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() */
63 /* For TIOCGWINSZ and friends: */
64 #ifdef HAVE_SYS_IOCTL_H
65 # include <sys/ioctl.h>
71 /* Needed for run_with_timeout. */
77 #ifndef HAVE_SIGSETJMP
78 /* If sigsetjmp is a macro, configure won't pick it up. */
80 # define HAVE_SIGSETJMP
84 #undef USE_SIGNAL_TIMEOUT
85 #if defined(HAVE_SIGSETJMP) || defined(HAVE_SIGBLOCK)
86 # define USE_SIGNAL_TIMEOUT
93 /* Utility function: like xstrdup(), but also lowercases S. */
96 xstrdup_lower (const char *s)
98 char *copy = xstrdup (s);
105 /* Copy the string formed by two pointers (one on the beginning, other
106 on the char after the last char) to a new, malloc-ed location.
109 strdupdelim (const char *beg, const char *end)
111 char *res = xmalloc (end - beg + 1);
112 memcpy (res, beg, end - beg);
113 res[end - beg] = '\0';
117 /* Parse a string containing comma-separated elements, and return a
118 vector of char pointers with the elements. Spaces following the
119 commas are ignored. */
121 sepstring (const char *s)
135 res = xrealloc (res, (i + 2) * sizeof (char *));
136 res[i] = strdupdelim (p, s);
139 /* Skip the blanks following the ','. */
147 res = xrealloc (res, (i + 2) * sizeof (char *));
148 res[i] = strdupdelim (p, s);
153 /* Like sprintf, but allocates a string of sufficient size with malloc
154 and returns it. GNU libc has a similar function named asprintf,
155 which requires the pointer to the string to be passed. */
158 aprintf (const char *fmt, ...)
160 /* This function is implemented using vsnprintf, which we provide
161 for the systems that don't have it. Therefore, it should be 100%
165 char *str = xmalloc (size);
172 /* See log_vprintf_internal for explanation why it's OK to rely
173 on the return value of vsnprintf. */
175 va_start (args, fmt);
176 n = vsnprintf (str, size, fmt, args);
179 /* If the printing worked, return the string. */
180 if (n > -1 && n < size)
183 /* Else try again with a larger buffer. */
184 if (n > -1) /* C99 */
185 size = n + 1; /* precisely what is needed */
187 size <<= 1; /* twice the old size */
188 str = xrealloc (str, size);
192 /* Concatenate the NULL-terminated list of string arguments into
193 freshly allocated space. */
196 concat_strings (const char *str0, ...)
199 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
202 const char *next_str;
203 int total_length = 0;
206 /* Calculate the length of and allocate the resulting string. */
209 va_start (args, str0);
210 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
212 int len = strlen (next_str);
213 if (argcount < countof (saved_lengths))
214 saved_lengths[argcount++] = len;
218 p = ret = xmalloc (total_length + 1);
220 /* Copy the strings into the allocated space. */
223 va_start (args, str0);
224 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
227 if (argcount < countof (saved_lengths))
228 len = saved_lengths[argcount++];
230 len = strlen (next_str);
231 memcpy (p, next_str, len);
240 /* Return pointer to a static char[] buffer in which zero-terminated
241 string-representation of TM (in form hh:mm:ss) is printed.
243 If TM is NULL, the current time will be used. */
246 time_str (time_t *tm)
248 static char output[15];
250 time_t secs = tm ? *tm : time (NULL);
254 /* In case of error, return the empty string. Maybe we should
255 just abort if this happens? */
259 ptm = localtime (&secs);
260 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
264 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
267 datetime_str (time_t *tm)
269 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
271 time_t secs = tm ? *tm : time (NULL);
275 /* In case of error, return the empty string. Maybe we should
276 just abort if this happens? */
280 ptm = localtime (&secs);
281 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
282 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
283 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
287 /* The Windows versions of the following two functions are defined in
292 fork_to_background (void)
295 /* Whether we arrange our own version of opt.lfilename here. */
296 bool logfile_changed = false;
300 /* We must create the file immediately to avoid either a race
301 condition (which arises from using unique_name and failing to
302 use fopen_excl) or lying to the user about the log file name
303 (which arises from using unique_name, printing the name, and
304 using fopen_excl later on.) */
305 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
308 logfile_changed = true;
321 /* parent, no error */
322 printf (_("Continuing in background, pid %d.\n"), (int) pid);
324 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
325 exit (0); /* #### should we use _exit()? */
328 /* child: give up the privileges and keep running. */
330 freopen ("/dev/null", "r", stdin);
331 freopen ("/dev/null", "w", stdout);
332 freopen ("/dev/null", "w", stderr);
334 #endif /* not WINDOWS */
336 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
337 specified with TM. The atime ("access time") is set to the current
341 touch (const char *file, time_t tm)
343 #ifdef HAVE_STRUCT_UTIMBUF
344 struct utimbuf times;
352 times.actime = time (NULL);
353 if (utime (file, ×) == -1)
354 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
357 /* Checks if FILE is a symbolic link, and removes it if it is. Does
358 nothing under MS-Windows. */
360 remove_link (const char *file)
365 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
367 DEBUGP (("Unlinking %s (symlink).\n", file));
370 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
371 file, strerror (errno));
376 /* Does FILENAME exist? This is quite a lousy implementation, since
377 it supplies no error codes -- only a yes-or-no answer. Thus it
378 will return that a file does not exist if, e.g., the directory is
379 unreadable. I don't mind it too much currently, though. The
380 proper way should, of course, be to have a third, error state,
381 other than true/false, but that would introduce uncalled-for
382 additional complexity to the callers. */
384 file_exists_p (const char *filename)
387 return access (filename, F_OK) >= 0;
390 return stat (filename, &buf) >= 0;
394 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
395 Returns 0 on error. */
397 file_non_directory_p (const char *path)
400 /* Use lstat() rather than stat() so that symbolic links pointing to
401 directories can be identified correctly. */
402 if (lstat (path, &buf) != 0)
404 return S_ISDIR (buf.st_mode) ? false : true;
407 /* Return the size of file named by FILENAME, or -1 if it cannot be
408 opened or seeked into. */
410 file_size (const char *filename)
412 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
414 /* We use fseek rather than stat to determine the file size because
415 that way we can also verify that the file is readable without
416 explicitly checking for permissions. Inspired by the POST patch
418 FILE *fp = fopen (filename, "rb");
421 fseeko (fp, 0, SEEK_END);
427 if (stat (filename, &st) < 0)
433 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
434 doesn't exist is found. Return a freshly allocated copy of the
438 unique_name_1 (const char *prefix)
441 int plen = strlen (prefix);
442 char *template = (char *)alloca (plen + 1 + 24);
443 char *template_tail = template + plen;
445 memcpy (template, prefix, plen);
446 *template_tail++ = '.';
449 number_to_string (template_tail, count++);
450 while (file_exists_p (template));
452 return xstrdup (template);
455 /* Return a unique file name, based on FILE.
457 More precisely, if FILE doesn't exist, it is returned unmodified.
458 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
459 file name that doesn't exist is returned.
461 The resulting file is not created, only verified that it didn't
462 exist at the point in time when the function was called.
463 Therefore, where security matters, don't rely that the file created
464 by this function exists until you open it with O_EXCL or
467 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
468 string. Otherwise, it may return FILE if the file doesn't exist
469 (and therefore doesn't need changing). */
472 unique_name (const char *file, bool allow_passthrough)
474 /* If the FILE itself doesn't exist, return it without
476 if (!file_exists_p (file))
477 return allow_passthrough ? (char *)file : xstrdup (file);
479 /* Otherwise, find a numeric suffix that results in unused file name
481 return unique_name_1 (file);
484 /* Create a file based on NAME, except without overwriting an existing
485 file with that name. Providing O_EXCL is correctly implemented,
486 this function does not have the race condition associated with
487 opening the file returned by unique_name. */
490 unique_create (const char *name, bool binary, char **opened_name)
492 /* unique file name, based on NAME */
493 char *uname = unique_name (name, false);
495 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
498 uname = unique_name (name, false);
500 if (opened_name && fp != NULL)
503 *opened_name = uname;
515 /* Open the file for writing, with the addition that the file is
516 opened "exclusively". This means that, if the file already exists,
517 this function will *fail* and errno will be set to EEXIST. If
518 BINARY is set, the file will be opened in binary mode, equivalent
521 If opening the file fails for any reason, including the file having
522 previously existed, this function returns NULL and sets errno
526 fopen_excl (const char *fname, bool binary)
530 int flags = O_WRONLY | O_CREAT | O_EXCL;
535 fd = open (fname, flags, 0666);
538 return fdopen (fd, binary ? "wb" : "w");
539 #else /* not O_EXCL */
540 /* Manually check whether the file exists. This is prone to race
541 conditions, but systems without O_EXCL haven't deserved
543 if (file_exists_p (fname))
548 return fopen (fname, binary ? "wb" : "w");
549 #endif /* not O_EXCL */
552 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
553 are missing, create them first. In case any mkdir() call fails,
554 return its error status. Returns 0 on successful completion.
556 The behaviour of this function should be identical to the behaviour
557 of `mkdir -p' on systems where mkdir supports the `-p' option. */
559 make_directory (const char *directory)
561 int i, ret, quit = 0;
564 /* Make a copy of dir, to be able to write to it. Otherwise, the
565 function is unsafe if called with a read-only char *argument. */
566 STRDUP_ALLOCA (dir, directory);
568 /* If the first character of dir is '/', skip it (and thus enable
569 creation of absolute-pathname directories. */
570 for (i = (*dir == '/'); 1; ++i)
572 for (; dir[i] && dir[i] != '/'; i++)
577 /* Check whether the directory already exists. Allow creation of
578 of intermediate directories to fail, as the initial path components
579 are not necessarily directories! */
580 if (!file_exists_p (dir))
581 ret = mkdir (dir, 0777);
592 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
593 should be a file name.
595 file_merge("/foo/bar", "baz") => "/foo/baz"
596 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
597 file_merge("foo", "bar") => "bar"
599 In other words, it's a simpler and gentler version of uri_merge_1. */
602 file_merge (const char *base, const char *file)
605 const char *cut = (const char *)strrchr (base, '/');
608 return xstrdup (file);
610 result = xmalloc (cut - base + 1 + strlen (file) + 1);
611 memcpy (result, base, cut - base);
612 result[cut - base] = '/';
613 strcpy (result + (cut - base) + 1, file);
618 static bool in_acclist (const char *const *, const char *, bool);
620 /* Determine whether a file is acceptable to be followed, according to
621 lists of patterns to accept/reject. */
623 acceptable (const char *s)
627 while (l && s[l] != '/')
634 return (in_acclist ((const char *const *)opt.accepts, s, true)
635 && !in_acclist ((const char *const *)opt.rejects, s, true));
637 return in_acclist ((const char *const *)opt.accepts, s, true);
639 else if (opt.rejects)
640 return !in_acclist ((const char *const *)opt.rejects, s, true);
644 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
645 `/something', frontcmp() will return 1 only if S2 begins with
646 `/something'. Otherwise, 0 is returned. */
648 frontcmp (const char *s1, const char *s2)
650 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
654 /* Iterate through STRLIST, and return the first element that matches
655 S, through wildcards or front comparison (as appropriate). */
657 proclist (char **strlist, const char *s, enum accd flags)
660 for (x = strlist; *x; x++)
662 /* Remove leading '/' if ALLABS */
663 char *p = *x + ((flags & ALLABS) && (**x == '/'));
664 if (has_wildcards_p (p))
666 if (fnmatch (p, s, FNM_PATHNAME) == 0)
678 /* Returns whether DIRECTORY is acceptable for download, wrt the
679 include/exclude lists.
681 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
682 and absolute paths may be freely intermixed. */
684 accdir (const char *directory, enum accd flags)
686 /* Remove starting '/'. */
687 if (flags & ALLABS && *directory == '/')
691 if (!proclist (opt.includes, directory, flags))
696 if (proclist (opt.excludes, directory, flags))
702 /* Return true if STRING ends with TAIL. For instance:
704 match_tail ("abc", "bc", false) -> 1
705 match_tail ("abc", "ab", false) -> 0
706 match_tail ("abc", "abc", false) -> 1
708 If FOLD_CASE is true, the comparison will be case-insensitive. */
711 match_tail (const char *string, const char *tail, bool fold_case)
715 /* We want this to be fast, so we code two loops, one with
716 case-folding, one without. */
720 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
721 if (string[i] != tail[j])
726 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
727 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
731 /* If the tail was exhausted, the match was succesful. */
738 /* Checks whether string S matches each element of ACCEPTS. A list
739 element are matched either with fnmatch() or match_tail(),
740 according to whether the element contains wildcards or not.
742 If the BACKWARD is false, don't do backward comparison -- just compare
745 in_acclist (const char *const *accepts, const char *s, bool backward)
747 for (; *accepts; accepts++)
749 if (has_wildcards_p (*accepts))
751 /* fnmatch returns 0 if the pattern *does* match the
753 if (fnmatch (*accepts, s, 0) == 0)
760 if (match_tail (s, *accepts, 0))
765 if (!strcmp (s, *accepts))
773 /* Return the location of STR's suffix (file extension). Examples:
774 suffix ("foo.bar") -> "bar"
775 suffix ("foo.bar.baz") -> "baz"
776 suffix ("/foo/bar") -> NULL
777 suffix ("/foo.bar/baz") -> NULL */
779 suffix (const char *str)
783 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
787 return (char *)str + i;
792 /* Return true if S contains globbing wildcards (`*', `?', `[' or
796 has_wildcards_p (const char *s)
799 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
804 /* Return true if FNAME ends with a typical HTML suffix. The
805 following (case-insensitive) suffixes are presumed to be HTML
810 ?html (`?' matches one character)
812 #### CAVEAT. This is not necessarily a good indication that FNAME
813 refers to a file that contains HTML! */
815 has_html_suffix_p (const char *fname)
819 if ((suf = suffix (fname)) == NULL)
821 if (!strcasecmp (suf, "html"))
823 if (!strcasecmp (suf, "htm"))
825 if (suf[0] && !strcasecmp (suf + 1, "html"))
830 /* Read a line from FP and return the pointer to freshly allocated
831 storage. The storage space is obtained through malloc() and should
832 be freed with free() when it is no longer needed.
834 The length of the line is not limited, except by available memory.
835 The newline character at the end of line is retained. The line is
836 terminated with a zero character.
838 After end-of-file is encountered without anything being read, NULL
839 is returned. NULL is also returned on error. To distinguish
840 between these two cases, use the stdio function ferror(). */
843 read_whole_line (FILE *fp)
847 char *line = xmalloc (bufsize);
849 while (fgets (line + length, bufsize - length, fp))
851 length += strlen (line + length);
853 /* Possible for example when reading from a binary file where
854 a line begins with \0. */
857 if (line[length - 1] == '\n')
860 /* fgets() guarantees to read the whole line, or to use up the
861 space we've given it. We can double the buffer
864 line = xrealloc (line, bufsize);
866 if (length == 0 || ferror (fp))
871 if (length + 1 < bufsize)
872 /* Relieve the memory from our exponential greediness. We say
873 `length + 1' because the terminating \0 is not included in
874 LENGTH. We don't need to zero-terminate the string ourselves,
875 though, because fgets() does that. */
876 line = xrealloc (line, length + 1);
880 /* Read FILE into memory. A pointer to `struct file_memory' are
881 returned; use struct element `content' to access file contents, and
882 the element `length' to know the file length. `content' is *not*
883 zero-terminated, and you should *not* read or write beyond the [0,
884 length) range of characters.
886 After you are done with the file contents, call read_file_free to
889 Depending on the operating system and the type of file that is
890 being read, read_file() either mmap's the file into memory, or
891 reads the file into the core using read().
893 If file is named "-", fileno(stdin) is used for reading instead.
894 If you want to read from a real file named "-", use "./-" instead. */
897 read_file (const char *file)
900 struct file_memory *fm;
902 bool inhibit_close = false;
904 /* Some magic in the finest tradition of Perl and its kin: if FILE
905 is "-", just use stdin. */
909 inhibit_close = true;
910 /* Note that we don't inhibit mmap() in this case. If stdin is
911 redirected from a regular file, mmap() will still work. */
914 fd = open (file, O_RDONLY);
917 fm = xnew (struct file_memory);
922 if (fstat (fd, &buf) < 0)
924 fm->length = buf.st_size;
925 /* NOTE: As far as I know, the callers of this function never
926 modify the file text. Relying on this would enable us to
927 specify PROT_READ and MAP_SHARED for a marginal gain in
928 efficiency, but at some cost to generality. */
929 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
931 if (fm->content == (char *)MAP_FAILED)
941 /* The most common reason why mmap() fails is that FD does not point
942 to a plain file. However, it's also possible that mmap() doesn't
943 work for a particular type of file. Therefore, whenever mmap()
944 fails, we just fall back to the regular method. */
945 #endif /* HAVE_MMAP */
948 size = 512; /* number of bytes fm->contents can
949 hold at any given time. */
950 fm->content = xmalloc (size);
954 if (fm->length > size / 2)
956 /* #### I'm not sure whether the whole exponential-growth
957 thing makes sense with kernel read. On Linux at least,
958 read() refuses to read more than 4K from a file at a
959 single chunk anyway. But other Unixes might optimize it
960 better, and it doesn't *hurt* anything, so I'm leaving
963 /* Normally, we grow SIZE exponentially to make the number
964 of calls to read() and realloc() logarithmic in relation
965 to file size. However, read() can read an amount of data
966 smaller than requested, and it would be unreasonable to
967 double SIZE every time *something* was read. Therefore,
968 we double SIZE only when the length exceeds half of the
969 entire allocated size. */
971 fm->content = xrealloc (fm->content, size);
973 nread = read (fd, fm->content + fm->length, size - fm->length);
975 /* Successful read. */
986 if (size > fm->length && fm->length != 0)
987 /* Due to exponential growth of fm->content, the allocated region
988 might be much larger than what is actually needed. */
989 fm->content = xrealloc (fm->content, fm->length);
1001 /* Release the resources held by FM. Specifically, this calls
1002 munmap() or xfree() on fm->content, depending whether mmap or
1003 malloc/read were used to read in the file. It also frees the
1004 memory needed to hold the FM structure itself. */
1007 read_file_free (struct file_memory *fm)
1012 munmap (fm->content, fm->length);
1017 xfree (fm->content);
1022 /* Free the pointers in a NULL-terminated vector of pointers, then
1023 free the pointer itself. */
1025 free_vec (char **vec)
1036 /* Append vector V2 to vector V1. The function frees V2 and
1037 reallocates V1 (thus you may not use the contents of neither
1038 pointer after the call). If V1 is NULL, V2 is returned. */
1040 merge_vecs (char **v1, char **v2)
1050 /* To avoid j == 0 */
1055 for (i = 0; v1[i]; i++);
1057 for (j = 0; v2[j]; j++);
1058 /* Reallocate v1. */
1059 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1060 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1065 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1066 is allocated as needed. Return the new value of the vector. */
1069 vec_append (char **vec, const char *str)
1071 int cnt; /* count of vector elements, including
1072 the one we're about to append */
1075 for (cnt = 0; vec[cnt]; cnt++)
1081 /* Reallocate the array to fit the new element and the NULL. */
1082 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1083 /* Append a copy of STR to the vector. */
1084 vec[cnt - 1] = xstrdup (str);
1089 /* Sometimes it's useful to create "sets" of strings, i.e. special
1090 hash tables where you want to store strings as keys and merely
1091 query for their existence. Here is a set of utility routines that
1092 makes that transparent. */
1095 string_set_add (struct hash_table *ht, const char *s)
1097 /* First check whether the set element already exists. If it does,
1098 do nothing so that we don't have to free() the old element and
1099 then strdup() a new one. */
1100 if (hash_table_contains (ht, s))
1103 /* We use "1" as value. It provides us a useful and clear arbitrary
1104 value, and it consumes no memory -- the pointers to the same
1105 string "1" will be shared by all the key-value pairs in all `set'
1107 hash_table_put (ht, xstrdup (s), "1");
1110 /* Synonym for hash_table_contains... */
1113 string_set_contains (struct hash_table *ht, const char *s)
1115 return hash_table_contains (ht, s);
1119 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1121 char ***arrayptr = (char ***) arg;
1122 *(*arrayptr)++ = (char *) key;
1126 /* Convert the specified string set to array. ARRAY should be large
1127 enough to hold hash_table_count(ht) char pointers. */
1129 void string_set_to_array (struct hash_table *ht, char **array)
1131 hash_table_map (ht, string_set_to_array_mapper, &array);
1135 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1142 string_set_free (struct hash_table *ht)
1144 hash_table_map (ht, string_set_free_mapper, NULL);
1145 hash_table_destroy (ht);
1149 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1156 /* Another utility function: call free() on all keys and values of HT. */
1159 free_keys_and_values (struct hash_table *ht)
1161 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1165 /* Get grouping data, the separator and grouping info, by calling
1166 localeconv(). The information is cached after the first call to
1169 In locales that don't set a thousand separator (such as the "C"
1170 locale), this forces it to be ",". We are now only showing
1171 thousand separators in one place, so this shouldn't be a problem in
1175 get_grouping_data (const char **sep, const char **grouping)
1177 static const char *cached_sep;
1178 static const char *cached_grouping;
1179 static bool initialized;
1182 /* Get the grouping info from the locale. */
1183 struct lconv *lconv = localeconv ();
1184 cached_sep = lconv->thousands_sep;
1185 cached_grouping = lconv->grouping;
1188 /* Many locales (such as "C" or "hr_HR") don't specify
1189 grouping, which we still want to use it for legibility.
1190 In those locales set the sep char to ',', unless that
1191 character is used for decimal point, in which case set it
1193 if (*lconv->decimal_point != ',')
1197 cached_grouping = "\x03";
1202 *grouping = cached_grouping;
1205 /* Return a printed representation of N with thousand separators.
1206 This should respect locale settings, with the exception of the "C"
1207 locale which mandates no separator, but we use one anyway.
1209 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1210 the separators because it's too non-portable, and it's hard to test
1211 for this feature at configure time. Besides, it wouldn't work in
1212 the "C" locale, which many Unix users still work in. */
1215 with_thousand_seps (wgint n)
1217 static char outbuf[48];
1218 char *p = outbuf + sizeof outbuf;
1220 /* Info received from locale */
1221 const char *grouping, *sep;
1224 /* State information */
1225 int i = 0, groupsize;
1226 const char *atgroup;
1228 bool negative = n < 0;
1230 /* Initialize grouping data. */
1231 get_grouping_data (&sep, &grouping);
1232 seplen = strlen (sep);
1234 groupsize = *atgroup++;
1236 /* This will overflow on WGINT_MIN, but we're not using this to
1237 print negative numbers anyway. */
1241 /* Write the number into the buffer, backwards, inserting the
1242 separators as necessary. */
1246 *--p = n % 10 + '0';
1250 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1251 if (++i == groupsize)
1256 memcpy (p -= seplen, sep, seplen);
1259 groupsize = *atgroup++;
1268 /* N, a byte quantity, is converted to a human-readable abberviated
1269 form a la sizes printed by `ls -lh'. The result is written to a
1270 static buffer, a pointer to which is returned.
1272 Unlike `with_thousand_seps', this approximates to the nearest unit.
1273 Quoting GNU libit: "Most people visually process strings of 3-4
1274 digits effectively, but longer strings of digits are more prone to
1275 misinterpretation. Hence, converting to an abbreviated form
1276 usually improves readability."
1278 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1279 original computer-related meaning of "powers of 1024". Powers of
1280 1000 would be useless since Wget already displays sizes with
1281 thousand separators. We don't use the "*bibyte" names invented in
1282 1998, and seldom used in practice. Wikipedia's entry on kilobyte
1283 discusses this in some detail. */
1286 human_readable (HR_NUMTYPE n)
1288 /* These suffixes are compatible with those of GNU `ls -lh'. */
1289 static char powers[] =
1291 'K', /* kilobyte, 2^10 bytes */
1292 'M', /* megabyte, 2^20 bytes */
1293 'G', /* gigabyte, 2^30 bytes */
1294 'T', /* terabyte, 2^40 bytes */
1295 'P', /* petabyte, 2^50 bytes */
1296 'E', /* exabyte, 2^60 bytes */
1301 /* If the quantity is smaller than 1K, just print it. */
1304 snprintf (buf, sizeof (buf), "%d", (int) n);
1308 /* Loop over powers, dividing N with 1024 in each iteration. This
1309 works unchanged for all sizes of wgint, while still avoiding
1310 non-portable `long double' arithmetic. */
1311 for (i = 0; i < countof (powers); i++)
1313 /* At each iteration N is greater than the *subsequent* power.
1314 That way N/1024.0 produces a decimal number in the units of
1316 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1318 double val = n / 1024.0;
1319 /* Print values smaller than 10 with one decimal digits, and
1320 others without any decimals. */
1321 snprintf (buf, sizeof (buf), "%.*f%c",
1322 val < 10 ? 1 : 0, val, powers[i]);
1327 return NULL; /* unreached */
1330 /* Count the digits in the provided number. Used to allocate space
1331 when printing numbers. */
1334 numdigit (wgint number)
1338 ++cnt; /* accomodate '-' */
1339 while ((number /= 10) != 0)
1344 #define PR(mask) *p++ = n / (mask) + '0'
1346 /* DIGITS_<D> is used to print a D-digit number and should be called
1347 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1348 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1349 Recursively this continues until DIGITS_1 is invoked. */
1351 #define DIGITS_1(mask) PR (mask)
1352 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1353 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1354 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1355 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1356 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1357 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1358 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1359 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1360 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1362 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1364 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1365 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1366 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1367 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1368 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1369 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1370 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1371 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1372 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1374 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1375 cases and to portably support strange sizes of wgint. Ideally this
1376 would just use "%j" and intmax_t, but many systems don't support
1377 it, so it's used only if nothing else works. */
1378 #if SIZEOF_LONG >= SIZEOF_WGINT
1379 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1380 #elif SIZEOF_LONG_LONG >= SIZEOF_WGINT
1381 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1382 #elif defined(WINDOWS)
1383 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64d", (__int64) (n))
1385 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
1388 /* Shorthand for casting to wgint. */
1391 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1392 `sprintf(buffer, "%lld", (long long) number)', only typically much
1393 faster and portable to machines without long long.
1395 The speedup may make a difference in programs that frequently
1396 convert numbers to strings. Some implementations of sprintf,
1397 particularly the one in GNU libc, have been known to be extremely
1398 slow when converting integers to strings.
1400 Return the pointer to the location where the terminating zero was
1401 printed. (Equivalent to calling buffer+strlen(buffer) after the
1404 BUFFER should be big enough to accept as many bytes as you expect
1405 the number to take up. On machines with 64-bit longs the maximum
1406 needed size is 24 bytes. That includes the digits needed for the
1407 largest 64-bit number, the `-' sign in case it's negative, and the
1408 terminating '\0'. */
1411 number_to_string (char *buffer, wgint number)
1416 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1417 /* We are running in a strange or misconfigured environment. Let
1418 sprintf cope with it. */
1419 SPRINTF_WGINT (buffer, n);
1420 p += strlen (buffer);
1421 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1427 /* -n would overflow. Have sprintf deal with this. */
1428 SPRINTF_WGINT (buffer, n);
1429 p += strlen (buffer);
1437 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1438 way printing any N is fully open-coded without a loop or jump.
1439 (Also see description of DIGITS_*.) */
1441 if (n < 10) DIGITS_1 (1);
1442 else if (n < 100) DIGITS_2 (10);
1443 else if (n < 1000) DIGITS_3 (100);
1444 else if (n < 10000) DIGITS_4 (1000);
1445 else if (n < 100000) DIGITS_5 (10000);
1446 else if (n < 1000000) DIGITS_6 (100000);
1447 else if (n < 10000000) DIGITS_7 (1000000);
1448 else if (n < 100000000) DIGITS_8 (10000000);
1449 else if (n < 1000000000) DIGITS_9 (100000000);
1450 #if SIZEOF_WGINT == 4
1451 /* wgint is 32 bits wide: no number has more than 10 digits. */
1452 else DIGITS_10 (1000000000);
1454 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1455 Constants are constructed by compile-time multiplication to avoid
1456 dealing with different notations for 64-bit constants
1457 (nL/nLL/nI64, depending on the compiler and architecture). */
1458 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1459 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1460 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1461 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1462 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1463 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1464 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1465 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1466 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1467 else DIGITS_19 (1000000000*(W)1000000000);
1471 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1478 #undef SPRINTF_WGINT
1501 /* Print NUMBER to a statically allocated string and return a pointer
1502 to the printed representation.
1504 This function is intended to be used in conjunction with printf.
1505 It is hard to portably print wgint values:
1506 a) you cannot use printf("%ld", number) because wgint can be long
1507 long on 32-bit machines with LFS.
1508 b) you cannot use printf("%lld", number) because NUMBER could be
1509 long on 32-bit machines without LFS, or on 64-bit machines,
1510 which do not require LFS. Also, Windows doesn't support %lld.
1511 c) you cannot use printf("%j", (int_max_t) number) because not all
1512 versions of printf support "%j", the most notable being the one
1514 d) you cannot #define WGINT_FMT to the appropriate format and use
1515 printf(WGINT_FMT, number) because that would break translations
1516 for user-visible messages, such as printf("Downloaded: %d
1519 What you should use instead is printf("%s", number_to_static_string
1522 CAVEAT: since the function returns pointers to static data, you
1523 must be careful to copy its result before calling it again.
1524 However, to make it more useful with printf, the function maintains
1525 an internal ring of static buffers to return. That way things like
1526 printf("%s %s", number_to_static_string (num1),
1527 number_to_static_string (num2)) work as expected. Three buffers
1528 are currently used, which means that "%s %s %s" will work, but "%s
1529 %s %s %s" won't. If you need to print more than three wgints,
1530 bump the RING_SIZE (or rethink your message.) */
1533 number_to_static_string (wgint number)
1535 static char ring[RING_SIZE][24];
1537 char *buf = ring[ringpos];
1538 number_to_string (buf, number);
1539 ringpos = (ringpos + 1) % RING_SIZE;
1543 /* Determine the width of the terminal we're running on. If that's
1544 not possible, return 0. */
1547 determine_screen_width (void)
1549 /* If there's a way to get the terminal size using POSIX
1550 tcgetattr(), somebody please tell me. */
1555 if (opt.lfilename != NULL)
1558 fd = fileno (stderr);
1559 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1560 return 0; /* most likely ENOTTY */
1563 #elif defined(WINDOWS)
1564 CONSOLE_SCREEN_BUFFER_INFO csbi;
1565 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1567 return csbi.dwSize.X;
1568 #else /* neither TIOCGWINSZ nor WINDOWS */
1570 #endif /* neither TIOCGWINSZ nor WINDOWS */
1573 /* Whether the rnd system (either rand or [dl]rand48) has been
1575 static int rnd_seeded;
1577 /* Return a random number between 0 and MAX-1, inclusive.
1579 If the system does not support lrand48 and MAX is greater than the
1580 value of RAND_MAX+1 on the system, the returned value will be in
1581 the range [0, RAND_MAX]. This may be fixed in a future release.
1582 The random number generator is seeded automatically the first time
1585 This uses lrand48 where available, rand elsewhere. DO NOT use it
1586 for cryptography. It is only meant to be used in situations where
1587 quality of the random numbers returned doesn't really matter. */
1590 random_number (int max)
1595 srand48 ((long) time (NULL) ^ (long) getpid ());
1598 return lrand48 () % max;
1599 #else /* not HAVE_DRAND48 */
1605 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1610 /* Like rand() % max, but uses the high-order bits for better
1611 randomness on architectures where rand() is implemented using a
1612 simple congruential generator. */
1614 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1615 return (int) bounded;
1617 #endif /* not HAVE_DRAND48 */
1620 /* Return a random uniformly distributed floating point number in the
1621 [0, 1) range. Uses drand48 where available, and a really lame
1622 kludge elsewhere. */
1630 srand48 ((long) time (NULL) ^ (long) getpid ());
1634 #else /* not HAVE_DRAND48 */
1635 return ( random_number (10000) / 10000.0
1636 + random_number (10000) / (10000.0 * 10000.0)
1637 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1638 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1639 #endif /* not HAVE_DRAND48 */
1642 /* Implementation of run_with_timeout, a generic timeout-forcing
1643 routine for systems with Unix-like signal handling. */
1645 #ifdef USE_SIGNAL_TIMEOUT
1646 # ifdef HAVE_SIGSETJMP
1647 # define SETJMP(env) sigsetjmp (env, 1)
1649 static sigjmp_buf run_with_timeout_env;
1652 abort_run_with_timeout (int sig)
1654 assert (sig == SIGALRM);
1655 siglongjmp (run_with_timeout_env, -1);
1657 # else /* not HAVE_SIGSETJMP */
1658 # define SETJMP(env) setjmp (env)
1660 static jmp_buf run_with_timeout_env;
1663 abort_run_with_timeout (int sig)
1665 assert (sig == SIGALRM);
1666 /* We don't have siglongjmp to preserve the set of blocked signals;
1667 if we longjumped out of the handler at this point, SIGALRM would
1668 remain blocked. We must unblock it manually. */
1669 int mask = siggetmask ();
1670 mask &= ~sigmask (SIGALRM);
1673 /* Now it's safe to longjump. */
1674 longjmp (run_with_timeout_env, -1);
1676 # endif /* not HAVE_SIGSETJMP */
1678 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1679 setitimer where available, alarm otherwise.
1681 TIMEOUT should be non-zero. If the timeout value is so small that
1682 it would be rounded to zero, it is rounded to the least legal value
1683 instead (1us for setitimer, 1s for alarm). That ensures that
1684 SIGALRM will be delivered in all cases. */
1687 alarm_set (double timeout)
1690 /* Use the modern itimer interface. */
1691 struct itimerval itv;
1693 itv.it_value.tv_sec = (long) timeout;
1694 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1695 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1696 /* Ensure that we wait for at least the minimum interval.
1697 Specifying zero would mean "wait forever". */
1698 itv.it_value.tv_usec = 1;
1699 setitimer (ITIMER_REAL, &itv, NULL);
1700 #else /* not ITIMER_REAL */
1701 /* Use the old alarm() interface. */
1702 int secs = (int) timeout;
1704 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1705 because alarm(0) means "never deliver the alarm", i.e. "wait
1706 forever", which is not what someone who specifies a 0.5s
1707 timeout would expect. */
1710 #endif /* not ITIMER_REAL */
1713 /* Cancel the alarm set with alarm_set. */
1719 struct itimerval disable;
1721 setitimer (ITIMER_REAL, &disable, NULL);
1722 #else /* not ITIMER_REAL */
1724 #endif /* not ITIMER_REAL */
1727 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1728 seconds. Returns true if the function was interrupted with a
1729 timeout, false otherwise.
1731 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1732 using setitimer() or alarm(). The timeout is enforced by
1733 longjumping out of the SIGALRM handler. This has several
1734 advantages compared to the traditional approach of relying on
1735 signals causing system calls to exit with EINTR:
1737 * The callback function is *forcibly* interrupted after the
1738 timeout expires, (almost) regardless of what it was doing and
1739 whether it was in a syscall. For example, a calculation that
1740 takes a long time is interrupted as reliably as an IO
1743 * It works with both SYSV and BSD signals because it doesn't
1744 depend on the default setting of SA_RESTART.
1746 * It doesn't require special handler setup beyond a simple call
1747 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1750 The only downside is that, if FUN allocates internal resources that
1751 are normally freed prior to exit from the functions, they will be
1752 lost in case of timeout. */
1755 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1765 signal (SIGALRM, abort_run_with_timeout);
1766 if (SETJMP (run_with_timeout_env) != 0)
1768 /* Longjumped out of FUN with a timeout. */
1769 signal (SIGALRM, SIG_DFL);
1772 alarm_set (timeout);
1775 /* Preserve errno in case alarm() or signal() modifies it. */
1776 saved_errno = errno;
1778 signal (SIGALRM, SIG_DFL);
1779 errno = saved_errno;
1784 #else /* not USE_SIGNAL_TIMEOUT */
1787 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1788 define it under Windows, because Windows has its own version of
1789 run_with_timeout that uses threads. */
1792 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1797 #endif /* not WINDOWS */
1798 #endif /* not USE_SIGNAL_TIMEOUT */
1802 /* Sleep the specified amount of seconds. On machines without
1803 nanosleep(), this may sleep shorter if interrupted by signals. */
1806 xsleep (double seconds)
1808 #ifdef HAVE_NANOSLEEP
1809 /* nanosleep is the preferred interface because it offers high
1810 accuracy and, more importantly, because it allows us to reliably
1811 restart receiving a signal such as SIGWINCH. (There was an
1812 actual Debian bug report about --limit-rate malfunctioning while
1813 the terminal was being resized.) */
1814 struct timespec sleep, remaining;
1815 sleep.tv_sec = (long) seconds;
1816 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1817 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1818 /* If nanosleep has been interrupted by a signal, adjust the
1819 sleeping period and return to sleep. */
1821 #elif defined(HAVE_USLEEP)
1822 /* If usleep is available, use it in preference to select. */
1825 /* On some systems, usleep cannot handle values larger than
1826 1,000,000. If the period is larger than that, use sleep
1827 first, then add usleep for subsecond accuracy. */
1829 seconds -= (long) seconds;
1831 usleep (seconds * 1000000);
1832 #else /* fall back select */
1833 /* Note that, although Windows supports select, it can't be used to
1834 implement sleeping because Winsock's select doesn't implement
1835 timeout when it is passed NULL pointers for all fd sets. (But it
1836 does under Cygwin, which implements Unix-compatible select.) */
1837 struct timeval sleep;
1838 sleep.tv_sec = (long) seconds;
1839 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1840 select (0, NULL, NULL, NULL, &sleep);
1841 /* If select returns -1 and errno is EINTR, it means we were
1842 interrupted by a signal. But without knowing how long we've
1843 actually slept, we can't return to sleep. Using gettimeofday to
1844 track sleeps is slow and unreliable due to clock skew. */
1848 #endif /* not WINDOWS */
1850 /* Encode the string STR of length LENGTH to base64 format and place it
1851 to B64STORE. The output will be \0-terminated, and must point to a
1852 writable buffer of at least 1+BASE64_LENGTH(length) bytes. It
1853 returns the length of the resulting base64 data, not counting the
1856 This implementation will not emit newlines after 76 characters of
1860 base64_encode (const char *str, int length, char *b64store)
1862 /* Conversion table. */
1863 static char tbl[64] = {
1864 'A','B','C','D','E','F','G','H',
1865 'I','J','K','L','M','N','O','P',
1866 'Q','R','S','T','U','V','W','X',
1867 'Y','Z','a','b','c','d','e','f',
1868 'g','h','i','j','k','l','m','n',
1869 'o','p','q','r','s','t','u','v',
1870 'w','x','y','z','0','1','2','3',
1871 '4','5','6','7','8','9','+','/'
1874 const unsigned char *s = (const unsigned char *) str;
1877 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1878 for (i = 0; i < length; i += 3)
1880 *p++ = tbl[s[0] >> 2];
1881 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1882 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1883 *p++ = tbl[s[2] & 0x3f];
1887 /* Pad the result if necessary... */
1888 if (i == length + 1)
1890 else if (i == length + 2)
1891 *(p - 1) = *(p - 2) = '=';
1893 /* ...and zero-terminate it. */
1896 return p - b64store;
1899 #define IS_ASCII(c) (((c) & 0x80) == 0)
1900 #define IS_BASE64(c) ((IS_ASCII (c) && base64_char_to_value[c] >= 0) || c == '=')
1902 /* Get next character from the string, except that non-base64
1903 characters are ignored, as mandated by rfc2045. */
1904 #define NEXT_BASE64_CHAR(c, p) do { \
1906 } while (c != '\0' && !IS_BASE64 (c))
1908 /* Decode data from BASE64 (assumed to be encoded as base64) into
1909 memory pointed to by TO. TO should be large enough to accomodate
1910 the decoded data, which is guaranteed to be less than
1913 Since TO is assumed to contain binary data, it is not
1914 NUL-terminated. The function returns the length of the data
1915 written to TO. -1 is returned in case of error caused by malformed
1919 base64_decode (const char *base64, char *to)
1921 /* Table of base64 values for first 128 characters. Note that this
1922 assumes ASCII (but so does Wget in other places). */
1923 static short base64_char_to_value[128] =
1925 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1926 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1927 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1928 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1929 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1930 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1931 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1932 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1933 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1934 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1935 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1936 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1937 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1940 const char *p = base64;
1946 unsigned long value;
1948 /* Process first byte of a quadruplet. */
1949 NEXT_BASE64_CHAR (c, p);
1953 return -1; /* illegal '=' while decoding base64 */
1954 value = base64_char_to_value[c] << 18;
1956 /* Process scond byte of a quadruplet. */
1957 NEXT_BASE64_CHAR (c, p);
1959 return -1; /* premature EOF while decoding base64 */
1961 return -1; /* illegal `=' while decoding base64 */
1962 value |= base64_char_to_value[c] << 12;
1965 /* Process third byte of a quadruplet. */
1966 NEXT_BASE64_CHAR (c, p);
1968 return -1; /* premature EOF while decoding base64 */
1972 NEXT_BASE64_CHAR (c, p);
1974 return -1; /* premature EOF while decoding base64 */
1976 return -1; /* padding `=' expected but not found */
1980 value |= base64_char_to_value[c] << 6;
1981 *q++ = 0xff & value >> 8;
1983 /* Process fourth byte of a quadruplet. */
1984 NEXT_BASE64_CHAR (c, p);
1986 return -1; /* premature EOF while decoding base64 */
1990 value |= base64_char_to_value[c];
1991 *q++ = 0xff & value;
1999 #undef NEXT_BASE64_CHAR
2001 /* Simple merge sort for use by stable_sort. Implementation courtesy
2002 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2005 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2006 int (*cmpfun) (const void *, const void *))
2008 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2012 size_t mid = (to + from) / 2;
2013 mergesort_internal (base, temp, size, from, mid, cmpfun);
2014 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2017 for (k = from; (i <= mid) && (j <= to); k++)
2018 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2019 memcpy (ELT (temp, k), ELT (base, i++), size);
2021 memcpy (ELT (temp, k), ELT (base, j++), size);
2023 memcpy (ELT (temp, k++), ELT (base, i++), size);
2025 memcpy (ELT (temp, k++), ELT (base, j++), size);
2026 for (k = from; k <= to; k++)
2027 memcpy (ELT (base, k), ELT (temp, k), size);
2032 /* Stable sort with interface exactly like standard library's qsort.
2033 Uses mergesort internally, allocating temporary storage with
2037 stable_sort (void *base, size_t nmemb, size_t size,
2038 int (*cmpfun) (const void *, const void *))
2042 void *temp = alloca (nmemb * size * sizeof (void *));
2043 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2047 /* Print a decimal number. If it is equal to or larger than ten, the
2048 number is rounded. Otherwise it is printed with one significant
2049 digit without trailing zeros and with no more than three fractional
2050 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2051 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2053 This is useful for displaying durations because it provides
2054 order-of-magnitude information without unnecessary clutter --
2055 long-running downloads are shown without the fractional part, and
2056 short ones still retain one significant digit. */
2059 print_decimal (double number)
2061 static char buf[32];
2062 double n = number >= 0 ? number : -number;
2065 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2066 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2067 snprintf (buf, sizeof buf, "%.0f", number);
2069 snprintf (buf, sizeof buf, "%.1f", number);
2070 else if (n >= 0.001)
2071 snprintf (buf, sizeof buf, "%.1g", number);
2072 else if (n >= 0.0005)
2073 /* round [0.0005, 0.001) to 0.001 */
2074 snprintf (buf, sizeof buf, "%.3f", number);
2076 /* print numbers close to 0 as 0, not 0.000 */