1 /* Various utility functions.
2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
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>
46 # include <process.h> /* getpid() */
51 #ifdef HAVE_SYS_UTIME_H
52 # include <sys/utime.h>
60 /* For TIOCGWINSZ and friends: */
61 #ifdef HAVE_SYS_IOCTL_H
62 # include <sys/ioctl.h>
68 /* Needed for Unix version of run_with_timeout. */
72 #ifndef HAVE_SIGSETJMP
73 /* If sigsetjmp is a macro, configure won't pick it up. */
75 # define HAVE_SIGSETJMP
79 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
80 # 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 prints into a string of sufficient size freshly
152 allocated with malloc, which is returned. If unable to print due
153 to invalid format, returns NULL. Inability to allocate needed
154 memory results in abort, as with xmalloc. This is in spirit
155 similar to the GNU/BSD extension asprintf, but somewhat easier to
158 Internally the function either calls vasprintf or loops around
159 vsnprintf until the correct size is found. Since Wget also ships a
160 fallback implementation of vsnprintf, this should be portable. */
163 aprintf (const char *fmt, ...)
165 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
170 va_start (args, fmt);
171 ret = vasprintf (&str, fmt, args);
173 if (ret < 0 && errno == ENOMEM)
174 abort (); /* for consistency with xmalloc/xrealloc */
178 #else /* not HAVE_VASPRINTF */
180 /* vasprintf is unavailable. snprintf into a small buffer and
181 resize it as necessary. */
183 char *str = xmalloc (size);
185 /* #### This code will infloop and eventually abort in xrealloc if
186 passed a FMT that causes snprintf to consistently return -1. */
193 va_start (args, fmt);
194 n = vsnprintf (str, size, fmt, args);
197 /* If the printing worked, return the string. */
198 if (n > -1 && n < size)
201 /* Else try again with a larger buffer. */
202 if (n > -1) /* C99 */
203 size = n + 1; /* precisely what is needed */
205 size <<= 1; /* twice the old size */
206 str = xrealloc (str, size);
208 #endif /* not HAVE_VASPRINTF */
211 /* Concatenate the NULL-terminated list of string arguments into
212 freshly allocated space. */
215 concat_strings (const char *str0, ...)
218 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
221 const char *next_str;
222 int total_length = 0;
225 /* Calculate the length of and allocate the resulting string. */
228 va_start (args, str0);
229 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
231 int len = strlen (next_str);
232 if (argcount < countof (saved_lengths))
233 saved_lengths[argcount++] = len;
237 p = ret = xmalloc (total_length + 1);
239 /* Copy the strings into the allocated space. */
242 va_start (args, str0);
243 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
246 if (argcount < countof (saved_lengths))
247 len = saved_lengths[argcount++];
249 len = strlen (next_str);
250 memcpy (p, next_str, len);
259 /* Format the provided time according to the specified format. The
260 format is a string with format elements supported by strftime. */
263 fmttime (time_t t, const char *fmt)
265 static char output[32];
266 struct tm *tm = localtime(&t);
269 if (!strftime(output, sizeof(output), fmt, tm))
274 /* Return pointer to a static char[] buffer in which zero-terminated
275 string-representation of TM (in form hh:mm:ss) is printed.
277 If TM is NULL, the current time will be used. */
282 return fmttime(t, "%H:%M:%S");
285 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
288 datetime_str (time_t t)
290 return fmttime(t, "%Y-%m-%d %H:%M:%S");
293 /* The Windows versions of the following two functions are defined in
294 mswindows.c. On MSDOS this function should never be called. */
296 #if !defined(WINDOWS) && !defined(MSDOS)
298 fork_to_background (void)
301 /* Whether we arrange our own version of opt.lfilename here. */
302 bool logfile_changed = false;
306 /* We must create the file immediately to avoid either a race
307 condition (which arises from using unique_name and failing to
308 use fopen_excl) or lying to the user about the log file name
309 (which arises from using unique_name, printing the name, and
310 using fopen_excl later on.) */
311 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
314 logfile_changed = true;
327 /* parent, no error */
328 printf (_("Continuing in background, pid %d.\n"), (int) pid);
330 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
331 exit (0); /* #### should we use _exit()? */
334 /* child: give up the privileges and keep running. */
336 freopen ("/dev/null", "r", stdin);
337 freopen ("/dev/null", "w", stdout);
338 freopen ("/dev/null", "w", stderr);
340 #endif /* !WINDOWS && !MSDOS */
342 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
343 specified with TM. The atime ("access time") is set to the current
347 touch (const char *file, time_t tm)
349 #ifdef HAVE_STRUCT_UTIMBUF
350 struct utimbuf times;
358 times.actime = time (NULL);
359 if (utime (file, ×) == -1)
360 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
363 /* Checks if FILE is a symbolic link, and removes it if it is. Does
364 nothing under MS-Windows. */
366 remove_link (const char *file)
371 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
373 DEBUGP (("Unlinking %s (symlink).\n", file));
376 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
377 file, strerror (errno));
382 /* Does FILENAME exist? This is quite a lousy implementation, since
383 it supplies no error codes -- only a yes-or-no answer. Thus it
384 will return that a file does not exist if, e.g., the directory is
385 unreadable. I don't mind it too much currently, though. The
386 proper way should, of course, be to have a third, error state,
387 other than true/false, but that would introduce uncalled-for
388 additional complexity to the callers. */
390 file_exists_p (const char *filename)
393 return access (filename, F_OK) >= 0;
396 return stat (filename, &buf) >= 0;
400 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
401 Returns 0 on error. */
403 file_non_directory_p (const char *path)
406 /* Use lstat() rather than stat() so that symbolic links pointing to
407 directories can be identified correctly. */
408 if (lstat (path, &buf) != 0)
410 return S_ISDIR (buf.st_mode) ? false : true;
413 /* Return the size of file named by FILENAME, or -1 if it cannot be
414 opened or seeked into. */
416 file_size (const char *filename)
418 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
420 /* We use fseek rather than stat to determine the file size because
421 that way we can also verify that the file is readable without
422 explicitly checking for permissions. Inspired by the POST patch
424 FILE *fp = fopen (filename, "rb");
427 fseeko (fp, 0, SEEK_END);
433 if (stat (filename, &st) < 0)
439 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
440 doesn't exist is found. Return a freshly allocated copy of the
444 unique_name_1 (const char *prefix)
447 int plen = strlen (prefix);
448 char *template = (char *)alloca (plen + 1 + 24);
449 char *template_tail = template + plen;
451 memcpy (template, prefix, plen);
452 *template_tail++ = '.';
455 number_to_string (template_tail, count++);
456 while (file_exists_p (template));
458 return xstrdup (template);
461 /* Return a unique file name, based on FILE.
463 More precisely, if FILE doesn't exist, it is returned unmodified.
464 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
465 file name that doesn't exist is returned.
467 The resulting file is not created, only verified that it didn't
468 exist at the point in time when the function was called.
469 Therefore, where security matters, don't rely that the file created
470 by this function exists until you open it with O_EXCL or
473 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
474 string. Otherwise, it may return FILE if the file doesn't exist
475 (and therefore doesn't need changing). */
478 unique_name (const char *file, bool allow_passthrough)
480 /* If the FILE itself doesn't exist, return it without
482 if (!file_exists_p (file))
483 return allow_passthrough ? (char *)file : xstrdup (file);
485 /* Otherwise, find a numeric suffix that results in unused file name
487 return unique_name_1 (file);
490 /* Create a file based on NAME, except without overwriting an existing
491 file with that name. Providing O_EXCL is correctly implemented,
492 this function does not have the race condition associated with
493 opening the file returned by unique_name. */
496 unique_create (const char *name, bool binary, char **opened_name)
498 /* unique file name, based on NAME */
499 char *uname = unique_name (name, false);
501 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
504 uname = unique_name (name, false);
506 if (opened_name && fp != NULL)
509 *opened_name = uname;
521 /* Open the file for writing, with the addition that the file is
522 opened "exclusively". This means that, if the file already exists,
523 this function will *fail* and errno will be set to EEXIST. If
524 BINARY is set, the file will be opened in binary mode, equivalent
527 If opening the file fails for any reason, including the file having
528 previously existed, this function returns NULL and sets errno
532 fopen_excl (const char *fname, bool binary)
536 int flags = O_WRONLY | O_CREAT | O_EXCL;
541 fd = open (fname, flags, 0666);
544 return fdopen (fd, binary ? "wb" : "w");
545 #else /* not O_EXCL */
546 /* Manually check whether the file exists. This is prone to race
547 conditions, but systems without O_EXCL haven't deserved
549 if (file_exists_p (fname))
554 return fopen (fname, binary ? "wb" : "w");
555 #endif /* not O_EXCL */
558 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
559 are missing, create them first. In case any mkdir() call fails,
560 return its error status. Returns 0 on successful completion.
562 The behaviour of this function should be identical to the behaviour
563 of `mkdir -p' on systems where mkdir supports the `-p' option. */
565 make_directory (const char *directory)
567 int i, ret, quit = 0;
570 /* Make a copy of dir, to be able to write to it. Otherwise, the
571 function is unsafe if called with a read-only char *argument. */
572 STRDUP_ALLOCA (dir, directory);
574 /* If the first character of dir is '/', skip it (and thus enable
575 creation of absolute-pathname directories. */
576 for (i = (*dir == '/'); 1; ++i)
578 for (; dir[i] && dir[i] != '/'; i++)
583 /* Check whether the directory already exists. Allow creation of
584 of intermediate directories to fail, as the initial path components
585 are not necessarily directories! */
586 if (!file_exists_p (dir))
587 ret = mkdir (dir, 0777);
598 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
599 should be a file name.
601 file_merge("/foo/bar", "baz") => "/foo/baz"
602 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
603 file_merge("foo", "bar") => "bar"
605 In other words, it's a simpler and gentler version of uri_merge. */
608 file_merge (const char *base, const char *file)
611 const char *cut = (const char *)strrchr (base, '/');
614 return xstrdup (file);
616 result = xmalloc (cut - base + 1 + strlen (file) + 1);
617 memcpy (result, base, cut - base);
618 result[cut - base] = '/';
619 strcpy (result + (cut - base) + 1, file);
624 /* Like fnmatch, but performs a case-insensitive match. */
627 fnmatch_nocase (const char *pattern, const char *string, int flags)
630 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
631 also present on *BSD platforms, and possibly elsewhere. */
632 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
634 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
635 char *patcopy = (char *) alloca (strlen (pattern) + 1);
636 char *strcopy = (char *) alloca (strlen (string) + 1);
638 for (p = patcopy; *pattern; pattern++, p++)
639 *p = TOLOWER (*pattern);
641 for (p = strcopy; *string; string++, p++)
642 *p = TOLOWER (*string);
644 return fnmatch (patcopy, strcopy, flags);
648 static bool in_acclist (const char *const *, const char *, bool);
650 /* Determine whether a file is acceptable to be followed, according to
651 lists of patterns to accept/reject. */
653 acceptable (const char *s)
657 while (l && s[l] != '/')
664 return (in_acclist ((const char *const *)opt.accepts, s, true)
665 && !in_acclist ((const char *const *)opt.rejects, s, true));
667 return in_acclist ((const char *const *)opt.accepts, s, true);
669 else if (opt.rejects)
670 return !in_acclist ((const char *const *)opt.rejects, s, true);
674 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
675 will return true if and only if D2 begins with `/something/' or is exactly
678 subdir_p (const char *d1, const char *d2)
680 if (!opt.ignore_case)
681 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
684 for (; *d1 && *d2 && (TOLOWER (*d1) == TOLOWER (*d2)); ++d1, ++d2)
687 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
690 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
691 first element that matches DIR, through wildcards or front comparison (as
694 dir_matches_p (char **dirlist, const char *dir)
697 int (*matcher) (const char *, const char *, int)
698 = opt.ignore_case ? fnmatch_nocase : fnmatch;
700 for (x = dirlist; *x; x++)
702 /* Remove leading '/' */
703 char *p = *x + (**x == '/');
704 if (has_wildcards_p (p))
706 if (matcher (p, dir, FNM_PATHNAME) == 0)
711 if (subdir_p (p, dir))
716 return *x ? true : false;
719 /* Returns whether DIRECTORY is acceptable for download, wrt the
720 include/exclude lists.
722 The leading `/' is ignored in paths; relative and absolute paths
723 may be freely intermixed. */
726 accdir (const char *directory)
728 /* Remove starting '/'. */
729 if (*directory == '/')
733 if (!dir_matches_p (opt.includes, directory))
738 if (dir_matches_p (opt.excludes, directory))
744 /* Return true if STRING ends with TAIL. For instance:
746 match_tail ("abc", "bc", false) -> 1
747 match_tail ("abc", "ab", false) -> 0
748 match_tail ("abc", "abc", false) -> 1
750 If FOLD_CASE is true, the comparison will be case-insensitive. */
753 match_tail (const char *string, const char *tail, bool fold_case)
757 /* We want this to be fast, so we code two loops, one with
758 case-folding, one without. */
762 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
763 if (string[i] != tail[j])
768 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
769 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
773 /* If the tail was exhausted, the match was succesful. */
780 /* Checks whether string S matches each element of ACCEPTS. A list
781 element are matched either with fnmatch() or match_tail(),
782 according to whether the element contains wildcards or not.
784 If the BACKWARD is false, don't do backward comparison -- just compare
787 in_acclist (const char *const *accepts, const char *s, bool backward)
789 for (; *accepts; accepts++)
791 if (has_wildcards_p (*accepts))
793 int res = opt.ignore_case
794 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
795 /* fnmatch returns 0 if the pattern *does* match the string. */
803 if (match_tail (s, *accepts, opt.ignore_case))
808 int cmp = opt.ignore_case
809 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
818 /* Return the location of STR's suffix (file extension). Examples:
819 suffix ("foo.bar") -> "bar"
820 suffix ("foo.bar.baz") -> "baz"
821 suffix ("/foo/bar") -> NULL
822 suffix ("/foo.bar/baz") -> NULL */
824 suffix (const char *str)
828 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
832 return (char *)str + i;
837 /* Return true if S contains globbing wildcards (`*', `?', `[' or
841 has_wildcards_p (const char *s)
844 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
849 /* Return true if FNAME ends with a typical HTML suffix. The
850 following (case-insensitive) suffixes are presumed to be HTML
855 ?html (`?' matches one character)
857 #### CAVEAT. This is not necessarily a good indication that FNAME
858 refers to a file that contains HTML! */
860 has_html_suffix_p (const char *fname)
864 if ((suf = suffix (fname)) == NULL)
866 if (!strcasecmp (suf, "html"))
868 if (!strcasecmp (suf, "htm"))
870 if (suf[0] && !strcasecmp (suf + 1, "html"))
875 /* Read a line from FP and return the pointer to freshly allocated
876 storage. The storage space is obtained through malloc() and should
877 be freed with free() when it is no longer needed.
879 The length of the line is not limited, except by available memory.
880 The newline character at the end of line is retained. The line is
881 terminated with a zero character.
883 After end-of-file is encountered without anything being read, NULL
884 is returned. NULL is also returned on error. To distinguish
885 between these two cases, use the stdio function ferror(). */
888 read_whole_line (FILE *fp)
892 char *line = xmalloc (bufsize);
894 while (fgets (line + length, bufsize - length, fp))
896 length += strlen (line + length);
898 /* Possible for example when reading from a binary file where
899 a line begins with \0. */
902 if (line[length - 1] == '\n')
905 /* fgets() guarantees to read the whole line, or to use up the
906 space we've given it. We can double the buffer
909 line = xrealloc (line, bufsize);
911 if (length == 0 || ferror (fp))
916 if (length + 1 < bufsize)
917 /* Relieve the memory from our exponential greediness. We say
918 `length + 1' because the terminating \0 is not included in
919 LENGTH. We don't need to zero-terminate the string ourselves,
920 though, because fgets() does that. */
921 line = xrealloc (line, length + 1);
925 /* Read FILE into memory. A pointer to `struct file_memory' are
926 returned; use struct element `content' to access file contents, and
927 the element `length' to know the file length. `content' is *not*
928 zero-terminated, and you should *not* read or write beyond the [0,
929 length) range of characters.
931 After you are done with the file contents, call read_file_free to
934 Depending on the operating system and the type of file that is
935 being read, read_file() either mmap's the file into memory, or
936 reads the file into the core using read().
938 If file is named "-", fileno(stdin) is used for reading instead.
939 If you want to read from a real file named "-", use "./-" instead. */
942 read_file (const char *file)
945 struct file_memory *fm;
947 bool inhibit_close = false;
949 /* Some magic in the finest tradition of Perl and its kin: if FILE
950 is "-", just use stdin. */
954 inhibit_close = true;
955 /* Note that we don't inhibit mmap() in this case. If stdin is
956 redirected from a regular file, mmap() will still work. */
959 fd = open (file, O_RDONLY);
962 fm = xnew (struct file_memory);
967 if (fstat (fd, &buf) < 0)
969 fm->length = buf.st_size;
970 /* NOTE: As far as I know, the callers of this function never
971 modify the file text. Relying on this would enable us to
972 specify PROT_READ and MAP_SHARED for a marginal gain in
973 efficiency, but at some cost to generality. */
974 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
976 if (fm->content == (char *)MAP_FAILED)
986 /* The most common reason why mmap() fails is that FD does not point
987 to a plain file. However, it's also possible that mmap() doesn't
988 work for a particular type of file. Therefore, whenever mmap()
989 fails, we just fall back to the regular method. */
990 #endif /* HAVE_MMAP */
993 size = 512; /* number of bytes fm->contents can
994 hold at any given time. */
995 fm->content = xmalloc (size);
999 if (fm->length > size / 2)
1001 /* #### I'm not sure whether the whole exponential-growth
1002 thing makes sense with kernel read. On Linux at least,
1003 read() refuses to read more than 4K from a file at a
1004 single chunk anyway. But other Unixes might optimize it
1005 better, and it doesn't *hurt* anything, so I'm leaving
1008 /* Normally, we grow SIZE exponentially to make the number
1009 of calls to read() and realloc() logarithmic in relation
1010 to file size. However, read() can read an amount of data
1011 smaller than requested, and it would be unreasonable to
1012 double SIZE every time *something* was read. Therefore,
1013 we double SIZE only when the length exceeds half of the
1014 entire allocated size. */
1016 fm->content = xrealloc (fm->content, size);
1018 nread = read (fd, fm->content + fm->length, size - fm->length);
1020 /* Successful read. */
1021 fm->length += nread;
1031 if (size > fm->length && fm->length != 0)
1032 /* Due to exponential growth of fm->content, the allocated region
1033 might be much larger than what is actually needed. */
1034 fm->content = xrealloc (fm->content, fm->length);
1041 xfree (fm->content);
1046 /* Release the resources held by FM. Specifically, this calls
1047 munmap() or xfree() on fm->content, depending whether mmap or
1048 malloc/read were used to read in the file. It also frees the
1049 memory needed to hold the FM structure itself. */
1052 read_file_free (struct file_memory *fm)
1057 munmap (fm->content, fm->length);
1062 xfree (fm->content);
1067 /* Free the pointers in a NULL-terminated vector of pointers, then
1068 free the pointer itself. */
1070 free_vec (char **vec)
1081 /* Append vector V2 to vector V1. The function frees V2 and
1082 reallocates V1 (thus you may not use the contents of neither
1083 pointer after the call). If V1 is NULL, V2 is returned. */
1085 merge_vecs (char **v1, char **v2)
1095 /* To avoid j == 0 */
1100 for (i = 0; v1[i]; i++)
1103 for (j = 0; v2[j]; j++)
1105 /* Reallocate v1. */
1106 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1107 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1112 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1113 is allocated as needed. Return the new value of the vector. */
1116 vec_append (char **vec, const char *str)
1118 int cnt; /* count of vector elements, including
1119 the one we're about to append */
1122 for (cnt = 0; vec[cnt]; cnt++)
1128 /* Reallocate the array to fit the new element and the NULL. */
1129 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1130 /* Append a copy of STR to the vector. */
1131 vec[cnt - 1] = xstrdup (str);
1136 /* Sometimes it's useful to create "sets" of strings, i.e. special
1137 hash tables where you want to store strings as keys and merely
1138 query for their existence. Here is a set of utility routines that
1139 makes that transparent. */
1142 string_set_add (struct hash_table *ht, const char *s)
1144 /* First check whether the set element already exists. If it does,
1145 do nothing so that we don't have to free() the old element and
1146 then strdup() a new one. */
1147 if (hash_table_contains (ht, s))
1150 /* We use "1" as value. It provides us a useful and clear arbitrary
1151 value, and it consumes no memory -- the pointers to the same
1152 string "1" will be shared by all the key-value pairs in all `set'
1154 hash_table_put (ht, xstrdup (s), "1");
1157 /* Synonym for hash_table_contains... */
1160 string_set_contains (struct hash_table *ht, const char *s)
1162 return hash_table_contains (ht, s);
1165 /* Convert the specified string set to array. ARRAY should be large
1166 enough to hold hash_table_count(ht) char pointers. */
1168 void string_set_to_array (struct hash_table *ht, char **array)
1170 hash_table_iterator iter;
1171 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1172 *array++ = iter.key;
1175 /* Free the string set. This frees both the storage allocated for
1176 keys and the actual hash table. (hash_table_destroy would only
1177 destroy the hash table.) */
1180 string_set_free (struct hash_table *ht)
1182 hash_table_iterator iter;
1183 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1185 hash_table_destroy (ht);
1188 /* Utility function: simply call xfree() on all keys and values of HT. */
1191 free_keys_and_values (struct hash_table *ht)
1193 hash_table_iterator iter;
1194 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1201 /* Get digit grouping data for thousand separors by calling
1202 localeconv(). The data includes separator string and grouping info
1203 and is cached after the first call to the function.
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 display
1248 separators in the "C" locale, still used by many Unix users. */
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 would overflow on WGINT_MIN, but printing negative numbers
1273 is not an important goal of this fuinction. */
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". We don't
1316 use the "*bibyte" names invented in 1998, and seldom used in
1317 practice. Wikipedia's entry on "binary prefix" discusses this in
1321 human_readable (HR_NUMTYPE n)
1323 /* These suffixes are compatible with those of GNU `ls -lh'. */
1324 static char powers[] =
1326 'K', /* kilobyte, 2^10 bytes */
1327 'M', /* megabyte, 2^20 bytes */
1328 'G', /* gigabyte, 2^30 bytes */
1329 'T', /* terabyte, 2^40 bytes */
1330 'P', /* petabyte, 2^50 bytes */
1331 'E', /* exabyte, 2^60 bytes */
1336 /* If the quantity is smaller than 1K, just print it. */
1339 snprintf (buf, sizeof (buf), "%d", (int) n);
1343 /* Loop over powers, dividing N with 1024 in each iteration. This
1344 works unchanged for all sizes of wgint, while still avoiding
1345 non-portable `long double' arithmetic. */
1346 for (i = 0; i < countof (powers); i++)
1348 /* At each iteration N is greater than the *subsequent* power.
1349 That way N/1024.0 produces a decimal number in the units of
1351 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1353 double val = n / 1024.0;
1354 /* Print values smaller than 10 with one decimal digits, and
1355 others without any decimals. */
1356 snprintf (buf, sizeof (buf), "%.*f%c",
1357 val < 10 ? 1 : 0, val, powers[i]);
1362 return NULL; /* unreached */
1365 /* Count the digits in the provided number. Used to allocate space
1366 when printing numbers. */
1369 numdigit (wgint number)
1373 ++cnt; /* accomodate '-' */
1374 while ((number /= 10) != 0)
1379 #define PR(mask) *p++ = n / (mask) + '0'
1381 /* DIGITS_<D> is used to print a D-digit number and should be called
1382 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1383 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1384 Recursively this continues until DIGITS_1 is invoked. */
1386 #define DIGITS_1(mask) PR (mask)
1387 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1388 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1389 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1390 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1391 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1392 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1393 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1394 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1395 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1397 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1399 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1400 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1401 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1402 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1403 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1404 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1405 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1406 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1407 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1409 /* Shorthand for casting to wgint. */
1412 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1413 `sprintf(buffer, "%lld", (long long) number)', only typically much
1414 faster and portable to machines without long long.
1416 The speedup may make a difference in programs that frequently
1417 convert numbers to strings. Some implementations of sprintf,
1418 particularly the one in some versions of GNU libc, have been known
1419 to be quite slow when converting integers to strings.
1421 Return the pointer to the location where the terminating zero was
1422 printed. (Equivalent to calling buffer+strlen(buffer) after the
1425 BUFFER should be large enough to accept as many bytes as you expect
1426 the number to take up. On machines with 64-bit wgints the maximum
1427 needed size is 24 bytes. That includes the digits needed for the
1428 largest 64-bit number, the `-' sign in case it's negative, and the
1429 terminating '\0'. */
1432 number_to_string (char *buffer, wgint number)
1437 int last_digit_char = 0;
1439 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1440 /* We are running in a very strange environment. Leave the correct
1441 printing to sprintf. */
1442 p += sprintf (buf, "%j", (intmax_t) (n));
1443 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1449 /* n = -n would overflow because -n would evaluate to a
1450 wgint value larger than WGINT_MAX. Need to make n
1451 smaller and handle the last digit separately. */
1452 int last_digit = n % 10;
1453 /* The sign of n%10 is implementation-defined. */
1455 last_digit_char = '0' - last_digit;
1457 last_digit_char = '0' + last_digit;
1458 /* After n is made smaller, -n will not overflow. */
1466 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1467 way printing any N is fully open-coded without a loop or jump.
1468 (Also see description of DIGITS_*.) */
1470 if (n < 10) DIGITS_1 (1);
1471 else if (n < 100) DIGITS_2 (10);
1472 else if (n < 1000) DIGITS_3 (100);
1473 else if (n < 10000) DIGITS_4 (1000);
1474 else if (n < 100000) DIGITS_5 (10000);
1475 else if (n < 1000000) DIGITS_6 (100000);
1476 else if (n < 10000000) DIGITS_7 (1000000);
1477 else if (n < 100000000) DIGITS_8 (10000000);
1478 else if (n < 1000000000) DIGITS_9 (100000000);
1479 #if SIZEOF_WGINT == 4
1480 /* wgint is 32 bits wide: no number has more than 10 digits. */
1481 else DIGITS_10 (1000000000);
1483 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1484 Constants are constructed by compile-time multiplication to avoid
1485 dealing with different notations for 64-bit constants
1486 (nL/nLL/nI64, depending on the compiler and architecture). */
1487 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1488 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1489 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1490 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1491 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1492 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1493 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1494 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1495 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1496 else DIGITS_19 (1000000000*(W)1000000000);
1499 if (last_digit_char)
1500 *p++ = last_digit_char;
1503 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1510 #undef SPRINTF_WGINT
1533 /* Print NUMBER to a statically allocated string and return a pointer
1534 to the printed representation.
1536 This function is intended to be used in conjunction with printf.
1537 It is hard to portably print wgint values:
1538 a) you cannot use printf("%ld", number) because wgint can be long
1539 long on 32-bit machines with LFS.
1540 b) you cannot use printf("%lld", number) because NUMBER could be
1541 long on 32-bit machines without LFS, or on 64-bit machines,
1542 which do not require LFS. Also, Windows doesn't support %lld.
1543 c) you cannot use printf("%j", (int_max_t) number) because not all
1544 versions of printf support "%j", the most notable being the one
1546 d) you cannot #define WGINT_FMT to the appropriate format and use
1547 printf(WGINT_FMT, number) because that would break translations
1548 for user-visible messages, such as printf("Downloaded: %d
1551 What you should use instead is printf("%s", number_to_static_string
1554 CAVEAT: since the function returns pointers to static data, you
1555 must be careful to copy its result before calling it again.
1556 However, to make it more useful with printf, the function maintains
1557 an internal ring of static buffers to return. That way things like
1558 printf("%s %s", number_to_static_string (num1),
1559 number_to_static_string (num2)) work as expected. Three buffers
1560 are currently used, which means that "%s %s %s" will work, but "%s
1561 %s %s %s" won't. If you need to print more than three wgints,
1562 bump the RING_SIZE (or rethink your message.) */
1565 number_to_static_string (wgint number)
1567 static char ring[RING_SIZE][24];
1569 char *buf = ring[ringpos];
1570 number_to_string (buf, number);
1571 ringpos = (ringpos + 1) % RING_SIZE;
1575 /* Determine the width of the terminal we're running on. If that's
1576 not possible, return 0. */
1579 determine_screen_width (void)
1581 /* If there's a way to get the terminal size using POSIX
1582 tcgetattr(), somebody please tell me. */
1587 if (opt.lfilename != NULL)
1590 fd = fileno (stderr);
1591 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1592 return 0; /* most likely ENOTTY */
1595 #elif defined(WINDOWS)
1596 CONSOLE_SCREEN_BUFFER_INFO csbi;
1597 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1599 return csbi.dwSize.X;
1600 #else /* neither TIOCGWINSZ nor WINDOWS */
1602 #endif /* neither TIOCGWINSZ nor WINDOWS */
1605 /* Whether the rnd system (either rand or [dl]rand48) has been
1607 static int rnd_seeded;
1609 /* Return a random number between 0 and MAX-1, inclusive.
1611 If the system does not support lrand48 and MAX is greater than the
1612 value of RAND_MAX+1 on the system, the returned value will be in
1613 the range [0, RAND_MAX]. This may be fixed in a future release.
1614 The random number generator is seeded automatically the first time
1617 This uses lrand48 where available, rand elsewhere. DO NOT use it
1618 for cryptography. It is only meant to be used in situations where
1619 quality of the random numbers returned doesn't really matter. */
1622 random_number (int max)
1627 srand48 ((long) time (NULL) ^ (long) getpid ());
1630 return lrand48 () % max;
1631 #else /* not HAVE_DRAND48 */
1637 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1642 /* Like rand() % max, but uses the high-order bits for better
1643 randomness on architectures where rand() is implemented using a
1644 simple congruential generator. */
1646 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1647 return (int) bounded;
1649 #endif /* not HAVE_DRAND48 */
1652 /* Return a random uniformly distributed floating point number in the
1653 [0, 1) range. Uses drand48 where available, and a really lame
1654 kludge elsewhere. */
1662 srand48 ((long) time (NULL) ^ (long) getpid ());
1666 #else /* not HAVE_DRAND48 */
1667 return ( random_number (10000) / 10000.0
1668 + random_number (10000) / (10000.0 * 10000.0)
1669 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1670 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1671 #endif /* not HAVE_DRAND48 */
1674 /* Implementation of run_with_timeout, a generic timeout-forcing
1675 routine for systems with Unix-like signal handling. */
1677 #ifdef USE_SIGNAL_TIMEOUT
1678 # ifdef HAVE_SIGSETJMP
1679 # define SETJMP(env) sigsetjmp (env, 1)
1681 static sigjmp_buf run_with_timeout_env;
1684 abort_run_with_timeout (int sig)
1686 assert (sig == SIGALRM);
1687 siglongjmp (run_with_timeout_env, -1);
1689 # else /* not HAVE_SIGSETJMP */
1690 # define SETJMP(env) setjmp (env)
1692 static jmp_buf run_with_timeout_env;
1695 abort_run_with_timeout (int sig)
1697 assert (sig == SIGALRM);
1698 /* We don't have siglongjmp to preserve the set of blocked signals;
1699 if we longjumped out of the handler at this point, SIGALRM would
1700 remain blocked. We must unblock it manually. */
1701 int mask = siggetmask ();
1702 mask &= ~sigmask (SIGALRM);
1705 /* Now it's safe to longjump. */
1706 longjmp (run_with_timeout_env, -1);
1708 # endif /* not HAVE_SIGSETJMP */
1710 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1711 setitimer where available, alarm otherwise.
1713 TIMEOUT should be non-zero. If the timeout value is so small that
1714 it would be rounded to zero, it is rounded to the least legal value
1715 instead (1us for setitimer, 1s for alarm). That ensures that
1716 SIGALRM will be delivered in all cases. */
1719 alarm_set (double timeout)
1722 /* Use the modern itimer interface. */
1723 struct itimerval itv;
1725 itv.it_value.tv_sec = (long) timeout;
1726 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1727 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1728 /* Ensure that we wait for at least the minimum interval.
1729 Specifying zero would mean "wait forever". */
1730 itv.it_value.tv_usec = 1;
1731 setitimer (ITIMER_REAL, &itv, NULL);
1732 #else /* not ITIMER_REAL */
1733 /* Use the old alarm() interface. */
1734 int secs = (int) timeout;
1736 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1737 because alarm(0) means "never deliver the alarm", i.e. "wait
1738 forever", which is not what someone who specifies a 0.5s
1739 timeout would expect. */
1742 #endif /* not ITIMER_REAL */
1745 /* Cancel the alarm set with alarm_set. */
1751 struct itimerval disable;
1753 setitimer (ITIMER_REAL, &disable, NULL);
1754 #else /* not ITIMER_REAL */
1756 #endif /* not ITIMER_REAL */
1759 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1760 seconds. Returns true if the function was interrupted with a
1761 timeout, false otherwise.
1763 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1764 using setitimer() or alarm(). The timeout is enforced by
1765 longjumping out of the SIGALRM handler. This has several
1766 advantages compared to the traditional approach of relying on
1767 signals causing system calls to exit with EINTR:
1769 * The callback function is *forcibly* interrupted after the
1770 timeout expires, (almost) regardless of what it was doing and
1771 whether it was in a syscall. For example, a calculation that
1772 takes a long time is interrupted as reliably as an IO
1775 * It works with both SYSV and BSD signals because it doesn't
1776 depend on the default setting of SA_RESTART.
1778 * It doesn't require special handler setup beyond a simple call
1779 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1782 The only downside is that, if FUN allocates internal resources that
1783 are normally freed prior to exit from the functions, they will be
1784 lost in case of timeout. */
1787 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1797 signal (SIGALRM, abort_run_with_timeout);
1798 if (SETJMP (run_with_timeout_env) != 0)
1800 /* Longjumped out of FUN with a timeout. */
1801 signal (SIGALRM, SIG_DFL);
1804 alarm_set (timeout);
1807 /* Preserve errno in case alarm() or signal() modifies it. */
1808 saved_errno = errno;
1810 signal (SIGALRM, SIG_DFL);
1811 errno = saved_errno;
1816 #else /* not USE_SIGNAL_TIMEOUT */
1819 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1820 define it under Windows, because Windows has its own version of
1821 run_with_timeout that uses threads. */
1824 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1829 #endif /* not WINDOWS */
1830 #endif /* not USE_SIGNAL_TIMEOUT */
1834 /* Sleep the specified amount of seconds. On machines without
1835 nanosleep(), this may sleep shorter if interrupted by signals. */
1838 xsleep (double seconds)
1840 #ifdef HAVE_NANOSLEEP
1841 /* nanosleep is the preferred interface because it offers high
1842 accuracy and, more importantly, because it allows us to reliably
1843 restart receiving a signal such as SIGWINCH. (There was an
1844 actual Debian bug report about --limit-rate malfunctioning while
1845 the terminal was being resized.) */
1846 struct timespec sleep, remaining;
1847 sleep.tv_sec = (long) seconds;
1848 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1849 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1850 /* If nanosleep has been interrupted by a signal, adjust the
1851 sleeping period and return to sleep. */
1853 #elif defined(HAVE_USLEEP)
1854 /* If usleep is available, use it in preference to select. */
1857 /* On some systems, usleep cannot handle values larger than
1858 1,000,000. If the period is larger than that, use sleep
1859 first, then add usleep for subsecond accuracy. */
1861 seconds -= (long) seconds;
1863 usleep (seconds * 1000000);
1864 #else /* fall back select */
1865 /* Note that, although Windows supports select, it can't be used to
1866 implement sleeping because Winsock's select doesn't implement
1867 timeout when it is passed NULL pointers for all fd sets. (But it
1868 does under Cygwin, which implements Unix-compatible select.) */
1869 struct timeval sleep;
1870 sleep.tv_sec = (long) seconds;
1871 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1872 select (0, NULL, NULL, NULL, &sleep);
1873 /* If select returns -1 and errno is EINTR, it means we were
1874 interrupted by a signal. But without knowing how long we've
1875 actually slept, we can't return to sleep. Using gettimeofday to
1876 track sleeps is slow and unreliable due to clock skew. */
1880 #endif /* not WINDOWS */
1882 /* Encode the octets in DATA of length LENGTH to base64 format,
1883 storing the result to DEST. The output will be zero-terminated,
1884 and must point to a writable buffer of at least
1885 1+BASE64_LENGTH(length) bytes. The function returns the length of
1886 the resulting base64 data, not counting the terminating zero.
1888 This implementation does not emit newlines after 76 characters of
1892 base64_encode (const void *data, int length, char *dest)
1894 /* Conversion table. */
1895 static const char tbl[64] = {
1896 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
1897 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
1898 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
1899 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
1901 /* Access bytes in DATA as unsigned char, otherwise the shifts below
1902 don't work for data with MSB set. */
1903 const unsigned char *s = data;
1904 /* Theoretical ANSI violation when length < 3. */
1905 const unsigned char *end = (const unsigned char *) data + length - 2;
1908 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1909 for (; s < end; s += 3)
1911 *p++ = tbl[s[0] >> 2];
1912 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1913 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1914 *p++ = tbl[s[2] & 0x3f];
1917 /* Pad the result if necessary... */
1921 *p++ = tbl[s[0] >> 2];
1922 *p++ = tbl[(s[0] & 3) << 4];
1927 *p++ = tbl[s[0] >> 2];
1928 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1929 *p++ = tbl[((s[1] & 0xf) << 2)];
1933 /* ...and zero-terminate it. */
1939 /* Store in C the next non-whitespace character from the string, or \0
1940 when end of string is reached. */
1941 #define NEXT_CHAR(c, p) do { \
1942 c = (unsigned char) *p++; \
1943 } while (ISSPACE (c))
1945 #define IS_ASCII(c) (((c) & 0x80) == 0)
1947 /* Decode data from BASE64 (a null-terminated string) into memory
1948 pointed to by DEST. DEST is assumed to be large enough to
1949 accomodate the decoded data, which is guaranteed to be no more than
1952 Since DEST is assumed to contain binary data, it is not
1953 NUL-terminated. The function returns the length of the data
1954 written to TO. -1 is returned in case of error caused by malformed
1957 This function originates from Free Recode. */
1960 base64_decode (const char *base64, void *dest)
1962 /* Table of base64 values for first 128 characters. Note that this
1963 assumes ASCII (but so does Wget in other places). */
1964 static const signed char base64_char_to_value[128] =
1966 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1967 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1968 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1969 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1970 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1971 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1972 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1973 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1974 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1975 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1976 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1977 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1978 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1980 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
1981 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
1983 const char *p = base64;
1989 unsigned long value;
1991 /* Process first byte of a quadruplet. */
1995 if (c == '=' || !IS_BASE64 (c))
1996 return -1; /* illegal char while decoding base64 */
1997 value = BASE64_CHAR_TO_VALUE (c) << 18;
1999 /* Process second byte of a quadruplet. */
2002 return -1; /* premature EOF while decoding base64 */
2003 if (c == '=' || !IS_BASE64 (c))
2004 return -1; /* illegal char while decoding base64 */
2005 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2008 /* Process third byte of a quadruplet. */
2011 return -1; /* premature EOF while decoding base64 */
2013 return -1; /* illegal char while decoding base64 */
2019 return -1; /* premature EOF while decoding base64 */
2021 return -1; /* padding `=' expected but not found */
2025 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2026 *q++ = 0xff & value >> 8;
2028 /* Process fourth byte of a quadruplet. */
2031 return -1; /* premature EOF while decoding base64 */
2035 return -1; /* illegal char while decoding base64 */
2037 value |= BASE64_CHAR_TO_VALUE (c);
2038 *q++ = 0xff & value;
2041 #undef BASE64_CHAR_TO_VALUE
2043 return q - (char *) dest;
2049 /* Simple merge sort for use by stable_sort. Implementation courtesy
2050 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2053 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2054 int (*cmpfun) (const void *, const void *))
2056 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2060 size_t mid = (to + from) / 2;
2061 mergesort_internal (base, temp, size, from, mid, cmpfun);
2062 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2065 for (k = from; (i <= mid) && (j <= to); k++)
2066 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2067 memcpy (ELT (temp, k), ELT (base, i++), size);
2069 memcpy (ELT (temp, k), ELT (base, j++), size);
2071 memcpy (ELT (temp, k++), ELT (base, i++), size);
2073 memcpy (ELT (temp, k++), ELT (base, j++), size);
2074 for (k = from; k <= to; k++)
2075 memcpy (ELT (base, k), ELT (temp, k), size);
2080 /* Stable sort with interface exactly like standard library's qsort.
2081 Uses mergesort internally, allocating temporary storage with
2085 stable_sort (void *base, size_t nmemb, size_t size,
2086 int (*cmpfun) (const void *, const void *))
2090 void *temp = alloca (nmemb * size * sizeof (void *));
2091 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2095 /* Print a decimal number. If it is equal to or larger than ten, the
2096 number is rounded. Otherwise it is printed with one significant
2097 digit without trailing zeros and with no more than three fractional
2098 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2099 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2101 This is useful for displaying durations because it provides
2102 order-of-magnitude information without unnecessary clutter --
2103 long-running downloads are shown without the fractional part, and
2104 short ones still retain one significant digit. */
2107 print_decimal (double number)
2109 static char buf[32];
2110 double n = number >= 0 ? number : -number;
2113 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2114 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2115 snprintf (buf, sizeof buf, "%.0f", number);
2117 snprintf (buf, sizeof buf, "%.1f", number);
2118 else if (n >= 0.001)
2119 snprintf (buf, sizeof buf, "%.1g", number);
2120 else if (n >= 0.0005)
2121 /* round [0.0005, 0.001) to 0.001 */
2122 snprintf (buf, sizeof buf, "%.3f", number);
2124 /* print numbers close to 0 as 0, not 0.000 */
2141 { "/somedir", "/somedir", true },
2142 { "/somedir", "/somedir/d2", true },
2143 { "/somedir/d1", "/somedir", false },
2146 for (i = 0; i < countof(test_array); ++i)
2148 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2150 mu_assert ("test_subdir_p: wrong result",
2151 res == test_array[i].result);
2158 test_dir_matches_p()
2166 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2167 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2168 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2169 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2170 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2173 for (i = 0; i < countof(test_array); ++i)
2175 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2177 mu_assert ("test_dir_matches_p: wrong result",
2178 res == test_array[i].result);
2184 #endif /* TESTING */