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>
54 #ifdef HAVE_SYS_UTIME_H
55 # include <sys/utime.h>
59 # include <libc.h> /* for access() */
65 /* For TIOCGWINSZ and friends: */
66 #ifdef HAVE_SYS_IOCTL_H
67 # include <sys/ioctl.h>
73 /* Needed for run_with_timeout. */
74 #undef USE_SIGNAL_TIMEOUT
82 #ifndef HAVE_SIGSETJMP
83 /* If sigsetjmp is a macro, configure won't pick it up. */
85 # define HAVE_SIGSETJMP
90 # ifdef HAVE_SIGSETJMP
91 # define USE_SIGNAL_TIMEOUT
94 # define USE_SIGNAL_TIMEOUT
102 /* Utility function: like xstrdup(), but also lowercases S. */
105 xstrdup_lower (const char *s)
107 char *copy = xstrdup (s);
114 /* Copy the string formed by two pointers (one on the beginning, other
115 on the char after the last char) to a new, malloc-ed location.
118 strdupdelim (const char *beg, const char *end)
120 char *res = (char *)xmalloc (end - beg + 1);
121 memcpy (res, beg, end - beg);
122 res[end - beg] = '\0';
126 /* Parse a string containing comma-separated elements, and return a
127 vector of char pointers with the elements. Spaces following the
128 commas are ignored. */
130 sepstring (const char *s)
144 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
145 res[i] = strdupdelim (p, s);
148 /* Skip the blanks following the ','. */
156 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
157 res[i] = strdupdelim (p, s);
162 /* Like sprintf, but allocates a string of sufficient size with malloc
163 and returns it. GNU libc has a similar function named asprintf,
164 which requires the pointer to the string to be passed. */
167 aprintf (const char *fmt, ...)
169 /* This function is implemented using vsnprintf, which we provide
170 for the systems that don't have it. Therefore, it should be 100%
174 char *str = xmalloc (size);
181 /* See log_vprintf_internal for explanation why it's OK to rely
182 on the return value of vsnprintf. */
184 va_start (args, fmt);
185 n = vsnprintf (str, size, fmt, args);
188 /* If the printing worked, return the string. */
189 if (n > -1 && n < size)
192 /* Else try again with a larger buffer. */
193 if (n > -1) /* C99 */
194 size = n + 1; /* precisely what is needed */
196 size <<= 1; /* twice the old size */
197 str = xrealloc (str, size);
201 /* Concatenate the NULL-terminated list of string arguments into
202 freshly allocated space. */
205 concat_strings (const char *str0, ...)
208 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
211 const char *next_str;
212 int total_length = 0;
215 /* Calculate the length of and allocate the resulting string. */
218 va_start (args, str0);
219 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
221 int len = strlen (next_str);
222 if (argcount < countof (saved_lengths))
223 saved_lengths[argcount++] = len;
227 p = ret = xmalloc (total_length + 1);
229 /* Copy the strings into the allocated space. */
232 va_start (args, str0);
233 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
236 if (argcount < countof (saved_lengths))
237 len = saved_lengths[argcount++];
239 len = strlen (next_str);
240 memcpy (p, next_str, len);
249 /* Return pointer to a static char[] buffer in which zero-terminated
250 string-representation of TM (in form hh:mm:ss) is printed.
252 If TM is NULL, the current time will be used. */
255 time_str (time_t *tm)
257 static char output[15];
259 time_t secs = tm ? *tm : time (NULL);
263 /* In case of error, return the empty string. Maybe we should
264 just abort if this happens? */
268 ptm = localtime (&secs);
269 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
273 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
276 datetime_str (time_t *tm)
278 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
280 time_t secs = tm ? *tm : time (NULL);
284 /* In case of error, return the empty string. Maybe we should
285 just abort if this happens? */
289 ptm = localtime (&secs);
290 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
291 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
292 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
296 /* The Windows versions of the following two functions are defined in
301 fork_to_background (void)
304 /* Whether we arrange our own version of opt.lfilename here. */
305 int logfile_changed = 0;
309 /* We must create the file immediately to avoid either a race
310 condition (which arises from using unique_name and failing to
311 use fopen_excl) or lying to the user about the log file name
312 (which arises from using unique_name, printing the name, and
313 using fopen_excl later on.) */
314 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, 0, &opt.lfilename);
330 /* parent, no error */
331 printf (_("Continuing in background, pid %d.\n"), (int)pid);
333 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
334 exit (0); /* #### should we use _exit()? */
337 /* child: give up the privileges and keep running. */
339 freopen ("/dev/null", "r", stdin);
340 freopen ("/dev/null", "w", stdout);
341 freopen ("/dev/null", "w", stderr);
343 #endif /* not WINDOWS */
345 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
346 specified with TM. The atime ("access time") is set to the current
350 touch (const char *file, time_t tm)
352 #ifdef HAVE_STRUCT_UTIMBUF
353 struct utimbuf times;
361 times.actime = time (NULL);
362 if (utime (file, ×) == -1)
363 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
366 /* Checks if FILE is a symbolic link, and removes it if it is. Does
367 nothing under MS-Windows. */
369 remove_link (const char *file)
374 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
376 DEBUGP (("Unlinking %s (symlink).\n", file));
379 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
380 file, strerror (errno));
385 /* Does FILENAME exist? This is quite a lousy implementation, since
386 it supplies no error codes -- only a yes-or-no answer. Thus it
387 will return that a file does not exist if, e.g., the directory is
388 unreadable. I don't mind it too much currently, though. The
389 proper way should, of course, be to have a third, error state,
390 other than true/false, but that would introduce uncalled-for
391 additional complexity to the callers. */
393 file_exists_p (const char *filename)
396 return access (filename, F_OK) >= 0;
399 return stat (filename, &buf) >= 0;
403 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
404 Returns 0 on error. */
406 file_non_directory_p (const char *path)
409 /* Use lstat() rather than stat() so that symbolic links pointing to
410 directories can be identified correctly. */
411 if (lstat (path, &buf) != 0)
413 return S_ISDIR (buf.st_mode) ? 0 : 1;
416 /* Return the size of file named by FILENAME, or -1 if it cannot be
417 opened or seeked into. */
419 file_size (const char *filename)
421 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
423 /* We use fseek rather than stat to determine the file size because
424 that way we can also verify that the file is readable without
425 explicitly checking for permissions. Inspired by the POST patch
427 FILE *fp = fopen (filename, "rb");
430 fseeko (fp, 0, SEEK_END);
436 if (stat (filename, &st) < 0)
442 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
443 doesn't exist is found. Return a freshly allocated copy of the
447 unique_name_1 (const char *prefix)
450 int plen = strlen (prefix);
451 char *template = (char *)alloca (plen + 1 + 24);
452 char *template_tail = template + plen;
454 memcpy (template, prefix, plen);
455 *template_tail++ = '.';
458 number_to_string (template_tail, count++);
459 while (file_exists_p (template));
461 return xstrdup (template);
464 /* Return a unique file name, based on FILE.
466 More precisely, if FILE doesn't exist, it is returned unmodified.
467 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
468 file name that doesn't exist is returned.
470 The resulting file is not created, only verified that it didn't
471 exist at the point in time when the function was called.
472 Therefore, where security matters, don't rely that the file created
473 by this function exists until you open it with O_EXCL or
476 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
477 string. Otherwise, it may return FILE if the file doesn't exist
478 (and therefore doesn't need changing). */
481 unique_name (const char *file, int allow_passthrough)
483 /* If the FILE itself doesn't exist, return it without
485 if (!file_exists_p (file))
486 return allow_passthrough ? (char *)file : xstrdup (file);
488 /* Otherwise, find a numeric suffix that results in unused file name
490 return unique_name_1 (file);
493 /* Create a file based on NAME, except without overwriting an existing
494 file with that name. Providing O_EXCL is correctly implemented,
495 this function does not have the race condition associated with
496 opening the file returned by unique_name. */
499 unique_create (const char *name, int binary, char **opened_name)
501 /* unique file name, based on NAME */
502 char *uname = unique_name (name, 0);
504 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
507 uname = unique_name (name, 0);
509 if (opened_name && fp != NULL)
512 *opened_name = uname;
524 /* Open the file for writing, with the addition that the file is
525 opened "exclusively". This means that, if the file already exists,
526 this function will *fail* and errno will be set to EEXIST. If
527 BINARY is set, the file will be opened in binary mode, equivalent
530 If opening the file fails for any reason, including the file having
531 previously existed, this function returns NULL and sets errno
535 fopen_excl (const char *fname, int binary)
539 int flags = O_WRONLY | O_CREAT | O_EXCL;
544 fd = open (fname, flags, 0666);
547 return fdopen (fd, binary ? "wb" : "w");
548 #else /* not O_EXCL */
549 /* Manually check whether the file exists. This is prone to race
550 conditions, but systems without O_EXCL haven't deserved
552 if (file_exists_p (fname))
557 return fopen (fname, binary ? "wb" : "w");
558 #endif /* not O_EXCL */
561 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
562 are missing, create them first. In case any mkdir() call fails,
563 return its error status. Returns 0 on successful completion.
565 The behaviour of this function should be identical to the behaviour
566 of `mkdir -p' on systems where mkdir supports the `-p' option. */
568 make_directory (const char *directory)
570 int i, ret, quit = 0;
573 /* Make a copy of dir, to be able to write to it. Otherwise, the
574 function is unsafe if called with a read-only char *argument. */
575 STRDUP_ALLOCA (dir, directory);
577 /* If the first character of dir is '/', skip it (and thus enable
578 creation of absolute-pathname directories. */
579 for (i = (*dir == '/'); 1; ++i)
581 for (; dir[i] && dir[i] != '/'; i++)
586 /* Check whether the directory already exists. Allow creation of
587 of intermediate directories to fail, as the initial path components
588 are not necessarily directories! */
589 if (!file_exists_p (dir))
590 ret = mkdir (dir, 0777);
601 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
602 should be a file name.
604 file_merge("/foo/bar", "baz") => "/foo/baz"
605 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
606 file_merge("foo", "bar") => "bar"
608 In other words, it's a simpler and gentler version of uri_merge_1. */
611 file_merge (const char *base, const char *file)
614 const char *cut = (const char *)strrchr (base, '/');
617 return xstrdup (file);
619 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
620 memcpy (result, base, cut - base);
621 result[cut - base] = '/';
622 strcpy (result + (cut - base) + 1, file);
627 static int in_acclist (const char *const *, const char *, int);
629 /* Determine whether a file is acceptable to be followed, according to
630 lists of patterns to accept/reject. */
632 acceptable (const char *s)
636 while (l && s[l] != '/')
643 return (in_acclist ((const char *const *)opt.accepts, s, 1)
644 && !in_acclist ((const char *const *)opt.rejects, s, 1));
646 return in_acclist ((const char *const *)opt.accepts, s, 1);
648 else if (opt.rejects)
649 return !in_acclist ((const char *const *)opt.rejects, s, 1);
653 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
654 `/something', frontcmp() will return 1 only if S2 begins with
655 `/something'. Otherwise, 0 is returned. */
657 frontcmp (const char *s1, const char *s2)
659 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
663 /* Iterate through STRLIST, and return the first element that matches
664 S, through wildcards or front comparison (as appropriate). */
666 proclist (char **strlist, const char *s, enum accd flags)
669 for (x = strlist; *x; x++)
671 /* Remove leading '/' if ALLABS */
672 char *p = *x + ((flags & ALLABS) && (**x == '/'));
673 if (has_wildcards_p (p))
675 if (fnmatch (p, s, FNM_PATHNAME) == 0)
687 /* Returns whether DIRECTORY is acceptable for download, wrt the
688 include/exclude lists.
690 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
691 and absolute paths may be freely intermixed. */
693 accdir (const char *directory, enum accd flags)
695 /* Remove starting '/'. */
696 if (flags & ALLABS && *directory == '/')
700 if (!proclist (opt.includes, directory, flags))
705 if (proclist (opt.excludes, directory, flags))
711 /* Return non-zero if STRING ends with TAIL. For instance:
713 match_tail ("abc", "bc", 0) -> 1
714 match_tail ("abc", "ab", 0) -> 0
715 match_tail ("abc", "abc", 0) -> 1
717 If FOLD_CASE_P is non-zero, the comparison will be
721 match_tail (const char *string, const char *tail, int fold_case_p)
725 /* We want this to be fast, so we code two loops, one with
726 case-folding, one without. */
730 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
731 if (string[i] != tail[j])
736 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
737 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
741 /* If the tail was exhausted, the match was succesful. */
748 /* Checks whether string S matches each element of ACCEPTS. A list
749 element are matched either with fnmatch() or match_tail(),
750 according to whether the element contains wildcards or not.
752 If the BACKWARD is 0, don't do backward comparison -- just compare
755 in_acclist (const char *const *accepts, const char *s, int backward)
757 for (; *accepts; accepts++)
759 if (has_wildcards_p (*accepts))
761 /* fnmatch returns 0 if the pattern *does* match the
763 if (fnmatch (*accepts, s, 0) == 0)
770 if (match_tail (s, *accepts, 0))
775 if (!strcmp (s, *accepts))
783 /* Return the location of STR's suffix (file extension). Examples:
784 suffix ("foo.bar") -> "bar"
785 suffix ("foo.bar.baz") -> "baz"
786 suffix ("/foo/bar") -> NULL
787 suffix ("/foo.bar/baz") -> NULL */
789 suffix (const char *str)
793 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
797 return (char *)str + i;
802 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
806 has_wildcards_p (const char *s)
809 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
814 /* Return non-zero if FNAME ends with a typical HTML suffix. The
815 following (case-insensitive) suffixes are presumed to be HTML files:
819 ?html (`?' matches one character)
821 #### CAVEAT. This is not necessarily a good indication that FNAME
822 refers to a file that contains HTML! */
824 has_html_suffix_p (const char *fname)
828 if ((suf = suffix (fname)) == NULL)
830 if (!strcasecmp (suf, "html"))
832 if (!strcasecmp (suf, "htm"))
834 if (suf[0] && !strcasecmp (suf + 1, "html"))
839 /* Read a line from FP and return the pointer to freshly allocated
840 storage. The storage space is obtained through malloc() and should
841 be freed with free() when it is no longer needed.
843 The length of the line is not limited, except by available memory.
844 The newline character at the end of line is retained. The line is
845 terminated with a zero character.
847 After end-of-file is encountered without anything being read, NULL
848 is returned. NULL is also returned on error. To distinguish
849 between these two cases, use the stdio function ferror(). */
852 read_whole_line (FILE *fp)
856 char *line = (char *)xmalloc (bufsize);
858 while (fgets (line + length, bufsize - length, fp))
860 length += strlen (line + length);
862 /* Possible for example when reading from a binary file where
863 a line begins with \0. */
866 if (line[length - 1] == '\n')
869 /* fgets() guarantees to read the whole line, or to use up the
870 space we've given it. We can double the buffer
873 line = xrealloc (line, bufsize);
875 if (length == 0 || ferror (fp))
880 if (length + 1 < bufsize)
881 /* Relieve the memory from our exponential greediness. We say
882 `length + 1' because the terminating \0 is not included in
883 LENGTH. We don't need to zero-terminate the string ourselves,
884 though, because fgets() does that. */
885 line = xrealloc (line, length + 1);
889 /* Read FILE into memory. A pointer to `struct file_memory' are
890 returned; use struct element `content' to access file contents, and
891 the element `length' to know the file length. `content' is *not*
892 zero-terminated, and you should *not* read or write beyond the [0,
893 length) range of characters.
895 After you are done with the file contents, call read_file_free to
898 Depending on the operating system and the type of file that is
899 being read, read_file() either mmap's the file into memory, or
900 reads the file into the core using read().
902 If file is named "-", fileno(stdin) is used for reading instead.
903 If you want to read from a real file named "-", use "./-" instead. */
906 read_file (const char *file)
909 struct file_memory *fm;
911 int inhibit_close = 0;
913 /* Some magic in the finest tradition of Perl and its kin: if FILE
914 is "-", just use stdin. */
919 /* Note that we don't inhibit mmap() in this case. If stdin is
920 redirected from a regular file, mmap() will still work. */
923 fd = open (file, O_RDONLY);
926 fm = xnew (struct file_memory);
931 if (fstat (fd, &buf) < 0)
933 fm->length = buf.st_size;
934 /* NOTE: As far as I know, the callers of this function never
935 modify the file text. Relying on this would enable us to
936 specify PROT_READ and MAP_SHARED for a marginal gain in
937 efficiency, but at some cost to generality. */
938 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
940 if (fm->content == (char *)MAP_FAILED)
950 /* The most common reason why mmap() fails is that FD does not point
951 to a plain file. However, it's also possible that mmap() doesn't
952 work for a particular type of file. Therefore, whenever mmap()
953 fails, we just fall back to the regular method. */
954 #endif /* HAVE_MMAP */
957 size = 512; /* number of bytes fm->contents can
958 hold at any given time. */
959 fm->content = xmalloc (size);
963 if (fm->length > size / 2)
965 /* #### I'm not sure whether the whole exponential-growth
966 thing makes sense with kernel read. On Linux at least,
967 read() refuses to read more than 4K from a file at a
968 single chunk anyway. But other Unixes might optimize it
969 better, and it doesn't *hurt* anything, so I'm leaving
972 /* Normally, we grow SIZE exponentially to make the number
973 of calls to read() and realloc() logarithmic in relation
974 to file size. However, read() can read an amount of data
975 smaller than requested, and it would be unreasonable to
976 double SIZE every time *something* was read. Therefore,
977 we double SIZE only when the length exceeds half of the
978 entire allocated size. */
980 fm->content = xrealloc (fm->content, size);
982 nread = read (fd, fm->content + fm->length, size - fm->length);
984 /* Successful read. */
995 if (size > fm->length && fm->length != 0)
996 /* Due to exponential growth of fm->content, the allocated region
997 might be much larger than what is actually needed. */
998 fm->content = xrealloc (fm->content, fm->length);
1005 xfree (fm->content);
1010 /* Release the resources held by FM. Specifically, this calls
1011 munmap() or xfree() on fm->content, depending whether mmap or
1012 malloc/read were used to read in the file. It also frees the
1013 memory needed to hold the FM structure itself. */
1016 read_file_free (struct file_memory *fm)
1021 munmap (fm->content, fm->length);
1026 xfree (fm->content);
1031 /* Free the pointers in a NULL-terminated vector of pointers, then
1032 free the pointer itself. */
1034 free_vec (char **vec)
1045 /* Append vector V2 to vector V1. The function frees V2 and
1046 reallocates V1 (thus you may not use the contents of neither
1047 pointer after the call). If V1 is NULL, V2 is returned. */
1049 merge_vecs (char **v1, char **v2)
1059 /* To avoid j == 0 */
1064 for (i = 0; v1[i]; i++);
1066 for (j = 0; v2[j]; j++);
1067 /* Reallocate v1. */
1068 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1069 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1074 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1075 is allocated as needed. Return the new value of the vector. */
1078 vec_append (char **vec, const char *str)
1080 int cnt; /* count of vector elements, including
1081 the one we're about to append */
1084 for (cnt = 0; vec[cnt]; cnt++)
1090 /* Reallocate the array to fit the new element and the NULL. */
1091 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1092 /* Append a copy of STR to the vector. */
1093 vec[cnt - 1] = xstrdup (str);
1098 /* Sometimes it's useful to create "sets" of strings, i.e. special
1099 hash tables where you want to store strings as keys and merely
1100 query for their existence. Here is a set of utility routines that
1101 makes that transparent. */
1104 string_set_add (struct hash_table *ht, const char *s)
1106 /* First check whether the set element already exists. If it does,
1107 do nothing so that we don't have to free() the old element and
1108 then strdup() a new one. */
1109 if (hash_table_contains (ht, s))
1112 /* We use "1" as value. It provides us a useful and clear arbitrary
1113 value, and it consumes no memory -- the pointers to the same
1114 string "1" will be shared by all the key-value pairs in all `set'
1116 hash_table_put (ht, xstrdup (s), "1");
1119 /* Synonym for hash_table_contains... */
1122 string_set_contains (struct hash_table *ht, const char *s)
1124 return hash_table_contains (ht, s);
1128 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1130 char ***arrayptr = (char ***) arg;
1131 *(*arrayptr)++ = (char *) key;
1135 /* Convert the specified string set to array. ARRAY should be large
1136 enough to hold hash_table_count(ht) char pointers. */
1138 void string_set_to_array (struct hash_table *ht, char **array)
1140 hash_table_map (ht, string_set_to_array_mapper, &array);
1144 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1151 string_set_free (struct hash_table *ht)
1153 hash_table_map (ht, string_set_free_mapper, NULL);
1154 hash_table_destroy (ht);
1158 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1165 /* Another utility function: call free() on all keys and values of HT. */
1168 free_keys_and_values (struct hash_table *ht)
1170 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1174 /* Add thousand separators to a number already in string form. Used
1175 by with_thousand_seps and with_thousand_seps_large. */
1178 add_thousand_seps (const char *repr)
1180 static char outbuf[48];
1185 /* Reset the pointers. */
1189 /* Ignore the sign for the purpose of adding thousand
1196 /* How many digits before the first separator? */
1197 mod = strlen (inptr) % 3;
1199 for (i = 0; i < mod; i++)
1200 *outptr++ = inptr[i];
1201 /* Now insert the rest of them, putting separator before every
1203 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1205 if (i % 3 == 0 && i1 != 0)
1207 *outptr++ = inptr[i1];
1209 /* Zero-terminate the string. */
1214 /* Return a static pointer to the number printed with thousand
1215 separators inserted at the right places. */
1218 with_thousand_seps (wgint l)
1221 /* Print the number into the buffer. */
1222 number_to_string (inbuf, l);
1223 return add_thousand_seps (inbuf);
1226 /* Write a string representation of LARGE_INT NUMBER into the provided
1229 It would be dangerous to use sprintf, because the code wouldn't
1230 work on a machine with gcc-provided long long support, but without
1231 libc support for "%lld". However, such old systems platforms
1232 typically lack snprintf and will end up using our version, which
1233 does support "%lld" whereever long longs are available. */
1236 large_int_to_string (char *buffer, int bufsize, LARGE_INT number)
1238 snprintf (buffer, bufsize, LARGE_INT_FMT, number);
1241 /* The same as with_thousand_seps, but works on LARGE_INT. */
1244 with_thousand_seps_large (LARGE_INT l)
1247 large_int_to_string (inbuf, sizeof (inbuf), l);
1248 return add_thousand_seps (inbuf);
1251 /* N, a byte quantity, is converted to a human-readable abberviated
1252 form a la sizes printed by `ls -lh'. The result is written to a
1253 static buffer, a pointer to which is returned.
1255 Unlike `with_thousand_seps', this approximates to the nearest unit.
1256 Quoting GNU libit: "Most people visually process strings of 3-4
1257 digits effectively, but longer strings of digits are more prone to
1258 misinterpretation. Hence, converting to an abbreviated form
1259 usually improves readability."
1261 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1262 original computer science meaning of "powers of 1024". Powers of
1263 1000 would be useless since Wget already displays sizes with
1264 thousand separators. We don't use the "*bibyte" names invented in
1265 1998, and seldom used in practice. Wikipedia's entry on kilobyte
1266 discusses this in some detail. */
1269 human_readable (wgint n)
1271 /* These suffixes are compatible with those of GNU `ls -lh'. */
1272 static char powers[] =
1274 'K', /* kilobyte, 2^10 bytes */
1275 'M', /* megabyte, 2^20 bytes */
1276 'G', /* gigabyte, 2^30 bytes */
1277 'T', /* terabyte, 2^40 bytes */
1278 'P', /* petabyte, 2^50 bytes */
1279 'E', /* exabyte, 2^60 bytes */
1284 /* If the quantity is smaller than 1K, just print it. */
1287 snprintf (buf, sizeof (buf), "%d", (int) n);
1291 /* Loop over powers, dividing N with 1024 in each iteration. This
1292 works unchanged for all sizes of wgint, while still avoiding
1293 non-portable `long double' arithmetic. */
1294 for (i = 0; i < countof (powers); i++)
1296 /* At each iteration N is greater than the *subsequent* power.
1297 That way N/1024.0 produces a decimal number in the units of
1299 if ((n >> 10) < 1024 || i == countof (powers) - 1)
1301 /* Must cast to long first because MS VC can't directly cast
1302 __int64 to double. (This is safe because N is known to
1304 double val = (double) (long) n / 1024.0;
1305 /* Print values smaller than 10 with one decimal digits, and
1306 others without any decimals. */
1307 snprintf (buf, sizeof (buf), "%.*f%c",
1308 val < 10 ? 1 : 0, val, powers[i]);
1313 return NULL; /* unreached */
1316 /* Count the digits in the provided number. Used to allocate space
1317 when printing numbers. */
1320 numdigit (wgint number)
1324 ++cnt; /* accomodate '-' */
1325 while ((number /= 10) != 0)
1330 #define PR(mask) *p++ = n / (mask) + '0'
1332 /* DIGITS_<D> is used to print a D-digit number and should be called
1333 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1334 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1335 Recursively this continues until DIGITS_1 is invoked. */
1337 #define DIGITS_1(mask) PR (mask)
1338 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1339 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1340 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1341 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1342 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1343 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1344 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1345 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1346 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1348 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1350 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1351 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1352 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1353 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1354 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1355 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1356 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1357 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1358 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1360 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1361 cases and to portably support strange sizes of wgint. Ideally this
1362 would just use "%j" and intmax_t, but many systems don't support
1363 it, so it's used only if nothing else works. */
1364 #if SIZEOF_LONG >= SIZEOF_WGINT
1365 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1367 # if SIZEOF_LONG_LONG >= SIZEOF_WGINT
1368 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1371 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64", (__int64) (n))
1373 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
1378 /* Shorthand for casting to wgint. */
1381 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1382 `sprintf(buffer, "%lld", (long long) number)', only typically much
1383 faster and portable to machines without long long.
1385 The speedup may make a difference in programs that frequently
1386 convert numbers to strings. Some implementations of sprintf,
1387 particularly the one in GNU libc, have been known to be extremely
1388 slow when converting integers to strings.
1390 Return the pointer to the location where the terminating zero was
1391 printed. (Equivalent to calling buffer+strlen(buffer) after the
1394 BUFFER should be big enough to accept as many bytes as you expect
1395 the number to take up. On machines with 64-bit longs the maximum
1396 needed size is 24 bytes. That includes the digits needed for the
1397 largest 64-bit number, the `-' sign in case it's negative, and the
1398 terminating '\0'. */
1401 number_to_string (char *buffer, wgint number)
1406 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1407 /* We are running in a strange or misconfigured environment. Let
1408 sprintf cope with it. */
1409 SPRINTF_WGINT (buffer, n);
1410 p += strlen (buffer);
1411 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1417 /* -n would overflow. Have sprintf deal with this. */
1418 SPRINTF_WGINT (buffer, n);
1419 p += strlen (buffer);
1427 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1428 way printing any N is fully open-coded without a loop or jump.
1429 (Also see description of DIGITS_*.) */
1431 if (n < 10) DIGITS_1 (1);
1432 else if (n < 100) DIGITS_2 (10);
1433 else if (n < 1000) DIGITS_3 (100);
1434 else if (n < 10000) DIGITS_4 (1000);
1435 else if (n < 100000) DIGITS_5 (10000);
1436 else if (n < 1000000) DIGITS_6 (100000);
1437 else if (n < 10000000) DIGITS_7 (1000000);
1438 else if (n < 100000000) DIGITS_8 (10000000);
1439 else if (n < 1000000000) DIGITS_9 (100000000);
1440 #if SIZEOF_WGINT == 4
1441 /* wgint is 32 bits wide: no number has more than 10 digits. */
1442 else DIGITS_10 (1000000000);
1444 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1445 Constants are constructed by compile-time multiplication to avoid
1446 dealing with different notations for 64-bit constants
1447 (nL/nLL/nI64, depending on the compiler and architecture). */
1448 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1449 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1450 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1451 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1452 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1453 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1454 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1455 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1456 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1457 else DIGITS_19 (1000000000*(W)1000000000);
1461 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1490 /* Print NUMBER to a statically allocated string and return a pointer
1491 to the printed representation.
1493 This function is intended to be used in conjunction with printf.
1494 It is hard to portably print wgint values:
1495 a) you cannot use printf("%ld", number) because wgint can be long
1496 long on 32-bit machines with LFS.
1497 b) you cannot use printf("%lld", number) because NUMBER could be
1498 long on 32-bit machines without LFS, or on 64-bit machines,
1499 which do not require LFS. Also, Windows doesn't support %lld.
1500 c) you cannot use printf("%j", (int_max_t) number) because not all
1501 versions of printf support "%j", the most notable being the one
1503 d) you cannot #define WGINT_FMT to the appropriate format and use
1504 printf(WGINT_FMT, number) because that would break translations
1505 for user-visible messages, such as printf("Downloaded: %d
1508 What you should use instead is printf("%s", number_to_static_string
1511 CAVEAT: since the function returns pointers to static data, you
1512 must be careful to copy its result before calling it again.
1513 However, to make it more useful with printf, the function maintains
1514 an internal ring of static buffers to return. That way things like
1515 printf("%s %s", number_to_static_string (num1),
1516 number_to_static_string (num2)) work as expected. Three buffers
1517 are currently used, which means that "%s %s %s" will work, but "%s
1518 %s %s %s" won't. If you need to print more than three wgints,
1519 bump the RING_SIZE (or rethink your message.) */
1522 number_to_static_string (wgint number)
1524 static char ring[RING_SIZE][24];
1526 char *buf = ring[ringpos];
1527 number_to_string (buf, number);
1528 ringpos = (ringpos + 1) % RING_SIZE;
1532 /* Determine the width of the terminal we're running on. If that's
1533 not possible, return 0. */
1536 determine_screen_width (void)
1538 /* If there's a way to get the terminal size using POSIX
1539 tcgetattr(), somebody please tell me. */
1544 if (opt.lfilename != NULL)
1547 fd = fileno (stderr);
1548 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1549 return 0; /* most likely ENOTTY */
1552 #else /* not TIOCGWINSZ */
1554 CONSOLE_SCREEN_BUFFER_INFO csbi;
1555 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1557 return csbi.dwSize.X;
1558 # else /* neither WINDOWS nor TIOCGWINSZ */
1560 #endif /* neither WINDOWS nor TIOCGWINSZ */
1561 #endif /* not TIOCGWINSZ */
1564 /* Return a random number between 0 and MAX-1, inclusive.
1566 If MAX is greater than the value of RAND_MAX+1 on the system, the
1567 returned value will be in the range [0, RAND_MAX]. This may be
1568 fixed in a future release.
1570 The random number generator is seeded automatically the first time
1573 This uses rand() for portability. It has been suggested that
1574 random() offers better randomness, but this is not required for
1575 Wget, so I chose to go for simplicity and use rand
1578 DO NOT use this for cryptographic purposes. It is only meant to be
1579 used in situations where quality of the random numbers returned
1580 doesn't really matter. */
1583 random_number (int max)
1591 srand (time (NULL));
1596 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1597 and enforce that assumption by masking other bits. */
1599 # define RAND_MAX 32767
1603 /* This is equivalent to rand() % max, but uses the high-order bits
1604 for better randomness on architecture where rand() is implemented
1605 using a simple congruential generator. */
1607 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1608 return (int)bounded;
1611 /* Return a random uniformly distributed floating point number in the
1612 [0, 1) range. The precision of returned numbers is 9 digits.
1614 Modify this to use erand48() where available! */
1619 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1620 sure it's greater than 1000. */
1621 int rnd1 = random_number (1000);
1622 int rnd2 = random_number (1000);
1623 int rnd3 = random_number (1000);
1624 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1627 /* Implementation of run_with_timeout, a generic timeout-forcing
1628 routine for systems with Unix-like signal handling. */
1630 #ifdef USE_SIGNAL_TIMEOUT
1631 # ifdef HAVE_SIGSETJMP
1632 # define SETJMP(env) sigsetjmp (env, 1)
1634 static sigjmp_buf run_with_timeout_env;
1637 abort_run_with_timeout (int sig)
1639 assert (sig == SIGALRM);
1640 siglongjmp (run_with_timeout_env, -1);
1642 # else /* not HAVE_SIGSETJMP */
1643 # define SETJMP(env) setjmp (env)
1645 static jmp_buf run_with_timeout_env;
1648 abort_run_with_timeout (int sig)
1650 assert (sig == SIGALRM);
1651 /* We don't have siglongjmp to preserve the set of blocked signals;
1652 if we longjumped out of the handler at this point, SIGALRM would
1653 remain blocked. We must unblock it manually. */
1654 int mask = siggetmask ();
1655 mask &= ~sigmask (SIGALRM);
1658 /* Now it's safe to longjump. */
1659 longjmp (run_with_timeout_env, -1);
1661 # endif /* not HAVE_SIGSETJMP */
1663 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1664 setitimer where available, alarm otherwise.
1666 TIMEOUT should be non-zero. If the timeout value is so small that
1667 it would be rounded to zero, it is rounded to the least legal value
1668 instead (1us for setitimer, 1s for alarm). That ensures that
1669 SIGALRM will be delivered in all cases. */
1672 alarm_set (double timeout)
1675 /* Use the modern itimer interface. */
1676 struct itimerval itv;
1678 itv.it_value.tv_sec = (long) timeout;
1679 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1680 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1681 /* Ensure that we wait for at least the minimum interval.
1682 Specifying zero would mean "wait forever". */
1683 itv.it_value.tv_usec = 1;
1684 setitimer (ITIMER_REAL, &itv, NULL);
1685 #else /* not ITIMER_REAL */
1686 /* Use the old alarm() interface. */
1687 int secs = (int) timeout;
1689 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1690 because alarm(0) means "never deliver the alarm", i.e. "wait
1691 forever", which is not what someone who specifies a 0.5s
1692 timeout would expect. */
1695 #endif /* not ITIMER_REAL */
1698 /* Cancel the alarm set with alarm_set. */
1704 struct itimerval disable;
1706 setitimer (ITIMER_REAL, &disable, NULL);
1707 #else /* not ITIMER_REAL */
1709 #endif /* not ITIMER_REAL */
1712 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1713 seconds. Returns non-zero if the function was interrupted with a
1714 timeout, zero otherwise.
1716 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1717 using setitimer() or alarm(). The timeout is enforced by
1718 longjumping out of the SIGALRM handler. This has several
1719 advantages compared to the traditional approach of relying on
1720 signals causing system calls to exit with EINTR:
1722 * The callback function is *forcibly* interrupted after the
1723 timeout expires, (almost) regardless of what it was doing and
1724 whether it was in a syscall. For example, a calculation that
1725 takes a long time is interrupted as reliably as an IO
1728 * It works with both SYSV and BSD signals because it doesn't
1729 depend on the default setting of SA_RESTART.
1731 * It doesn't require special handler setup beyond a simple call
1732 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1735 The only downside is that, if FUN allocates internal resources that
1736 are normally freed prior to exit from the functions, they will be
1737 lost in case of timeout. */
1740 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1750 signal (SIGALRM, abort_run_with_timeout);
1751 if (SETJMP (run_with_timeout_env) != 0)
1753 /* Longjumped out of FUN with a timeout. */
1754 signal (SIGALRM, SIG_DFL);
1757 alarm_set (timeout);
1760 /* Preserve errno in case alarm() or signal() modifies it. */
1761 saved_errno = errno;
1763 signal (SIGALRM, SIG_DFL);
1764 errno = saved_errno;
1769 #else /* not USE_SIGNAL_TIMEOUT */
1772 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1773 define it under Windows, because Windows has its own version of
1774 run_with_timeout that uses threads. */
1777 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1782 #endif /* not WINDOWS */
1783 #endif /* not USE_SIGNAL_TIMEOUT */
1787 /* Sleep the specified amount of seconds. On machines without
1788 nanosleep(), this may sleep shorter if interrupted by signals. */
1791 xsleep (double seconds)
1793 #ifdef HAVE_NANOSLEEP
1794 /* nanosleep is the preferred interface because it offers high
1795 accuracy and, more importantly, because it allows us to reliably
1796 restart receiving a signal such as SIGWINCH. (There was an
1797 actual Debian bug report about --limit-rate malfunctioning while
1798 the terminal was being resized.) */
1799 struct timespec sleep, remaining;
1800 sleep.tv_sec = (long) seconds;
1801 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1802 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1803 /* If nanosleep has been interrupted by a signal, adjust the
1804 sleeping period and return to sleep. */
1806 #else /* not HAVE_NANOSLEEP */
1808 /* If usleep is available, use it in preference to select. */
1811 /* On some systems, usleep cannot handle values larger than
1812 1,000,000. If the period is larger than that, use sleep
1813 first, then add usleep for subsecond accuracy. */
1815 seconds -= (long) seconds;
1817 usleep (seconds * 1000000);
1818 #else /* not HAVE_USLEEP */
1820 /* Note that, although Windows supports select, this sleeping
1821 strategy doesn't work there because Winsock's select doesn't
1822 implement timeout when it is passed NULL pointers for all fd
1823 sets. (But it does work under Cygwin, which implements its own
1825 struct timeval sleep;
1826 sleep.tv_sec = (long) seconds;
1827 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1828 select (0, NULL, NULL, NULL, &sleep);
1829 /* If select returns -1 and errno is EINTR, it means we were
1830 interrupted by a signal. But without knowing how long we've
1831 actually slept, we can't return to sleep. Using gettimeofday to
1832 track sleeps is slow and unreliable due to clock skew. */
1833 #else /* not HAVE_SELECT */
1835 #endif /* not HAVE_SELECT */
1836 #endif /* not HAVE_USLEEP */
1837 #endif /* not HAVE_NANOSLEEP */
1840 #endif /* not WINDOWS */
1842 /* Encode the string STR of length LENGTH to base64 format and place it
1843 to B64STORE. The output will be \0-terminated, and must point to a
1844 writable buffer of at least 1+BASE64_LENGTH(length) bytes. It
1845 returns the length of the resulting base64 data, not counting the
1848 This implementation will not emit newlines after 76 characters of
1852 base64_encode (const char *str, int length, char *b64store)
1854 /* Conversion table. */
1855 static char tbl[64] = {
1856 'A','B','C','D','E','F','G','H',
1857 'I','J','K','L','M','N','O','P',
1858 'Q','R','S','T','U','V','W','X',
1859 'Y','Z','a','b','c','d','e','f',
1860 'g','h','i','j','k','l','m','n',
1861 'o','p','q','r','s','t','u','v',
1862 'w','x','y','z','0','1','2','3',
1863 '4','5','6','7','8','9','+','/'
1866 const unsigned char *s = (const unsigned char *) str;
1869 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1870 for (i = 0; i < length; i += 3)
1872 *p++ = tbl[s[0] >> 2];
1873 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1874 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1875 *p++ = tbl[s[2] & 0x3f];
1879 /* Pad the result if necessary... */
1880 if (i == length + 1)
1882 else if (i == length + 2)
1883 *(p - 1) = *(p - 2) = '=';
1885 /* ...and zero-terminate it. */
1888 return p - b64store;
1891 #define IS_ASCII(c) (((c) & 0x80) == 0)
1892 #define IS_BASE64(c) ((IS_ASCII (c) && base64_char_to_value[c] >= 0) || c == '=')
1894 /* Get next character from the string, except that non-base64
1895 characters are ignored, as mandated by rfc2045. */
1896 #define NEXT_BASE64_CHAR(c, p) do { \
1898 } while (c != '\0' && !IS_BASE64 (c))
1900 /* Decode data from BASE64 (assumed to be encoded as base64) into
1901 memory pointed to by TO. TO should be large enough to accomodate
1902 the decoded data, which is guaranteed to be less than
1905 Since TO is assumed to contain binary data, it is not
1906 NUL-terminated. The function returns the length of the data
1907 written to TO. -1 is returned in case of error caused by malformed
1911 base64_decode (const char *base64, char *to)
1913 /* Table of base64 values for first 128 characters. Note that this
1914 assumes ASCII (but so does Wget in other places). */
1915 static short base64_char_to_value[128] =
1917 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1918 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1919 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1920 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1921 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1922 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1923 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1924 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1925 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1926 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1927 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1928 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1929 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1932 const char *p = base64;
1938 unsigned long value;
1940 /* Process first byte of a quadruplet. */
1941 NEXT_BASE64_CHAR (c, p);
1945 return -1; /* illegal '=' while decoding base64 */
1946 value = base64_char_to_value[c] << 18;
1948 /* Process scond byte of a quadruplet. */
1949 NEXT_BASE64_CHAR (c, p);
1951 return -1; /* premature EOF while decoding base64 */
1953 return -1; /* illegal `=' while decoding base64 */
1954 value |= base64_char_to_value[c] << 12;
1957 /* Process third byte of a quadruplet. */
1958 NEXT_BASE64_CHAR (c, p);
1960 return -1; /* premature EOF while decoding base64 */
1964 NEXT_BASE64_CHAR (c, p);
1966 return -1; /* premature EOF while decoding base64 */
1968 return -1; /* padding `=' expected but not found */
1972 value |= base64_char_to_value[c] << 6;
1973 *q++ = 0xff & value >> 8;
1975 /* Process fourth byte of a quadruplet. */
1976 NEXT_BASE64_CHAR (c, p);
1978 return -1; /* premature EOF while decoding base64 */
1982 value |= base64_char_to_value[c];
1983 *q++ = 0xff & value;
1991 #undef NEXT_BASE64_CHAR
1993 /* Simple merge sort for use by stable_sort. Implementation courtesy
1994 Zeljko Vrba with additional debugging by Nenad Barbutov. */
1997 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
1998 int (*cmpfun) (const void *, const void *))
2000 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2004 size_t mid = (to + from) / 2;
2005 mergesort_internal (base, temp, size, from, mid, cmpfun);
2006 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2009 for (k = from; (i <= mid) && (j <= to); k++)
2010 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2011 memcpy (ELT (temp, k), ELT (base, i++), size);
2013 memcpy (ELT (temp, k), ELT (base, j++), size);
2015 memcpy (ELT (temp, k++), ELT (base, i++), size);
2017 memcpy (ELT (temp, k++), ELT (base, j++), size);
2018 for (k = from; k <= to; k++)
2019 memcpy (ELT (base, k), ELT (temp, k), size);
2024 /* Stable sort with interface exactly like standard library's qsort.
2025 Uses mergesort internally, allocating temporary storage with
2029 stable_sort (void *base, size_t nmemb, size_t size,
2030 int (*cmpfun) (const void *, const void *))
2034 void *temp = alloca (nmemb * size * sizeof (void *));
2035 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);