates a string of sufficient size with malloc
153 and returns it. GNU libc has a similar function named asprintf,
154 which requires the pointer to the string to be passed. */
157 aprintf (const char *fmt, ...)
159 /* This function is implemented using vsnprintf, which we provide
160 for the systems that don't have it. Therefore, it should be 100%
164 char *str = xmalloc (size);
171 /* See log_vprintf_internal for explanation why it's OK to rely
172 on the return value of vsnprintf. */
174 va_start (args, fmt);
175 n = vsnprintf (str, size, fmt, args);
178 /* If the printing worked, return the string. */
179 if (n > -1 && n < size)
182 /* Else try again with a larger buffer. */
183 if (n > -1) /* C99 */
184 size = n + 1; /* precisely what is needed */
186 size <<= 1; /* twice the old size */
187 str = xrealloc (str, size);
191 /* Concatenate the NULL-terminated list of string arguments into
192 freshly allocated space. */
195 concat_strings (const char *str0, ...)
198 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
201 const char *next_str;
202 int total_length = 0;
205 /* Calculate the length of and allocate the resulting string. */
208 va_start (args, str0);
209 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
211 int len = strlen (next_str);
212 if (argcount < countof (saved_lengths))
213 saved_lengths[argcount++] = len;
217 p = ret = xmalloc (total_length + 1);
219 /* Copy the strings into the allocated space. */
222 va_start (args, str0);
223 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
226 if (argcount < countof (saved_lengths))
227 len = saved_lengths[argcount++];
229 len = strlen (next_str);
230 memcpy (p, next_str, len);
239 /* Return pointer to a static char[] buffer in which zero-terminated
240 string-representation of TM (in form hh:mm:ss) is printed.
242 If TM is NULL, the current time will be used. */
245 time_str (time_t *tm)
247 static char output[15];
249 time_t secs = tm ? *tm : time (NULL);
253 /* In case of error, return the empty string. Maybe we should
254 just abort if this happens? */
258 ptm = localtime (&secs);
259 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
263 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
266 datetime_str (time_t *tm)
268 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
270 time_t secs = tm ? *tm : time (NULL);
274 /* In case of error, return the empty string. Maybe we should
275 just abort if this happens? */
279 ptm = localtime (&secs);
280 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
281 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
282 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
286 /* The Windows versions of the following two functions are defined in
291 fork_to_background (void)
294 /* Whether we arrange our own version of opt.lfilename here. */
295 bool logfile_changed = false;
299 /* We must create the file immediately to avoid either a race
300 condition (which arises from using unique_name and failing to
301 use fopen_excl) or lying to the user about the log file name
302 (which arises from using unique_name, printing the name, and
303 using fopen_excl later on.) */
304 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
307 logfile_changed = true;
320 /* parent, no error */
321 printf (_("Continuing in background, pid %d.\n"), (int) pid);
323 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
324 exit (0); /* #### should we use _exit()? */
327 /* child: give up the privileges and keep running. */
329 freopen ("/dev/null", "r", stdin);
330 freopen ("/dev/null", "w", stdout);
331 freopen ("/dev/null", "w", stderr);
333 #endif /* not WINDOWS */
335 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
336 specified with TM. The atime ("access time") is set to the current
340 touch (const char *file, time_t tm)
342 #ifdef HAVE_STRUCT_UTIMBUF
343 struct utimbuf times;
351 times.actime = time (NULL);
352 if (utime (file, ×) == -1)
353 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
356 /* Checks if FILE is a symbolic link, and removes it if it is. Does
357 nothing under MS-Windows. */
359 remove_link (const char *file)
364 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
366 DEBUGP (("Unlinking %s (symlink).\n", file));
369 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
370 file, strerror (errno));
375 /* Does FILENAME exist? This is quite a lousy implementation, since
376 it supplies no error codes -- only a yes-or-no answer. Thus it
377 will return that a file does not exist if, e.g., the directory is
378 unreadable. I don't mind it too much currently, though. The
379 proper way should, of course, be to have a third, error state,
380 other than true/false, but that would introduce uncalled-for
381 additional complexity to the callers. */
383 file_exists_p (const char *filename)
386 return access (filename, F_OK) >= 0;
389 return stat (filename, &buf) >= 0;
393 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
394 Returns 0 on error. */
396 file_non_directory_p (const char *path)
399 /* Use lstat() rather than stat() so that symbolic links pointing to
400 directories can be identified correctly. */
401 if (lstat (path, &buf) != 0)
403 return S_ISDIR (buf.st_mode) ? false : true;
406 /* Return the size of file named by FILENAME, or -1 if it cannot be
407 opened or seeked into. */
409 file_size (const char *filename)
411 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
413 /* We use fseek rather than stat to determine the file size because
414 that way we can also verify that the file is readable without
415 explicitly checking for permissions. Inspired by the POST patch
417 FILE *fp = fopen (filename, "rb");
420 fseeko (fp, 0, SEEK_END);
426 if (stat (filename, &st) < 0)
432 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
433 doesn't exist is found. Return a freshly allocated copy of the
437 unique_name_1 (const char *prefix)
440 int plen = strlen (prefix);
441 char *template = (char *)alloca (plen + 1 + 24);
442 char *template_tail = template + plen;
444 memcpy (template, prefix, plen);
445 *template_tail++ = '.';
448 number_to_string (template_tail, count++);
449 while (file_exists_p (template));
451 return xstrdup (template);
454 /* Return a unique file name, based on FILE.
456 More precisely, if FILE doesn't exist, it is returned unmodified.
457 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
458 file name that doesn't exist is returned.
460 The resulting file is not created, only verified that it didn't
461 exist at the point in time when the function was called.
462 Therefore, where security matters, don't rely that the file created
463 by this function exists until you open it with O_EXCL or
466 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
467 string. Otherwise, it may return FILE if the file doesn't exist
468 (and therefore doesn't need changing). */
471 unique_name (const char *file, bool allow_passthrough)
473 /* If the FILE itself doesn't exist, return it without
475 if (!file_exists_p (file))
476 return allow_passthrough ? (char *)file : xstrdup (file);
478 /* Otherwise, find a numeric suffix that results in unused file name
480 return unique_name_1 (file);
483 /* Create a file based on NAME, except without overwriting an existing
484 file with that name. Providing O_EXCL is correctly implemented,
485 this function does not have the race condition associated with
486 opening the file returned by unique_name. */
489 unique_create (const char *name, bool binary, char **opened_name)
491 /* unique file name, based on NAME */
492 char *uname = unique_name (name, false);
494 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
497 uname = unique_name (name, false);
499 if (opened_name && fp != NULL)
502 *opened_name = uname;
514 /* Open the file for writing, with the addition that the file is
515 opened "exclusively". This means that, if the file already exists,
516 this function will *fail* and errno will be set to EEXIST. If
517 BINARY is set, the file will be opened in binary mode, equivalent
520 If opening the file fails for any reason, including the file having
521 previously existed, this function returns NULL and sets errno
525 fopen_excl (const char *fname, bool binary)
529 int flags = O_WRONLY | O_CREAT | O_EXCL;
534 fd = open (fname, flags, 0666);
537 return fdopen (fd, binary ? "wb" : "w");
538 #else /* not O_EXCL */
539 /* Manually check whether the file exists. This is prone to race
540 conditions, but systems without O_EXCL haven't deserved
542 if (file_exists_p (fname))
547 return fopen (fname, binary ? "wb" : "w");
548 #endif /* not O_EXCL */
551 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
552 are missing, create them first. In case any mkdir() call fails,
553 return its error status. Returns 0 on successful completion.
555 The behaviour of this function should be identical to the behaviour
556 of `mkdir -p' on systems where mkdir supports the `-p' option. */
558 make_directory (const char *directory)
560 int i, ret, quit = 0;
563 /* Make a copy of dir, to be able to write to it. Otherwise, the
564 function is unsafe if called with a read-only char *argument. */
565 STRDUP_ALLOCA (dir, directory);
567 /* If the first character of dir is '/', skip it (and thus enable
568 creation of absolute-pathname directories. */
569 for (i = (*dir == '/'); 1; ++i)
571 for (; dir[i] && dir[i] != '/'; i++)
576 /* Check whether the directory already exists. Allow creation of
577 of intermediate directories to fail, as the initial path components
578 are not necessarily directories! */
579 if (!file_exists_p (dir))
580 ret = mkdir (dir, 0777);
591 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
592 should be a file name.
594 file_merge("/foo/bar", "baz") => "/foo/baz"
595 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
596 file_merge("foo", "bar") => "bar"
598 In other words, it's a simpler and gentler version of uri_merge_1. */
601 file_merge (const char *base, const char *file)
604 const char *cut = (const char *)strrchr (base, '/');
607 return xstrdup (file);
609 result = xmalloc (cut - base + 1 + strlen (file) + 1);
610 memcpy (result, base, cut - base);
611 result[cut - base] = '/';
612 strcpy (result + (cut - base) + 1, file);
617 static bool in_acclist (const char *const *, const char *, bool);
619 /* Determine whether a file is acceptable to be followed, according to
620 lists of patterns to accept/reject. */
622 acceptable (const char *s)
626 while (l && s[l] != '/')
633 return (in_acclist ((const char *const *)opt.accepts, s, true)
634 && !in_acclist ((const char *const *)opt.rejects, s, true));
636 return in_acclist ((const char *const *)opt.accepts, s, true);
638 else if (opt.rejects)
639 return !in_acclist ((const char *const *)opt.rejects, s, true);
643 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
644 `/something', frontcmp() will return 1 only if S2 begins with
645 `/something'. Otherwise, 0 is returned. */
647 frontcmp (const char *s1, const char *s2)
649 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
653 /* Iterate through STRLIST, and return the first element that matches
654 S, through wildcards or front comparison (as appropriate). */
656 proclist (char **strlist, const char *s, enum accd flags)
659 for (x = strlist; *x; x++)
661 /* Remove leading '/' if ALLABS */
662 char *p = *x + ((flags & ALLABS) && (**x == '/'));
663 if (has_wildcards_p (p))
665 if (fnmatch (p, s, FNM_PATHNAME) == 0)
677 /* Returns whether DIRECTORY is acceptable for download, wrt the
678 include/exclude lists.
680 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
681 and absolute paths may be freely intermixed. */
683 accdir (const char *directory, enum accd flags)
685 /* Remove starting '/'. */
686 if (flags & ALLABS && *directory == '/')
690 if (!proclist (opt.includes, directory, flags))
695 if (proclist (opt.excludes, directory, flags))
701 /* Return true if STRING ends with TAIL. For instance:
703 match_tail ("abc", "bc", false) -> 1
704 match_tail ("abc", "ab", false) -> 0
705 match_tail ("abc", "abc", false) -> 1
707 If FOLD_CASE is true, the comparison will be case-insensitive. */
710 match_tail (const char *string, const char *tail, bool fold_case)
714 /* We want this to be fast, so we code two loops, one with
715 case-folding, one without. */
719 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
720 if (string[i] != tail[j])
725 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
726 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
730 /* If the tail was exhausted, the match was succesful. */
737 /* Checks whether string S matches each element of ACCEPTS. A list
738 element are matched either with fnmatch() or match_tail(),
739 according to whether the element contains wildcards or not.
741 If the BACKWARD is false, don't do backward comparison -- just compare
744 in_acclist (const char *const *accepts, const char *s, bool backward)
746 for (; *accepts; accepts++)
748 if (has_wildcards_p (*accepts))
750 /* fnmatch returns 0 if the pattern *does* match the
752 if (fnmatch (*accepts, s, 0) == 0)
759 if (match_tail (s, *accepts, 0))
764 if (!strcmp (s, *accepts))
772 /* Return the location of STR's suffix (file extension). Examples:
773 suffix ("foo.bar") -> "bar"
774 suffix ("foo.bar.baz") -> "baz"
775 suffix ("/foo/bar") -> NULL
776 suffix ("/foo.bar/baz") -> NULL */
778 suffix (const char *str)
782 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
786 return (char *)str + i;
791 /* Return true if S contains globbing wildcards (`*', `?', `[' or
795 has_wildcards_p (const char *s)
798 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
803 /* Return true if FNAME ends with a typical HTML suffix. The
804 following (case-insensitive) suffixes are presumed to be HTML
809 ?html (`?' matches one character)
811 #### CAVEAT. This is not necessarily a good indication that FNAME
812 refers to a file that contains HTML! */
814 has_html_suffix_p (const char *fname)
818 if ((suf = suffix (fname)) == NULL)
820 if (!strcasecmp (suf, "html"))
822 if (!strcasecmp (suf, "htm"))
824 if (suf[0] && !strcasecmp (suf + 1, "html"))
829 /* Read a line from FP and return the pointer to freshly allocated
830 storage. The storage space is obtained through malloc() and should
831 be freed with free() when it is no longer needed.
833 The length of the line is not limited, except by available memory.
834 The newline character at the end of line is retained. The line is
835 terminated with a zero character.
837 After end-of-file is encountered without anything being read, NULL
838 is returned. NULL is also returned on error. To distinguish
839 between these two cases, use the stdio function ferror(). */
842 read_whole_line (FILE *fp)
846 char *line = xmalloc (bufsize);
848 while (fgets (line + length, bufsize - length, fp))
850 length += strlen (line + length);
852 /* Possible for example when reading from a binary file where
853 a line begins with \0. */
856 if (line[length - 1] == '\n')
859 /* fgets() guarantees to read the whole line, or to use up the
860 space we've given it. We can double the buffer
863 line = xrealloc (line, bufsize);
865 if (length == 0 || ferror (fp))
870 if (length + 1 < bufsize)
871 /* Relieve the memory from our exponential greediness. We say
872 `length + 1' because the terminating \0 is not included in
873 LENGTH. We don't need to zero-terminate the string ourselves,
874 though, because fgets() does that. */
875 line = xrealloc (line, length + 1);
879 /* Read FILE into memory. A pointer to `struct file_memory' are
880 returned; use struct element `content' to access file contents, and
881 the element `length' to know the file length. `content' is *not*
882 zero-terminated, and you should *not* read or write beyond the [0,
883 length) range of characters.
885 After you are done with the file contents, call read_file_free to
888 Depending on the operating system and the type of file that is
889 being read, read_file() either mmap's the file into memory, or
890 reads the file into the core using read().
892 If file is named "-", fileno(stdin) is used for reading instead.
893 If you want to read from a real file named "-", use "./-" instead. */
896 read_file (const char *file)
899 struct file_memory *fm;
901 bool inhibit_close = false;
903 /* Some magic in the finest tradition of Perl and its kin: if FILE
904 is "-", just use stdin. */
908 inhibit_close = true;
909 /* Note that we don't inhibit mmap() in this case. If stdin is
910 redirected from a regular file, mmap() will still work. */
913 fd = open (file, O_RDONLY);
916 fm = xnew (struct file_memory);
921 if (fstat (fd, &buf) < 0)
923 fm->length = buf.st_size;
924 /* NOTE: As far as I know, the callers of this function never
925 modify the file text. Relying on this would enable us to
926 specify PROT_READ and MAP_SHARED for a marginal gain in
927 efficiency, but at some cost to generality. */
928 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
930 if (fm->content == (char *)MAP_FAILED)
940 /* The most common reason why mmap() fails is that FD does not point
941 to a plain file. However, it's also possible that mmap() doesn't
942 work for a particular type of file. Therefore, whenever mmap()
943 fails, we just fall back to the regular method. */
944 #endif /* HAVE_MMAP */
947 size = 512; /* number of bytes fm->contents can
948 hold at any given time. */
949 fm->content = xmalloc (size);
953 if (fm->length > size / 2)
955 /* #### I'm not sure whether the whole exponential-growth
956 thing makes sense with kernel read. On Linux at least,
957 read() refuses to read more than 4K from a file at a
958 single chunk anyway. But other Unixes might optimize it
959 better, and it doesn't *hurt* anything, so I'm leaving
962 /* Normally, we grow SIZE exponentially to make the number
963 of calls to read() and realloc() logarithmic in relation
964 to file size. However, read() can read an amount of data
965 smaller than requested, and it would be unreasonable to
966 double SIZE every time *something* was read. Therefore,
967 we double SIZE only when the length exceeds half of the
968 entire allocated size. */
970 fm->content = xrealloc (fm->content, size);
972 nread = read (fd, fm->content + fm->length, size - fm->length);
974 /* Successful read. */
985 if (size > fm->length && fm->length != 0)
986 /* Due to exponential growth of fm->content, the allocated region
987 might be much larger than what is actually needed. */
988 fm->content = xrealloc (fm->content, fm->length);
1000 /* Release the resources held by FM. Specifically, this calls
1001 munmap() or xfree() on fm->content, depending whether mmap or
1002 malloc/read were used to read in the file. It also frees the
1003 memory needed to hold the FM structure itself. */
1006 read_file_free (struct file_memory *fm)
1011 munmap (fm->content, fm->length);
1016 xfree (fm->content);
1021 /* Free the pointers in a NULL-terminated vector of pointers, then
1022 free the pointer itself. */
1024 free_vec (char **vec)
1035 /* Append vector V2 to vector V1. The function frees V2 and
1036 reallocates V1 (thus you may not use the contents of neither
1037 pointer after the call). If V1 is NULL, V2 is returned. */
1039 merge_vecs (char **v1, char **v2)
1049 /* To avoid j == 0 */
1054 for (i = 0; v1[i]; i++);
1056 for (j = 0; v2[j]; j++);
1057 /* Reallocate v1. */
1058 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1059 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1064 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1065 is allocated as needed. Return the new value of the vector. */
1068 vec_append (char **vec, const char *str)
1070 int cnt; /* count of vector elements, including
1071 the one we're about to append */
1074 for (cnt = 0; vec[cnt]; cnt++)
1080 /* Reallocate the array to fit the new element and the NULL. */
1081 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1082 /* Append a copy of STR to the vector. */
1083 vec[cnt - 1] = xstrdup (str);
1088 /* Sometimes it's useful to create "sets" of strings, i.e. special
1089 hash tables where you want to store strings as keys and merely
1090 query for their existence. Here is a set of utility routines that
1091 makes that transparent. */
1094 string_set_add (struct hash_table *ht, const char *s)
1096 /* First check whether the set element already exists. If it does,
1097 do nothing so that we don't have to free() the old element and
1098 then strdup() a new one. */
1099 if (hash_table_contains (ht, s))
1102 /* We use "1" as value. It provides us a useful and clear arbitrary
1103 value, and it consumes no memory -- the pointers to the same
1104 string "1" will be shared by all the key-value pairs in all `set'
1106 hash_table_put (ht, xstrdup (s), "1");
1109 /* Synonym for hash_table_contains... */
1112 string_set_contains (struct hash_table *ht, const char *s)
1114 return hash_table_contains (ht, s);
1118 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1120 char ***arrayptr = (char ***) arg;
1121 *(*arrayptr)++ = (char *) key;
1125 /* Convert the specified string set to array. ARRAY should be large
1126 enough to hold hash_table_count(ht) char pointers. */
1128 void string_set_to_array (struct hash_table *ht, char **array)
1130 hash_table_map (ht, string_set_to_array_mapper, &array);
1134 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1141 string_set_free (struct hash_table *ht)
1143 hash_table_map (ht, string_set_free_mapper, NULL);
1144 hash_table_destroy (ht);
1148 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1155 /* Another utility function: call free() on all keys and values of HT. */
1158 free_keys_and_values (struct hash_table *ht)
1160 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1164 /* Add thousand separators to a number already in string form. Used
1165 by with_thousand_seps and with_thousand_seps_large. */
1168 add_thousand_seps (const char *repr)
1170 static char outbuf[48];
1175 /* Reset the pointers. */
1179 /* Ignore the sign for the purpose of adding thousand
1186 /* How many digits before the first separator? */
1187 mod = strlen (inptr) % 3;
1189 for (i = 0; i < mod; i++)
1190 *outptr++ = inptr[i];
1191 /* Now insert the rest of them, putting separator before every
1193 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1195 if (i % 3 == 0 && i1 != 0)
1197 *outptr++ = inptr[i1];
1199 /* Zero-terminate the string. */
1204 /* Return a static pointer to the number printed with thousand
1205 separators inserted at the right places. */
1208 with_thousand_seps (wgint l)
1211 /* Print the number into the buffer. */
1212 number_to_string (inbuf, l);
1213 return add_thousand_seps (inbuf);
1216 /* Write a string representation of LARGE_INT NUMBER into the provided
1219 It would be dangerous to use sprintf, because the code wouldn't
1220 work on a machine with gcc-provided long long support, but without
1221 libc support for "%lld". However, such old systems platforms
1222 typically lack snprintf and will end up using our version, which
1223 does support "%lld" whereever long longs are available. */
1226 large_int_to_string (char *buffer, int bufsize, LARGE_INT number)
1228 snprintf (buffer, bufsize, LARGE_INT_FMT, number);
1231 /* The same as with_thousand_seps, but works on LARGE_INT. */
1234 with_thousand_seps_large (LARGE_INT l)
1237 large_int_to_string (inbuf, sizeof (inbuf), l);
1238 return add_thousand_seps (inbuf);
1241 /* N, a byte quantity, is converted to a human-readable abberviated
1242 form a la sizes printed by `ls -lh'. The result is written to a
1243 static buffer, a pointer to which is returned.
1245 Unlike `with_thousand_seps', this approximates to the nearest unit.
1246 Quoting GNU libit: "Most people visually process strings of 3-4
1247 digits effectively, but longer strings of digits are more prone to
1248 misinterpretation. Hence, converting to an abbreviated form
1249 usually improves readability."
1251 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1252 original computer science meaning of "powers of 1024". Powers of
1253 1000 would be useless since Wget already displays sizes with
1254 thousand separators. We don't use the "*bibyte" names invented in
1255 1998, and seldom used in practice. Wikipedia's entry on kilobyte
1256 discusses this in some detail. */
1259 human_readable (wgint n)
1261 /* These suffixes are compatible with those of GNU `ls -lh'. */
1262 static char powers[] =
1264 'K', /* kilobyte, 2^10 bytes */
1265 'M', /* megabyte, 2^20 bytes */
1266 'G', /* gigabyte, 2^30 bytes */
1267 'T', /* terabyte, 2^40 bytes */
1268 'P', /* petabyte, 2^50 bytes */
1269 'E', /* exabyte, 2^60 bytes */
1274 /* If the quantity is smaller than 1K, just print it. */
1277 snprintf (buf, sizeof (buf), "%d", (int) n);
1281 /* Loop over powers, dividing N with 1024 in each iteration. This
1282 works unchanged for all sizes of wgint, while still avoiding
1283 non-portable `long double' arithmetic. */
1284 for (i = 0; i < countof (powers); i++)
1286 /* At each iteration N is greater than the *subsequent* power.
1287 That way N/1024.0 produces a decimal number in the units of
1289 if ((n >> 10) < 1024 || i == countof (powers) - 1)
1291 /* Must cast to long first because MS VC can't directly cast
1292 __int64 to double. (This is safe because N is known to
1294 double val = (double) (long) n / 1024.0;
1295 /* Print values smaller than 10 with one decimal digits, and
1296 others without any decimals. */
1297 snprintf (buf, sizeof (buf), "%.*f%c",
1298 val < 10 ? 1 : 0, val, powers[i]);
1303 return NULL; /* unreached */
1306 /* Count the digits in the provided number. Used to allocate space
1307 when printing numbers. */
1310 numdigit (wgint number)
1314 ++cnt; /* accomodate '-' */
1315 while ((number /= 10) != 0)
1320 #define PR(mask) *p++ = n / (mask) + '0'
1322 /* DIGITS_<D> is used to print a D-digit number and should be called
1323 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1324 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1325 Recursively this continues until DIGITS_1 is invoked. */
1327 #define DIGITS_1(mask) PR (mask)
1328 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1329 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1330 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1331 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1332 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1333 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1334 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1335 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1336 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1338 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1340 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1341 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1342 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1343 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1344 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1345 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1346 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1347 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1348 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1350 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1351 cases and to portably support strange sizes of wgint. Ideally this
1352 would just use "%j" and intmax_t, but many systems don't support
1353 it, so it's used only if nothing else works. */
1354 #if SIZEOF_LONG >= SIZEOF_WGINT
1355 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1356 #elif SIZEOF_LONG_LONG >= SIZEOF_WGINT
1357 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1358 #elif defined(WINDOWS)
1359 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64", (__int64) (n))
1361 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
1364 /* Shorthand for casting to wgint. */
1367 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1368 `sprintf(buffer, "%lld", (long long) number)', only typically much
1369 faster and portable to machines without long long.
1371 The speedup may make a difference in programs that frequently
1372 convert numbers to strings. Some implementations of sprintf,
1373 particularly the one in GNU libc, have been known to be extremely
1374 slow when converting integers to strings.
1376 Return the pointer to the location where the terminating zero was
1377 printed. (Equivalent to calling buffer+strlen(buffer) after the
1380 BUFFER should be big enough to accept as many bytes as you expect
1381 the number to take up. On machines with 64-bit longs the maximum
1382 needed size is 24 bytes. That includes the digits needed for the
1383 largest 64-bit number, the `-' sign in case it's negative, and the
1384 terminating '\0'. */
1387 number_to_string (char *buffer, wgint number)
1392 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1393 /* We are running in a strange or misconfigured environment. Let
1394 sprintf cope with it. */
1395 SPRINTF_WGINT (buffer, n);
1396 p += strlen (buffer);
1397 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1403 /* -n would overflow. Have sprintf deal with this. */
1404 SPRINTF_WGINT (buffer, n);
1405 p += strlen (buffer);
1413 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1414 way printing any N is fully open-coded without a loop or jump.
1415 (Also see description of DIGITS_*.) */
1417 if (n < 10) DIGITS_1 (1);
1418 else if (n < 100) DIGITS_2 (10);
1419 else if (n < 1000) DIGITS_3 (100);
1420 else if (n < 10000) DIGITS_4 (1000);
1421 else if (n < 100000) DIGITS_5 (10000);
1422 else if (n < 1000000) DIGITS_6 (100000);
1423 else if (n < 10000000) DIGITS_7 (1000000);
1424 else if (n < 100000000) DIGITS_8 (10000000);
1425 else if (n < 1000000000) DIGITS_9 (100000000);
1426 #if SIZEOF_WGINT == 4
1427 /* wgint is 32 bits wide: no number has more than 10 digits. */
1428 else DIGITS_10 (1000000000);
1430 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1431 Constants are constructed by compile-time multiplication to avoid
1432 dealing with different notations for 64-bit constants
1433 (nL/nLL/nI64, depending on the compiler and architecture). */
1434 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1435 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1436 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1437 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1438 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1439 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1440 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1441 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1442 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1443 else DIGITS_19 (1000000000*(W)1000000000);
1447 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1476 /* Print NUMBER to a statically allocated string and return a pointer
1477 to the printed representation.
1479 This function is intended to be used in conjunction with printf.
1480 It is hard to portably print wgint values:
1481 a) you cannot use printf("%ld", number) because wgint can be long
1482 long on 32-bit machines with LFS.
1483 b) you cannot use printf("%lld", number) because NUMBER could be
1484 long on 32-bit machines without LFS, or on 64-bit machines,
1485 which do not require LFS. Also, Windows doesn't support %lld.
1486 c) you cannot use printf("%j", (int_max_t) number) because not all
1487 versions of printf support "%j", the most notable being the one
1489 d) you cannot #define WGINT_FMT to the appropriate format and use
1490 printf(WGINT_FMT, number) because that would break translations
1491 for user-visible messages, such as printf("Downloaded: %d
1494 What you should use instead is printf("%s", number_to_static_string
1497 CAVEAT: since the function returns pointers to static data, you
1498 must be careful to copy its result before calling it again.
1499 However, to make it more useful with printf, the function maintains
1500 an internal ring of static buffers to return. That way things like
1501 printf("%s %s", number_to_static_string (num1),
1502 number_to_static_string (num2)) work as expected. Three buffers
1503 are currently used, which means that "%s %s %s" will work, but "%s
1504 %s %s %s" won't. If you need to print more than three wgints,
1505 bump the RING_SIZE (or rethink your message.) */
1508 number_to_static_string (wgint number)
1510 static char ring[RING_SIZE][24];
1512 char *buf = ring[ringpos];
1513 number_to_string (buf, number);
1514 ringpos = (ringpos + 1) % RING_SIZE;
1518 /* Determine the width of the terminal we're running on. If that's
1519 not possible, return 0. */
1522 determine_screen_width (void)
1524 /* If there's a way to get the terminal size using POSIX
1525 tcgetattr(), somebody please tell me. */
1530 if (opt.lfilename != NULL)
1533 fd = fileno (stderr);
1534 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1535 return 0; /* most likely ENOTTY */
1538 #elif defined(WINDOWS)
1539 CONSOLE_SCREEN_BUFFER_INFO csbi;
1540 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1542 return csbi.dwSize.X;
1543 #else /* neither TIOCGWINSZ nor WINDOWS */
1545 #endif /* neither TIOCGWINSZ nor WINDOWS */
1548 /* Return a random number between 0 and MAX-1, inclusive.
1550 If MAX is greater than the value of RAND_MAX+1 on the system, the
1551 returned value will be in the range [0, RAND_MAX]. This may be
1552 fixed in a future release.
1554 The random number generator is seeded automatically the first time
1557 This uses rand() for portability. It has been suggested that
1558 random() offers better randomness, but this is not required for
1559 Wget, so I chose to go for simplicity and use rand
1562 DO NOT use this for cryptographic purposes. It is only meant to be
1563 used in situations where quality of the random numbers returned
1564 doesn't really matter. */
1567 random_number (int max)
1575 srand (time (NULL));
1580 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1581 and enforce that assumption by masking other bits. */
1583 # define RAND_MAX 32767
1587 /* This is equivalent to rand() % max, but uses the high-order bits
1588 for better randomness on architecture where rand() is implemented
1589 using a simple congruential generator. */
1591 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1592 return (int)bounded;
1595 /* Return a random uniformly distributed floating point number in the
1596 [0, 1) range. The precision of returned numbers is 9 digits.
1598 Modify this to use erand48() where available! */
1603 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1604 sure it's greater than 1000. */
1605 int rnd1 = random_number (1000);
1606 int rnd2 = random_number (1000);
1607 int rnd3 = random_number (1000);
1608 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1611 /* Implementation of run_with_timeout, a generic timeout-forcing
1612 routine for systems with Unix-like signal handling. */
1614 #ifdef USE_SIGNAL_TIMEOUT
1615 # ifdef HAVE_SIGSETJMP
1616 # define SETJMP(env) sigsetjmp (env, 1)
1618 static sigjmp_buf run_with_timeout_env;
1621 abort_run_with_timeout (int sig)
1623 assert (sig == SIGALRM);
1624 siglongjmp (run_with_timeout_env, -1);
1626 # else /* not HAVE_SIGSETJMP */
1627 # define SETJMP(env) setjmp (env)
1629 static jmp_buf run_with_timeout_env;
1632 abort_run_with_timeout (int sig)
1634 assert (sig == SIGALRM);
1635 /* We don't have siglongjmp to preserve the set of blocked signals;
1636 if we longjumped out of the handler at this point, SIGALRM would
1637 remain blocked. We must unblock it manually. */
1638 int mask = siggetmask ();
1639 mask &= ~sigmask (SIGALRM);
1642 /* Now it's safe to longjump. */
1643 longjmp (run_with_timeout_env, -1);
1645 # endif /* not HAVE_SIGSETJMP */
1647 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1648 setitimer where available, alarm otherwise.
1650 TIMEOUT should be non-zero. If the timeout value is so small that
1651 it would be rounded to zero, it is rounded to the least legal value
1652 instead (1us for setitimer, 1s for alarm). That ensures that
1653 SIGALRM will be delivered in all cases. */
1656 alarm_set (double timeout)
1659 /* Use the modern itimer interface. */
1660 struct itimerval itv;
1662 itv.it_value.tv_sec = (long) timeout;
1663 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1664 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1665 /* Ensure that we wait for at least the minimum interval.
1666 Specifying zero would mean "wait forever". */
1667 itv.it_value.tv_usec = 1;
1668 setitimer (ITIMER_REAL, &itv, NULL);
1669 #else /* not ITIMER_REAL */
1670 /* Use the old alarm() interface. */
1671 int secs = (int) timeout;
1673 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1674 because alarm(0) means "never deliver the alarm", i.e. "wait
1675 forever", which is not what someone who specifies a 0.5s
1676 timeout would expect. */
1679 #endif /* not ITIMER_REAL */
1682 /* Cancel the alarm set with alarm_set. */
1688 struct itimerval disable;
1690 setitimer (ITIMER_REAL, &disable, NULL);
1691 #else /* not ITIMER_REAL */
1693 #endif /* not ITIMER_REAL */
1696 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1697 seconds. Returns true if the function was interrupted with a
1698 timeout, false otherwise.
1700 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1701 using setitimer() or alarm(). The timeout is enforced by
1702 longjumping out of the SIGALRM handler. This has several
1703 advantages compared to the traditional approach of relying on
1704 signals causing system calls to exit with EINTR:
1706 * The callback function is *forcibly* interrupted after the
1707 timeout expires, (almost) regardless of what it was doing and
1708 whether it was in a syscall. For example, a calculation that
1709 takes a long time is interrupted as reliably as an IO
1712 * It works with both SYSV and BSD signals because it doesn't
1713 depend on the default setting of SA_RESTART.
1715 * It doesn't require special handler setup beyond a simple call
1716 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1719 The only downside is that, if FUN allocates internal resources that
1720 are normally freed prior to exit from the functions, they will be
1721 lost in case of timeout. */
1724 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1734 signal (SIGALRM, abort_run_with_timeout);
1735 if (SETJMP (run_with_timeout_env) != 0)
1737 /* Longjumped out of FUN with a timeout. */
1738 signal (SIGALRM, SIG_DFL);
1741 alarm_set (timeout);
1744 /* Preserve errno in case alarm() or signal() modifies it. */
1745 saved_errno = errno;
1747 signal (SIGALRM, SIG_DFL);
1748 errno = saved_errno;
1753 #else /* not USE_SIGNAL_TIMEOUT */
1756 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1757 define it under Windows, because Windows has its own version of
1758 run_with_timeout that uses threads. */
1761 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1766 #endif /* not WINDOWS */
1767 #endif /* not USE_SIGNAL_TIMEOUT */
1771 /* Sleep the specified amount of seconds. On machines without
1772 nanosleep(), this may sleep shorter if interrupted by signals. */
1775 xsleep (double seconds)
1777 #ifdef HAVE_NANOSLEEP
1778 /* nanosleep is the preferred interface because it offers high
1779 accuracy and, more importantly, because it allows us to reliably
1780 restart receiving a signal such as SIGWINCH. (There was an
1781 actual Debian bug report about --limit-rate malfunctioning while
1782 the terminal was being resized.) */
1783 struct timespec sleep, remaining;
1784 sleep.tv_sec = (long) seconds;
1785 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1786 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1787 /* If nanosleep has been interrupted by a signal, adjust the
1788 sleeping period and return to sleep. */
1790 #elif defined(HAVE_USLEEP)
1791 /* If usleep is available, use it in preference to select. */
1794 /* On some systems, usleep cannot handle values larger than
1795 1,000,000. If the period is larger than that, use sleep
1796 first, then add usleep for subsecond accuracy. */
1798 seconds -= (long) seconds;
1800 usleep (seconds * 1000000);
1801 #else /* fall back select */
1802 /* Note that, although Windows supports select, it can't be used to
1803 implement sleeping because Winsock's select doesn't implement
1804 timeout when it is passed NULL pointers for all fd sets. (But it
1805 does under Cygwin, which implements Unix-compatible select.) */
1806 struct timeval sleep;
1807 sleep.tv_sec = (long) seconds;
1808 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1809 select (0, NULL, NULL, NULL, &sleep);
1810 /* If select returns -1 and errno is EINTR, it means we were
1811 interrupted by a signal. But without knowing how long we've
1812 actually slept, we can't return to sleep. Using gettimeofday to
1813 track sleeps is slow and unreliable due to clock skew. */
1817 #endif /* not WINDOWS */
1819 /* Encode the string STR of length LENGTH to base64 format and place it
1820 to B64STORE. The output will be \0-terminated, and must point to a
1821 writable buffer of at least 1+BASE64_LENGTH(length) bytes. It
1822 returns the length of the resulting base64 data, not counting the
1825 This implementation will not emit newlines after 76 characters of
1829 base64_encode (const char *str, int length, char *b64store)
1831 /* Conversion table. */
1832 static char tbl[64] = {
1833 'A','B','C','D','E','F','G','H',
1834 'I','J','K','L','M','N','O','P',
1835 'Q','R','S','T','U','V','W','X',
1836 'Y','Z','a','b','c','d','e','f',
1837 'g','h','i','j','k','l','m','n',
1838 'o','p','q','r','s','t','u','v',
1839 'w','x','y','z','0','1','2','3',
1840 '4','5','6','7','8','9','+','/'
1843 const unsigned char *s = (const unsigned char *) str;
1846 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1847 for (i = 0; i < length; i += 3)
1849 *p++ = tbl[s[0] >> 2];
1850 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1851 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1852 *p++ = tbl[s[2] & 0x3f];
1856 /* Pad the result if necessary... */
1857 if (i == length + 1)
1859 else if (i == length + 2)
1860 *(p - 1) = *(p - 2) = '=';
1862 /* ...and zero-terminate it. */
1865 return p - b64store;
1868 #define IS_ASCII(c) (((c) & 0x80) == 0)
1869 #define IS_BASE64(c) ((IS_ASCII (c) && base64_char_to_value[c] >= 0) || c == '=')
1871 /* Get next character from the string, except that non-base64
1872 characters are ignored, as mandated by rfc2045. */
1873 #define NEXT_BASE64_CHAR(c, p) do { \
1875 } while (c != '\0' && !IS_BASE64 (c))
1877 /* Decode data from BASE64 (assumed to be encoded as base64) into
1878 memory pointed to by TO. TO should be large enough to accomodate
1879 the decoded data, which is guaranteed to be less than
1882 Since TO is assumed to contain binary data, it is not
1883 NUL-terminated. The function returns the length of the data
1884 written to TO. -1 is returned in case of error caused by malformed
1888 base64_decode (const char *base64, char *to)
1890 /* Table of base64 values for first 128 characters. Note that this
1891 assumes ASCII (but so does Wget in other places). */
1892 static short base64_char_to_value[128] =
1894 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1895 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1896 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1897 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1898 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1899 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1900 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1901 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1902 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1903 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1904 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1905 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1906 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1909 const char *p = base64;
1915 unsigned long value;
1917 /* Process first byte of a quadruplet. */
1918 NEXT_BASE64_CHAR (c, p);
1922 return -1; /* illegal '=' while decoding base64 */
1923 value = base64_char_to_value[c] << 18;
1925 /* Process scond byte of a quadruplet. */
1926 NEXT_BASE64_CHAR (c, p);
1928 return -1; /* premature EOF while decoding base64 */
1930 return -1; /* illegal `=' while decoding base64 */
1931 value |= base64_char_to_value[c] << 12;
1934 /* Process third byte of a quadruplet. */
1935 NEXT_BASE64_CHAR (c, p);
1937 return -1; /* premature EOF while decoding base64 */
1941 NEXT_BASE64_CHAR (c, p);
1943 return -1; /* premature EOF while decoding base64 */
1945 return -1; /* padding `=' expected but not found */
1949 value |= base64_char_to_value[c] << 6;
1950 *q++ = 0xff & value >> 8;
1952 /* Process fourth byte of a quadruplet. */
1953 NEXT_BASE64_CHAR (c, p);
1955 return -1; /* premature EOF while decoding base64 */
1959 value |= base64_char_to_value[c];
1960 *q++ = 0xff & value;
1968 #undef NEXT_BASE64_CHAR
1970 /* Simple merge sort for use by stable_sort. Implementation courtesy
1971 Zeljko Vrba with additional debugging by Nenad Barbutov. */
1974 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
1975 int (*cmpfun) (const void *, const void *))
1977 #define ELT(array, pos) ((char *)(array) + (pos) * size)
1981 size_t mid = (to + from) / 2;
1982 mergesort_internal (base, temp, size, from, mid, cmpfun);
1983 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
1986 for (k = from; (i <= mid) && (j <= to); k++)
1987 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
1988 memcpy (ELT (temp, k), ELT (base, i++), size);
1990 memcpy (ELT (temp, k), ELT (base, j++), size);
1992 memcpy (ELT (temp, k++), ELT (base, i++), size);
1994 memcpy (ELT (temp, k++), ELT (base, j++), size);
1995 for (k = from; k <= to; k++)
1996 memcpy (ELT (base, k), ELT (temp, k), size);
2001 /* Stable sort with interface exactly like standard library's qsort.
2002 Uses mergesort internally, allocating temporary storage with
2006 stable_sort (void *base, size_t nmemb, size_t size,
2007 int (*cmpfun) (const void *, const void *))
2011 void *temp = alloca (nmemb * size * sizeof (void *));
2012 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);