1 /* Various utility functions.
2 Copyright (C) 1996-2005 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 Foundation, Inc.,
18 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 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 Unix version of run_with_timeout. */
75 #ifndef HAVE_SIGSETJMP
76 /* If sigsetjmp is a macro, configure won't pick it up. */
78 # define HAVE_SIGSETJMP
82 #undef USE_SIGNAL_TIMEOUT
83 #if defined(HAVE_SIGSETJMP) || defined(HAVE_SIGBLOCK)
84 # define USE_SIGNAL_TIMEOUT
91 /* Utility function: like xstrdup(), but also lowercases S. */
94 xstrdup_lower (const char *s)
96 char *copy = xstrdup (s);
103 /* Copy the string formed by two pointers (one on the beginning, other
104 on the char after the last char) to a new, malloc-ed location.
107 strdupdelim (const char *beg, const char *end)
109 char *res = xmalloc (end - beg + 1);
110 memcpy (res, beg, end - beg);
111 res[end - beg] = '\0';
115 /* Parse a string containing comma-separated elements, and return a
116 vector of char pointers with the elements. Spaces following the
117 commas are ignored. */
119 sepstring (const char *s)
133 res = xrealloc (res, (i + 2) * sizeof (char *));
134 res[i] = strdupdelim (p, s);
137 /* Skip the blanks following the ','. */
145 res = xrealloc (res, (i + 2) * sizeof (char *));
146 res[i] = strdupdelim (p, s);
151 /* Like sprintf, but allocates a string of sufficient size with malloc
152 and returns it. GNU libc has a similar function named asprintf,
153 which requires the pointer to the string to be passed. */
156 aprintf (const char *fmt, ...)
158 /* This function is implemented using vsnprintf, which we provide
159 for the systems that don't have it. Therefore, it should be 100%
163 char *str = xmalloc (size);
170 /* See log_vprintf_internal for explanation why it's OK to rely
171 on the return value of vsnprintf. */
173 va_start (args, fmt);
174 n = vsnprintf (str, size, fmt, args);
177 /* If the printing worked, return the string. */
178 if (n > -1 && n < size)
181 /* Else try again with a larger buffer. */
182 if (n > -1) /* C99 */
183 size = n + 1; /* precisely what is needed */
185 size <<= 1; /* twice the old size */
186 str = xrealloc (str, size);
190 /* Concatenate the NULL-terminated list of string arguments into
191 freshly allocated space. */
194 concat_strings (const char *str0, ...)
197 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
200 const char *next_str;
201 int total_length = 0;
204 /* Calculate the length of and allocate the resulting string. */
207 va_start (args, str0);
208 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
210 int len = strlen (next_str);
211 if (argcount < countof (saved_lengths))
212 saved_lengths[argcount++] = len;
216 p = ret = xmalloc (total_length + 1);
218 /* Copy the strings into the allocated space. */
221 va_start (args, str0);
222 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
225 if (argcount < countof (saved_lengths))
226 len = saved_lengths[argcount++];
228 len = strlen (next_str);
229 memcpy (p, next_str, len);
238 /* Return pointer to a static char[] buffer in which zero-terminated
239 string-representation of TM (in form hh:mm:ss) is printed.
241 If TM is NULL, the current time will be used. */
244 time_str (time_t *tm)
246 static char output[15];
248 time_t secs = tm ? *tm : time (NULL);
252 /* In case of error, return the empty string. Maybe we should
253 just abort if this happens? */
257 ptm = localtime (&secs);
258 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
262 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
265 datetime_str (time_t *tm)
267 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
269 time_t secs = tm ? *tm : time (NULL);
273 /* In case of error, return the empty string. Maybe we should
274 just abort if this happens? */
278 ptm = localtime (&secs);
279 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
280 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
281 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
285 /* The Windows versions of the following two functions are defined in
290 fork_to_background (void)
293 /* Whether we arrange our own version of opt.lfilename here. */
294 bool logfile_changed = false;
298 /* We must create the file immediately to avoid either a race
299 condition (which arises from using unique_name and failing to
300 use fopen_excl) or lying to the user about the log file name
301 (which arises from using unique_name, printing the name, and
302 using fopen_excl later on.) */
303 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
306 logfile_changed = true;
319 /* parent, no error */
320 printf (_("Continuing in background, pid %d.\n"), (int) pid);
322 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
323 exit (0); /* #### should we use _exit()? */
326 /* child: give up the privileges and keep running. */
328 freopen ("/dev/null", "r", stdin);
329 freopen ("/dev/null", "w", stdout);
330 freopen ("/dev/null", "w", stderr);
332 #endif /* not WINDOWS */
334 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
335 specified with TM. The atime ("access time") is set to the current
339 touch (const char *file, time_t tm)
341 #ifdef HAVE_STRUCT_UTIMBUF
342 struct utimbuf times;
350 times.actime = time (NULL);
351 if (utime (file, ×) == -1)
352 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
355 /* Checks if FILE is a symbolic link, and removes it if it is. Does
356 nothing under MS-Windows. */
358 remove_link (const char *file)
363 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
365 DEBUGP (("Unlinking %s (symlink).\n", file));
368 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
369 file, strerror (errno));
374 /* Does FILENAME exist? This is quite a lousy implementation, since
375 it supplies no error codes -- only a yes-or-no answer. Thus it
376 will return that a file does not exist if, e.g., the directory is
377 unreadable. I don't mind it too much currently, though. The
378 proper way should, of course, be to have a third, error state,
379 other than true/false, but that would introduce uncalled-for
380 additional complexity to the callers. */
382 file_exists_p (const char *filename)
385 return access (filename, F_OK) >= 0;
388 return stat (filename, &buf) >= 0;
392 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
393 Returns 0 on error. */
395 file_non_directory_p (const char *path)
398 /* Use lstat() rather than stat() so that symbolic links pointing to
399 directories can be identified correctly. */
400 if (lstat (path, &buf) != 0)
402 return S_ISDIR (buf.st_mode) ? false : true;
405 /* Return the size of file named by FILENAME, or -1 if it cannot be
406 opened or seeked into. */
408 file_size (const char *filename)
410 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
412 /* We use fseek rather than stat to determine the file size because
413 that way we can also verify that the file is readable without
414 explicitly checking for permissions. Inspired by the POST patch
416 FILE *fp = fopen (filename, "rb");
419 fseeko (fp, 0, SEEK_END);
425 if (stat (filename, &st) < 0)
431 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
432 doesn't exist is found. Return a freshly allocated copy of the
436 unique_name_1 (const char *prefix)
439 int plen = strlen (prefix);
440 char *template = (char *)alloca (plen + 1 + 24);
441 char *template_tail = template + plen;
443 memcpy (template, prefix, plen);
444 *template_tail++ = '.';
447 number_to_string (template_tail, count++);
448 while (file_exists_p (template));
450 return xstrdup (template);
453 /* Return a unique file name, based on FILE.
455 More precisely, if FILE doesn't exist, it is returned unmodified.
456 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
457 file name that doesn't exist is returned.
459 The resulting file is not created, only verified that it didn't
460 exist at the point in time when the function was called.
461 Therefore, where security matters, don't rely that the file created
462 by this function exists until you open it with O_EXCL or
465 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
466 string. Otherwise, it may return FILE if the file doesn't exist
467 (and therefore doesn't need changing). */
470 unique_name (const char *file, bool allow_passthrough)
472 /* If the FILE itself doesn't exist, return it without
474 if (!file_exists_p (file))
475 return allow_passthrough ? (char *)file : xstrdup (file);
477 /* Otherwise, find a numeric suffix that results in unused file name
479 return unique_name_1 (file);
482 /* Create a file based on NAME, except without overwriting an existing
483 file with that name. Providing O_EXCL is correctly implemented,
484 this function does not have the race condition associated with
485 opening the file returned by unique_name. */
488 unique_create (const char *name, bool binary, char **opened_name)
490 /* unique file name, based on NAME */
491 char *uname = unique_name (name, false);
493 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
496 uname = unique_name (name, false);
498 if (opened_name && fp != NULL)
501 *opened_name = uname;
513 /* Open the file for writing, with the addition that the file is
514 opened "exclusively". This means that, if the file already exists,
515 this function will *fail* and errno will be set to EEXIST. If
516 BINARY is set, the file will be opened in binary mode, equivalent
519 If opening the file fails for any reason, including the file having
520 previously existed, this function returns NULL and sets errno
524 fopen_excl (const char *fname, bool binary)
528 int flags = O_WRONLY | O_CREAT | O_EXCL;
533 fd = open (fname, flags, 0666);
536 return fdopen (fd, binary ? "wb" : "w");
537 #else /* not O_EXCL */
538 /* Manually check whether the file exists. This is prone to race
539 conditions, but systems without O_EXCL haven't deserved
541 if (file_exists_p (fname))
546 return fopen (fname, binary ? "wb" : "w");
547 #endif /* not O_EXCL */
550 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
551 are missing, create them first. In case any mkdir() call fails,
552 return its error status. Returns 0 on successful completion.
554 The behaviour of this function should be identical to the behaviour
555 of `mkdir -p' on systems where mkdir supports the `-p' option. */
557 make_directory (const char *directory)
559 int i, ret, quit = 0;
562 /* Make a copy of dir, to be able to write to it. Otherwise, the
563 function is unsafe if called with a read-only char *argument. */
564 STRDUP_ALLOCA (dir, directory);
566 /* If the first character of dir is '/', skip it (and thus enable
567 creation of absolute-pathname directories. */
568 for (i = (*dir == '/'); 1; ++i)
570 for (; dir[i] && dir[i] != '/'; i++)
575 /* Check whether the directory already exists. Allow creation of
576 of intermediate directories to fail, as the initial path components
577 are not necessarily directories! */
578 if (!file_exists_p (dir))
579 ret = mkdir (dir, 0777);
590 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
591 should be a file name.
593 file_merge("/foo/bar", "baz") => "/foo/baz"
594 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
595 file_merge("foo", "bar") => "bar"
597 In other words, it's a simpler and gentler version of uri_merge. */
600 file_merge (const char *base, const char *file)
603 const char *cut = (const char *)strrchr (base, '/');
606 return xstrdup (file);
608 result = xmalloc (cut - base + 1 + strlen (file) + 1);
609 memcpy (result, base, cut - base);
610 result[cut - base] = '/';
611 strcpy (result + (cut - base) + 1, file);
616 /* Like fnmatch, but performs a case-insensitive match. */
619 fnmatch_nocase (const char *pattern, const char *string, int flags)
622 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
623 also present on *BSD platforms, and possibly elsewhere. */
624 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
626 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
627 char *patcopy = (char *) alloca (strlen (pattern) + 1);
628 char *strcopy = (char *) alloca (strlen (string) + 1);
630 for (p = patcopy; *pattern; pattern++, p++)
631 *p = TOLOWER (*pattern);
633 for (p = strcopy; *string; string++, p++)
634 *p = TOLOWER (*string);
636 return fnmatch (patcopy, strcopy, flags);
640 static bool in_acclist (const char *const *, const char *, bool);
642 /* Determine whether a file is acceptable to be followed, according to
643 lists of patterns to accept/reject. */
645 acceptable (const char *s)
649 while (l && s[l] != '/')
656 return (in_acclist ((const char *const *)opt.accepts, s, true)
657 && !in_acclist ((const char *const *)opt.rejects, s, true));
659 return in_acclist ((const char *const *)opt.accepts, s, true);
661 else if (opt.rejects)
662 return !in_acclist ((const char *const *)opt.rejects, s, true);
666 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
667 `/something', frontcmp() will return true only if S2 begins with
670 frontcmp (const char *s1, const char *s2)
672 if (!opt.ignore_case)
673 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2)
676 for (; *s1 && *s2 && (TOLOWER (*s1) == TOLOWER (*s2)); ++s1, ++s2)
681 /* Iterate through STRLIST, and return the first element that matches
682 S, through wildcards or front comparison (as appropriate). */
684 proclist (char **strlist, const char *s)
687 int (*matcher) (const char *, const char *, int)
688 = opt.ignore_case ? fnmatch_nocase : fnmatch;
690 for (x = strlist; *x; x++)
692 /* Remove leading '/' */
693 char *p = *x + (**x == '/');
694 if (has_wildcards_p (p))
696 if (matcher (p, s, FNM_PATHNAME) == 0)
708 /* Returns whether DIRECTORY is acceptable for download, wrt the
709 include/exclude lists.
711 The leading `/' is ignored in paths; relative and absolute paths
712 may be freely intermixed. */
715 accdir (const char *directory)
717 /* Remove starting '/'. */
718 if (*directory == '/')
722 if (!proclist (opt.includes, directory))
727 if (proclist (opt.excludes, directory))
733 /* Return true if STRING ends with TAIL. For instance:
735 match_tail ("abc", "bc", false) -> 1
736 match_tail ("abc", "ab", false) -> 0
737 match_tail ("abc", "abc", false) -> 1
739 If FOLD_CASE is true, the comparison will be case-insensitive. */
742 match_tail (const char *string, const char *tail, bool fold_case)
746 /* We want this to be fast, so we code two loops, one with
747 case-folding, one without. */
751 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
752 if (string[i] != tail[j])
757 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
758 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
762 /* If the tail was exhausted, the match was succesful. */
769 /* Checks whether string S matches each element of ACCEPTS. A list
770 element are matched either with fnmatch() or match_tail(),
771 according to whether the element contains wildcards or not.
773 If the BACKWARD is false, don't do backward comparison -- just compare
776 in_acclist (const char *const *accepts, const char *s, bool backward)
778 for (; *accepts; accepts++)
780 if (has_wildcards_p (*accepts))
782 int res = opt.ignore_case
783 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
784 /* fnmatch returns 0 if the pattern *does* match the string. */
792 if (match_tail (s, *accepts, opt.ignore_case))
797 int cmp = opt.ignore_case
798 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
807 /* Return the location of STR's suffix (file extension). Examples:
808 suffix ("foo.bar") -> "bar"
809 suffix ("foo.bar.baz") -> "baz"
810 suffix ("/foo/bar") -> NULL
811 suffix ("/foo.bar/baz") -> NULL */
813 suffix (const char *str)
817 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
821 return (char *)str + i;
826 /* Return true if S contains globbing wildcards (`*', `?', `[' or
830 has_wildcards_p (const char *s)
833 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
838 /* Return true if FNAME ends with a typical HTML suffix. The
839 following (case-insensitive) suffixes are presumed to be HTML
844 ?html (`?' matches one character)
846 #### CAVEAT. This is not necessarily a good indication that FNAME
847 refers to a file that contains HTML! */
849 has_html_suffix_p (const char *fname)
853 if ((suf = suffix (fname)) == NULL)
855 if (!strcasecmp (suf, "html"))
857 if (!strcasecmp (suf, "htm"))
859 if (suf[0] && !strcasecmp (suf + 1, "html"))
864 /* Read a line from FP and return the pointer to freshly allocated
865 storage. The storage space is obtained through malloc() and should
866 be freed with free() when it is no longer needed.
868 The length of the line is not limited, except by available memory.
869 The newline character at the end of line is retained. The line is
870 terminated with a zero character.
872 After end-of-file is encountered without anything being read, NULL
873 is returned. NULL is also returned on error. To distinguish
874 between these two cases, use the stdio function ferror(). */
877 read_whole_line (FILE *fp)
881 char *line = xmalloc (bufsize);
883 while (fgets (line + length, bufsize - length, fp))
885 length += strlen (line + length);
887 /* Possible for example when reading from a binary file where
888 a line begins with \0. */
891 if (line[length - 1] == '\n')
894 /* fgets() guarantees to read the whole line, or to use up the
895 space we've given it. We can double the buffer
898 line = xrealloc (line, bufsize);
900 if (length == 0 || ferror (fp))
905 if (length + 1 < bufsize)
906 /* Relieve the memory from our exponential greediness. We say
907 `length + 1' because the terminating \0 is not included in
908 LENGTH. We don't need to zero-terminate the string ourselves,
909 though, because fgets() does that. */
910 line = xrealloc (line, length + 1);
914 /* Read FILE into memory. A pointer to `struct file_memory' are
915 returned; use struct element `content' to access file contents, and
916 the element `length' to know the file length. `content' is *not*
917 zero-terminated, and you should *not* read or write beyond the [0,
918 length) range of characters.
920 After you are done with the file contents, call read_file_free to
923 Depending on the operating system and the type of file that is
924 being read, read_file() either mmap's the file into memory, or
925 reads the file into the core using read().
927 If file is named "-", fileno(stdin) is used for reading instead.
928 If you want to read from a real file named "-", use "./-" instead. */
931 read_file (const char *file)
934 struct file_memory *fm;
936 bool inhibit_close = false;
938 /* Some magic in the finest tradition of Perl and its kin: if FILE
939 is "-", just use stdin. */
943 inhibit_close = true;
944 /* Note that we don't inhibit mmap() in this case. If stdin is
945 redirected from a regular file, mmap() will still work. */
948 fd = open (file, O_RDONLY);
951 fm = xnew (struct file_memory);
956 if (fstat (fd, &buf) < 0)
958 fm->length = buf.st_size;
959 /* NOTE: As far as I know, the callers of this function never
960 modify the file text. Relying on this would enable us to
961 specify PROT_READ and MAP_SHARED for a marginal gain in
962 efficiency, but at some cost to generality. */
963 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
965 if (fm->content == (char *)MAP_FAILED)
975 /* The most common reason why mmap() fails is that FD does not point
976 to a plain file. However, it's also possible that mmap() doesn't
977 work for a particular type of file. Therefore, whenever mmap()
978 fails, we just fall back to the regular method. */
979 #endif /* HAVE_MMAP */
982 size = 512; /* number of bytes fm->contents can
983 hold at any given time. */
984 fm->content = xmalloc (size);
988 if (fm->length > size / 2)
990 /* #### I'm not sure whether the whole exponential-growth
991 thing makes sense with kernel read. On Linux at least,
992 read() refuses to read more than 4K from a file at a
993 single chunk anyway. But other Unixes might optimize it
994 better, and it doesn't *hurt* anything, so I'm leaving
997 /* Normally, we grow SIZE exponentially to make the number
998 of calls to read() and realloc() logarithmic in relation
999 to file size. However, read() can read an amount of data
1000 smaller than requested, and it would be unreasonable to
1001 double SIZE every time *something* was read. Therefore,
1002 we double SIZE only when the length exceeds half of the
1003 entire allocated size. */
1005 fm->content = xrealloc (fm->content, size);
1007 nread = read (fd, fm->content + fm->length, size - fm->length);
1009 /* Successful read. */
1010 fm->length += nread;
1020 if (size > fm->length && fm->length != 0)
1021 /* Due to exponential growth of fm->content, the allocated region
1022 might be much larger than what is actually needed. */
1023 fm->content = xrealloc (fm->content, fm->length);
1030 xfree (fm->content);
1035 /* Release the resources held by FM. Specifically, this calls
1036 munmap() or xfree() on fm->content, depending whether mmap or
1037 malloc/read were used to read in the file. It also frees the
1038 memory needed to hold the FM structure itself. */
1041 read_file_free (struct file_memory *fm)
1046 munmap (fm->content, fm->length);
1051 xfree (fm->content);
1056 /* Free the pointers in a NULL-terminated vector of pointers, then
1057 free the pointer itself. */
1059 free_vec (char **vec)
1070 /* Append vector V2 to vector V1. The function frees V2 and
1071 reallocates V1 (thus you may not use the contents of neither
1072 pointer after the call). If V1 is NULL, V2 is returned. */
1074 merge_vecs (char **v1, char **v2)
1084 /* To avoid j == 0 */
1089 for (i = 0; v1[i]; i++)
1092 for (j = 0; v2[j]; j++)
1094 /* Reallocate v1. */
1095 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1096 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1101 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1102 is allocated as needed. Return the new value of the vector. */
1105 vec_append (char **vec, const char *str)
1107 int cnt; /* count of vector elements, including
1108 the one we're about to append */
1111 for (cnt = 0; vec[cnt]; cnt++)
1117 /* Reallocate the array to fit the new element and the NULL. */
1118 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1119 /* Append a copy of STR to the vector. */
1120 vec[cnt - 1] = xstrdup (str);
1125 /* Sometimes it's useful to create "sets" of strings, i.e. special
1126 hash tables where you want to store strings as keys and merely
1127 query for their existence. Here is a set of utility routines that
1128 makes that transparent. */
1131 string_set_add (struct hash_table *ht, const char *s)
1133 /* First check whether the set element already exists. If it does,
1134 do nothing so that we don't have to free() the old element and
1135 then strdup() a new one. */
1136 if (hash_table_contains (ht, s))
1139 /* We use "1" as value. It provides us a useful and clear arbitrary
1140 value, and it consumes no memory -- the pointers to the same
1141 string "1" will be shared by all the key-value pairs in all `set'
1143 hash_table_put (ht, xstrdup (s), "1");
1146 /* Synonym for hash_table_contains... */
1149 string_set_contains (struct hash_table *ht, const char *s)
1151 return hash_table_contains (ht, s);
1155 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1157 char ***arrayptr = (char ***) arg;
1158 *(*arrayptr)++ = (char *) key;
1162 /* Convert the specified string set to array. ARRAY should be large
1163 enough to hold hash_table_count(ht) char pointers. */
1165 void string_set_to_array (struct hash_table *ht, char **array)
1167 hash_table_map (ht, string_set_to_array_mapper, &array);
1171 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1178 string_set_free (struct hash_table *ht)
1180 hash_table_map (ht, string_set_free_mapper, NULL);
1181 hash_table_destroy (ht);
1185 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1192 /* Another utility function: call free() on all keys and values of HT. */
1195 free_keys_and_values (struct hash_table *ht)
1197 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1201 /* Get grouping data, the separator and grouping info, by calling
1202 localeconv(). The information is cached after the first call to
1205 In locales that don't set a thousand separator (such as the "C"
1206 locale), this forces it to be ",". We are now only showing
1207 thousand separators in one place, so this shouldn't be a problem in
1211 get_grouping_data (const char **sep, const char **grouping)
1213 static const char *cached_sep;
1214 static const char *cached_grouping;
1215 static bool initialized;
1218 /* Get the grouping info from the locale. */
1219 struct lconv *lconv = localeconv ();
1220 cached_sep = lconv->thousands_sep;
1221 cached_grouping = lconv->grouping;
1224 /* Many locales (such as "C" or "hr_HR") don't specify
1225 grouping, which we still want to use it for legibility.
1226 In those locales set the sep char to ',', unless that
1227 character is used for decimal point, in which case set it
1229 if (*lconv->decimal_point != ',')
1233 cached_grouping = "\x03";
1238 *grouping = cached_grouping;
1241 /* Return a printed representation of N with thousand separators.
1242 This should respect locale settings, with the exception of the "C"
1243 locale which mandates no separator, but we use one anyway.
1245 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1246 the separators because it's too non-portable, and it's hard to test
1247 for this feature at configure time. Besides, it wouldn't work in
1248 the "C" locale, which many Unix users still work in. */
1251 with_thousand_seps (wgint n)
1253 static char outbuf[48];
1254 char *p = outbuf + sizeof outbuf;
1256 /* Info received from locale */
1257 const char *grouping, *sep;
1260 /* State information */
1261 int i = 0, groupsize;
1262 const char *atgroup;
1264 bool negative = n < 0;
1266 /* Initialize grouping data. */
1267 get_grouping_data (&sep, &grouping);
1268 seplen = strlen (sep);
1270 groupsize = *atgroup++;
1272 /* This will overflow on WGINT_MIN, but we're not using this to
1273 print negative numbers anyway. */
1277 /* Write the number into the buffer, backwards, inserting the
1278 separators as necessary. */
1282 *--p = n % 10 + '0';
1286 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1287 if (++i == groupsize)
1292 memcpy (p -= seplen, sep, seplen);
1295 groupsize = *atgroup++;
1304 /* N, a byte quantity, is converted to a human-readable abberviated
1305 form a la sizes printed by `ls -lh'. The result is written to a
1306 static buffer, a pointer to which is returned.
1308 Unlike `with_thousand_seps', this approximates to the nearest unit.
1309 Quoting GNU libit: "Most people visually process strings of 3-4
1310 digits effectively, but longer strings of digits are more prone to
1311 misinterpretation. Hence, converting to an abbreviated form
1312 usually improves readability."
1314 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1315 original computer-related meaning of "powers of 1024". Powers of
1316 1000 would be useless since Wget already displays sizes with
1317 thousand separators. We don't use the "*bibyte" names invented in
1318 1998, and seldom used in practice. Wikipedia's entry on kilobyte
1319 discusses this in some detail. */
1322 human_readable (HR_NUMTYPE n)
1324 /* These suffixes are compatible with those of GNU `ls -lh'. */
1325 static char powers[] =
1327 'K', /* kilobyte, 2^10 bytes */
1328 'M', /* megabyte, 2^20 bytes */
1329 'G', /* gigabyte, 2^30 bytes */
1330 'T', /* terabyte, 2^40 bytes */
1331 'P', /* petabyte, 2^50 bytes */
1332 'E', /* exabyte, 2^60 bytes */
1337 /* If the quantity is smaller than 1K, just print it. */
1340 snprintf (buf, sizeof (buf), "%d", (int) n);
1344 /* Loop over powers, dividing N with 1024 in each iteration. This
1345 works unchanged for all sizes of wgint, while still avoiding
1346 non-portable `long double' arithmetic. */
1347 for (i = 0; i < countof (powers); i++)
1349 /* At each iteration N is greater than the *subsequent* power.
1350 That way N/1024.0 produces a decimal number in the units of
1352 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1354 double val = n / 1024.0;
1355 /* Print values smaller than 10 with one decimal digits, and
1356 others without any decimals. */
1357 snprintf (buf, sizeof (buf), "%.*f%c",
1358 val < 10 ? 1 : 0, val, powers[i]);
1363 return NULL; /* unreached */
1366 /* Count the digits in the provided number. Used to allocate space
1367 when printing numbers. */
1370 numdigit (wgint number)
1374 ++cnt; /* accomodate '-' */
1375 while ((number /= 10) != 0)
1380 #define PR(mask) *p++ = n / (mask) + '0'
1382 /* DIGITS_<D> is used to print a D-digit number and should be called
1383 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1384 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1385 Recursively this continues until DIGITS_1 is invoked. */
1387 #define DIGITS_1(mask) PR (mask)
1388 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1389 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1390 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1391 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1392 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1393 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1394 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1395 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1396 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1398 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1400 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1401 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1402 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1403 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1404 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1405 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1406 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1407 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1408 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1410 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1411 cases and to portably support strange sizes of wgint. Ideally this
1412 would just use "%j" and intmax_t, but many systems don't support
1413 it, so it's used only if nothing else works. */
1414 #if SIZEOF_LONG >= SIZEOF_WGINT
1415 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1416 #elif SIZEOF_LONG_LONG >= SIZEOF_WGINT
1417 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1418 #elif defined(WINDOWS)
1419 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64d", (__int64) (n))
1421 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
1424 /* Shorthand for casting to wgint. */
1427 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1428 `sprintf(buffer, "%lld", (long long) number)', only typically much
1429 faster and portable to machines without long long.
1431 The speedup may make a difference in programs that frequently
1432 convert numbers to strings. Some implementations of sprintf,
1433 particularly the one in GNU libc, have been known to be extremely
1434 slow when converting integers to strings.
1436 Return the pointer to the location where the terminating zero was
1437 printed. (Equivalent to calling buffer+strlen(buffer) after the
1440 BUFFER should be big enough to accept as many bytes as you expect
1441 the number to take up. On machines with 64-bit longs the maximum
1442 needed size is 24 bytes. That includes the digits needed for the
1443 largest 64-bit number, the `-' sign in case it's negative, and the
1444 terminating '\0'. */
1447 number_to_string (char *buffer, wgint number)
1452 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1453 /* We are running in a strange or misconfigured environment. Let
1454 sprintf cope with it. */
1455 SPRINTF_WGINT (buffer, n);
1456 p += strlen (buffer);
1457 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1463 /* -n would overflow. Have sprintf deal with this. */
1464 SPRINTF_WGINT (buffer, n);
1465 p += strlen (buffer);
1473 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1474 way printing any N is fully open-coded without a loop or jump.
1475 (Also see description of DIGITS_*.) */
1477 if (n < 10) DIGITS_1 (1);
1478 else if (n < 100) DIGITS_2 (10);
1479 else if (n < 1000) DIGITS_3 (100);
1480 else if (n < 10000) DIGITS_4 (1000);
1481 else if (n < 100000) DIGITS_5 (10000);
1482 else if (n < 1000000) DIGITS_6 (100000);
1483 else if (n < 10000000) DIGITS_7 (1000000);
1484 else if (n < 100000000) DIGITS_8 (10000000);
1485 else if (n < 1000000000) DIGITS_9 (100000000);
1486 #if SIZEOF_WGINT == 4
1487 /* wgint is 32 bits wide: no number has more than 10 digits. */
1488 else DIGITS_10 (1000000000);
1490 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1491 Constants are constructed by compile-time multiplication to avoid
1492 dealing with different notations for 64-bit constants
1493 (nL/nLL/nI64, depending on the compiler and architecture). */
1494 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1495 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1496 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1497 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1498 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1499 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1500 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1501 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1502 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1503 else DIGITS_19 (1000000000*(W)1000000000);
1507 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1514 #undef SPRINTF_WGINT
1537 /* Print NUMBER to a statically allocated string and return a pointer
1538 to the printed representation.
1540 This function is intended to be used in conjunction with printf.
1541 It is hard to portably print wgint values:
1542 a) you cannot use printf("%ld", number) because wgint can be long
1543 long on 32-bit machines with LFS.
1544 b) you cannot use printf("%lld", number) because NUMBER could be
1545 long on 32-bit machines without LFS, or on 64-bit machines,
1546 which do not require LFS. Also, Windows doesn't support %lld.
1547 c) you cannot use printf("%j", (int_max_t) number) because not all
1548 versions of printf support "%j", the most notable being the one
1550 d) you cannot #define WGINT_FMT to the appropriate format and use
1551 printf(WGINT_FMT, number) because that would break translations
1552 for user-visible messages, such as printf("Downloaded: %d
1555 What you should use instead is printf("%s", number_to_static_string
1558 CAVEAT: since the function returns pointers to static data, you
1559 must be careful to copy its result before calling it again.
1560 However, to make it more useful with printf, the function maintains
1561 an internal ring of static buffers to return. That way things like
1562 printf("%s %s", number_to_static_string (num1),
1563 number_to_static_string (num2)) work as expected. Three buffers
1564 are currently used, which means that "%s %s %s" will work, but "%s
1565 %s %s %s" won't. If you need to print more than three wgints,
1566 bump the RING_SIZE (or rethink your message.) */
1569 number_to_static_string (wgint number)
1571 static char ring[RING_SIZE][24];
1573 char *buf = ring[ringpos];
1574 number_to_string (buf, number);
1575 ringpos = (ringpos + 1) % RING_SIZE;
1579 /* Determine the width of the terminal we're running on. If that's
1580 not possible, return 0. */
1583 determine_screen_width (void)
1585 /* If there's a way to get the terminal size using POSIX
1586 tcgetattr(), somebody please tell me. */
1591 if (opt.lfilename != NULL)
1594 fd = fileno (stderr);
1595 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1596 return 0; /* most likely ENOTTY */
1599 #elif defined(WINDOWS)
1600 CONSOLE_SCREEN_BUFFER_INFO csbi;
1601 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1603 return csbi.dwSize.X;
1604 #else /* neither TIOCGWINSZ nor WINDOWS */
1606 #endif /* neither TIOCGWINSZ nor WINDOWS */
1609 /* Whether the rnd system (either rand or [dl]rand48) has been
1611 static int rnd_seeded;
1613 /* Return a random number between 0 and MAX-1, inclusive.
1615 If the system does not support lrand48 and MAX is greater than the
1616 value of RAND_MAX+1 on the system, the returned value will be in
1617 the range [0, RAND_MAX]. This may be fixed in a future release.
1618 The random number generator is seeded automatically the first time
1621 This uses lrand48 where available, rand elsewhere. DO NOT use it
1622 for cryptography. It is only meant to be used in situations where
1623 quality of the random numbers returned doesn't really matter. */
1626 random_number (int max)
1631 srand48 ((long) time (NULL) ^ (long) getpid ());
1634 return lrand48 () % max;
1635 #else /* not HAVE_DRAND48 */
1641 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1646 /* Like rand() % max, but uses the high-order bits for better
1647 randomness on architectures where rand() is implemented using a
1648 simple congruential generator. */
1650 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1651 return (int) bounded;
1653 #endif /* not HAVE_DRAND48 */
1656 /* Return a random uniformly distributed floating point number in the
1657 [0, 1) range. Uses drand48 where available, and a really lame
1658 kludge elsewhere. */
1666 srand48 ((long) time (NULL) ^ (long) getpid ());
1670 #else /* not HAVE_DRAND48 */
1671 return ( random_number (10000) / 10000.0
1672 + random_number (10000) / (10000.0 * 10000.0)
1673 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1674 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1675 #endif /* not HAVE_DRAND48 */
1678 /* Implementation of run_with_timeout, a generic timeout-forcing
1679 routine for systems with Unix-like signal handling. */
1681 #ifdef USE_SIGNAL_TIMEOUT
1682 # ifdef HAVE_SIGSETJMP
1683 # define SETJMP(env) sigsetjmp (env, 1)
1685 static sigjmp_buf run_with_timeout_env;
1688 abort_run_with_timeout (int sig)
1690 assert (sig == SIGALRM);
1691 siglongjmp (run_with_timeout_env, -1);
1693 # else /* not HAVE_SIGSETJMP */
1694 # define SETJMP(env) setjmp (env)
1696 static jmp_buf run_with_timeout_env;
1699 abort_run_with_timeout (int sig)
1701 assert (sig == SIGALRM);
1702 /* We don't have siglongjmp to preserve the set of blocked signals;
1703 if we longjumped out of the handler at this point, SIGALRM would
1704 remain blocked. We must unblock it manually. */
1705 int mask = siggetmask ();
1706 mask &= ~sigmask (SIGALRM);
1709 /* Now it's safe to longjump. */
1710 longjmp (run_with_timeout_env, -1);
1712 # endif /* not HAVE_SIGSETJMP */
1714 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1715 setitimer where available, alarm otherwise.
1717 TIMEOUT should be non-zero. If the timeout value is so small that
1718 it would be rounded to zero, it is rounded to the least legal value
1719 instead (1us for setitimer, 1s for alarm). That ensures that
1720 SIGALRM will be delivered in all cases. */
1723 alarm_set (double timeout)
1726 /* Use the modern itimer interface. */
1727 struct itimerval itv;
1729 itv.it_value.tv_sec = (long) timeout;
1730 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1731 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1732 /* Ensure that we wait for at least the minimum interval.
1733 Specifying zero would mean "wait forever". */
1734 itv.it_value.tv_usec = 1;
1735 setitimer (ITIMER_REAL, &itv, NULL);
1736 #else /* not ITIMER_REAL */
1737 /* Use the old alarm() interface. */
1738 int secs = (int) timeout;
1740 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1741 because alarm(0) means "never deliver the alarm", i.e. "wait
1742 forever", which is not what someone who specifies a 0.5s
1743 timeout would expect. */
1746 #endif /* not ITIMER_REAL */
1749 /* Cancel the alarm set with alarm_set. */
1755 struct itimerval disable;
1757 setitimer (ITIMER_REAL, &disable, NULL);
1758 #else /* not ITIMER_REAL */
1760 #endif /* not ITIMER_REAL */
1763 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1764 seconds. Returns true if the function was interrupted with a
1765 timeout, false otherwise.
1767 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1768 using setitimer() or alarm(). The timeout is enforced by
1769 longjumping out of the SIGALRM handler. This has several
1770 advantages compared to the traditional approach of relying on
1771 signals causing system calls to exit with EINTR:
1773 * The callback function is *forcibly* interrupted after the
1774 timeout expires, (almost) regardless of what it was doing and
1775 whether it was in a syscall. For example, a calculation that
1776 takes a long time is interrupted as reliably as an IO
1779 * It works with both SYSV and BSD signals because it doesn't
1780 depend on the default setting of SA_RESTART.
1782 * It doesn't require special handler setup beyond a simple call
1783 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1786 The only downside is that, if FUN allocates internal resources that
1787 are normally freed prior to exit from the functions, they will be
1788 lost in case of timeout. */
1791 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1801 signal (SIGALRM, abort_run_with_timeout);
1802 if (SETJMP (run_with_timeout_env) != 0)
1804 /* Longjumped out of FUN with a timeout. */
1805 signal (SIGALRM, SIG_DFL);
1808 alarm_set (timeout);
1811 /* Preserve errno in case alarm() or signal() modifies it. */
1812 saved_errno = errno;
1814 signal (SIGALRM, SIG_DFL);
1815 errno = saved_errno;
1820 #else /* not USE_SIGNAL_TIMEOUT */
1823 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1824 define it under Windows, because Windows has its own version of
1825 run_with_timeout that uses threads. */
1828 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1833 #endif /* not WINDOWS */
1834 #endif /* not USE_SIGNAL_TIMEOUT */
1838 /* Sleep the specified amount of seconds. On machines without
1839 nanosleep(), this may sleep shorter if interrupted by signals. */
1842 xsleep (double seconds)
1844 #ifdef HAVE_NANOSLEEP
1845 /* nanosleep is the preferred interface because it offers high
1846 accuracy and, more importantly, because it allows us to reliably
1847 restart receiving a signal such as SIGWINCH. (There was an
1848 actual Debian bug report about --limit-rate malfunctioning while
1849 the terminal was being resized.) */
1850 struct timespec sleep, remaining;
1851 sleep.tv_sec = (long) seconds;
1852 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1853 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1854 /* If nanosleep has been interrupted by a signal, adjust the
1855 sleeping period and return to sleep. */
1857 #elif defined(HAVE_USLEEP)
1858 /* If usleep is available, use it in preference to select. */
1861 /* On some systems, usleep cannot handle values larger than
1862 1,000,000. If the period is larger than that, use sleep
1863 first, then add usleep for subsecond accuracy. */
1865 seconds -= (long) seconds;
1867 usleep (seconds * 1000000);
1868 #else /* fall back select */
1869 /* Note that, although Windows supports select, it can't be used to
1870 implement sleeping because Winsock's select doesn't implement
1871 timeout when it is passed NULL pointers for all fd sets. (But it
1872 does under Cygwin, which implements Unix-compatible select.) */
1873 struct timeval sleep;
1874 sleep.tv_sec = (long) seconds;
1875 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1876 select (0, NULL, NULL, NULL, &sleep);
1877 /* If select returns -1 and errno is EINTR, it means we were
1878 interrupted by a signal. But without knowing how long we've
1879 actually slept, we can't return to sleep. Using gettimeofday to
1880 track sleeps is slow and unreliable due to clock skew. */
1884 #endif /* not WINDOWS */
1886 /* Encode the string STR of length LENGTH to base64 format and place it
1887 to B64STORE. The output will be \0-terminated, and must point to a
1888 writable buffer of at least 1+BASE64_LENGTH(length) bytes. It
1889 returns the length of the resulting base64 data, not counting the
1892 This implementation will not emit newlines after 76 characters of
1896 base64_encode (const char *str, int length, char *b64store)
1898 /* Conversion table. */
1899 static char tbl[64] = {
1900 'A','B','C','D','E','F','G','H',
1901 'I','J','K','L','M','N','O','P',
1902 'Q','R','S','T','U','V','W','X',
1903 'Y','Z','a','b','c','d','e','f',
1904 'g','h','i','j','k','l','m','n',
1905 'o','p','q','r','s','t','u','v',
1906 'w','x','y','z','0','1','2','3',
1907 '4','5','6','7','8','9','+','/'
1910 const unsigned char *s = (const unsigned char *) str;
1913 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1914 for (i = 0; i < length; i += 3)
1916 *p++ = tbl[s[0] >> 2];
1917 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1918 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1919 *p++ = tbl[s[2] & 0x3f];
1923 /* Pad the result if necessary... */
1924 if (i == length + 1)
1926 else if (i == length + 2)
1927 *(p - 1) = *(p - 2) = '=';
1929 /* ...and zero-terminate it. */
1932 return p - b64store;
1935 /* Store in C the next non-whitespace character from the string, or \0
1936 when end of string is reached. */
1937 #define NEXT_CHAR(c, p) do { \
1938 c = (unsigned char) *p++; \
1939 } while (ISSPACE (c))
1941 #define IS_ASCII(c) (((c) & 0x80) == 0)
1943 /* Decode data from BASE64 (pointer to \0-terminated text) into memory
1944 pointed to by TO. TO should be large enough to accomodate the
1945 decoded data, which is guaranteed to be less than strlen(base64).
1947 Since TO is assumed to contain binary data, it is not
1948 NUL-terminated. The function returns the length of the data
1949 written to TO. -1 is returned in case of error caused by malformed
1953 base64_decode (const char *base64, char *to)
1955 /* Table of base64 values for first 128 characters. Note that this
1956 assumes ASCII (but so does Wget in other places). */
1957 static signed char base64_char_to_value[128] =
1959 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1960 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1961 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1962 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1963 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1964 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1965 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1966 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1967 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1968 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1969 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1970 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1971 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1973 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
1974 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
1976 const char *p = base64;
1982 unsigned long value;
1984 /* Process first byte of a quadruplet. */
1988 if (c == '=' || !IS_BASE64 (c))
1989 return -1; /* illegal char while decoding base64 */
1990 value = BASE64_CHAR_TO_VALUE (c) << 18;
1992 /* Process second byte of a quadruplet. */
1995 return -1; /* premature EOF while decoding base64 */
1996 if (c == '=' || !IS_BASE64 (c))
1997 return -1; /* illegal char while decoding base64 */
1998 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2001 /* Process third byte of a quadruplet. */
2004 return -1; /* premature EOF while decoding base64 */
2006 return -1; /* illegal char while decoding base64 */
2012 return -1; /* premature EOF while decoding base64 */
2014 return -1; /* padding `=' expected but not found */
2018 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2019 *q++ = 0xff & value >> 8;
2021 /* Process fourth byte of a quadruplet. */
2024 return -1; /* premature EOF while decoding base64 */
2028 return -1; /* illegal char while decoding base64 */
2030 value |= BASE64_CHAR_TO_VALUE (c);
2031 *q++ = 0xff & value;
2034 #undef BASE64_CHAR_TO_VALUE
2042 /* Simple merge sort for use by stable_sort. Implementation courtesy
2043 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2046 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2047 int (*cmpfun) (const void *, const void *))
2049 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2053 size_t mid = (to + from) / 2;
2054 mergesort_internal (base, temp, size, from, mid, cmpfun);
2055 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2058 for (k = from; (i <= mid) && (j <= to); k++)
2059 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2060 memcpy (ELT (temp, k), ELT (base, i++), size);
2062 memcpy (ELT (temp, k), ELT (base, j++), size);
2064 memcpy (ELT (temp, k++), ELT (base, i++), size);
2066 memcpy (ELT (temp, k++), ELT (base, j++), size);
2067 for (k = from; k <= to; k++)
2068 memcpy (ELT (base, k), ELT (temp, k), size);
2073 /* Stable sort with interface exactly like standard library's qsort.
2074 Uses mergesort internally, allocating temporary storage with
2078 stable_sort (void *base, size_t nmemb, size_t size,
2079 int (*cmpfun) (const void *, const void *))
2083 void *temp = alloca (nmemb * size * sizeof (void *));
2084 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2088 /* Print a decimal number. If it is equal to or larger than ten, the
2089 number is rounded. Otherwise it is printed with one significant
2090 digit without trailing zeros and with no more than three fractional
2091 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2092 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2094 This is useful for displaying durations because it provides
2095 order-of-magnitude information without unnecessary clutter --
2096 long-running downloads are shown without the fractional part, and
2097 short ones still retain one significant digit. */
2100 print_decimal (double number)
2102 static char buf[32];
2103 double n = number >= 0 ? number : -number;
2106 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2107 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2108 snprintf (buf, sizeof buf, "%.0f", number);
2110 snprintf (buf, sizeof buf, "%.1f", number);
2111 else if (n >= 0.001)
2112 snprintf (buf, sizeof buf, "%.1g", number);
2113 else if (n >= 0.0005)
2114 /* round [0.0005, 0.001) to 0.001 */
2115 snprintf (buf, sizeof buf, "%.3f", number);
2117 /* print numbers close to 0 as 0, not 0.000 */