1 /* Various utility functions.
2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 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 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
37 #ifdef HAVE_SYS_TIME_H
38 # include <sys/time.h>
44 # include <sys/mman.h>
47 # include <process.h> /* getpid() */
52 #ifdef HAVE_SYS_UTIME_H
53 # include <sys/utime.h>
61 /* For TIOCGWINSZ and friends: */
62 #ifdef HAVE_SYS_IOCTL_H
63 # include <sys/ioctl.h>
69 /* Needed for Unix version of run_with_timeout. */
73 #ifndef HAVE_SIGSETJMP
74 /* If sigsetjmp is a macro, configure won't pick it up. */
76 # define HAVE_SIGSETJMP
80 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
81 # 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 ','. */
138 while (c_isspace (*s))
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 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* 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;
304 if (opt.quiet && !opt.server_response)
306 /* Don't bother with a logfile, there are virtually no logs we
307 issue in quiet mode. (Server responses in FTP are the
308 exception, when enabled.) */
313 /* We must create the file immediately to avoid either a race
314 condition (which arises from using unique_name and failing to
315 use fopen_excl) or lying to the user about the log file name
316 (which arises from using unique_name, printing the name, and
317 using fopen_excl later on.) */
318 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
321 logfile_changed = true;
334 /* parent, no error */
336 printf (_("Continuing in background, pid %d.\n"), (int) pid);
338 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
339 exit (0); /* #### should we use _exit()? */
342 /* child: give up the privileges and keep running. */
344 freopen ("/dev/null", "r", stdin);
345 freopen ("/dev/null", "w", stdout);
346 freopen ("/dev/null", "w", stderr);
348 #endif /* !WINDOWS && !MSDOS */
350 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
351 specified with TM. The atime ("access time") is set to the current
355 touch (const char *file, time_t tm)
357 #ifdef HAVE_STRUCT_UTIMBUF
358 struct utimbuf times;
366 times.actime = time (NULL);
367 if (utime (file, ×) == -1)
368 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
371 /* Checks if FILE is a symbolic link, and removes it if it is. Does
372 nothing under MS-Windows. */
374 remove_link (const char *file)
379 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
381 DEBUGP (("Unlinking %s (symlink).\n", file));
384 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
385 file, strerror (errno));
390 /* Does FILENAME exist? This is quite a lousy implementation, since
391 it supplies no error codes -- only a yes-or-no answer. Thus it
392 will return that a file does not exist if, e.g., the directory is
393 unreadable. I don't mind it too much currently, though. The
394 proper way should, of course, be to have a third, error state,
395 other than true/false, but that would introduce uncalled-for
396 additional complexity to the callers. */
398 file_exists_p (const char *filename)
401 return access (filename, F_OK) >= 0;
404 return stat (filename, &buf) >= 0;
408 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
409 Returns 0 on error. */
411 file_non_directory_p (const char *path)
414 /* Use lstat() rather than stat() so that symbolic links pointing to
415 directories can be identified correctly. */
416 if (lstat (path, &buf) != 0)
418 return S_ISDIR (buf.st_mode) ? false : true;
421 /* Return the size of file named by FILENAME, or -1 if it cannot be
422 opened or seeked into. */
424 file_size (const char *filename)
426 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
428 /* We use fseek rather than stat to determine the file size because
429 that way we can also verify that the file is readable without
430 explicitly checking for permissions. Inspired by the POST patch
432 FILE *fp = fopen (filename, "rb");
435 fseeko (fp, 0, SEEK_END);
441 if (stat (filename, &st) < 0)
447 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
448 doesn't exist is found. Return a freshly allocated copy of the
452 unique_name_1 (const char *prefix)
455 int plen = strlen (prefix);
456 char *template = (char *)alloca (plen + 1 + 24);
457 char *template_tail = template + plen;
459 memcpy (template, prefix, plen);
460 *template_tail++ = '.';
463 number_to_string (template_tail, count++);
464 while (file_exists_p (template));
466 return xstrdup (template);
469 /* Return a unique file name, based on FILE.
471 More precisely, if FILE doesn't exist, it is returned unmodified.
472 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
473 file name that doesn't exist is returned.
475 The resulting file is not created, only verified that it didn't
476 exist at the point in time when the function was called.
477 Therefore, where security matters, don't rely that the file created
478 by this function exists until you open it with O_EXCL or
481 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
482 string. Otherwise, it may return FILE if the file doesn't exist
483 (and therefore doesn't need changing). */
486 unique_name (const char *file, bool allow_passthrough)
488 /* If the FILE itself doesn't exist, return it without
490 if (!file_exists_p (file))
491 return allow_passthrough ? (char *)file : xstrdup (file);
493 /* Otherwise, find a numeric suffix that results in unused file name
495 return unique_name_1 (file);
498 /* Create a file based on NAME, except without overwriting an existing
499 file with that name. Providing O_EXCL is correctly implemented,
500 this function does not have the race condition associated with
501 opening the file returned by unique_name. */
504 unique_create (const char *name, bool binary, char **opened_name)
506 /* unique file name, based on NAME */
507 char *uname = unique_name (name, false);
509 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
512 uname = unique_name (name, false);
514 if (opened_name && fp != NULL)
517 *opened_name = uname;
529 /* Open the file for writing, with the addition that the file is
530 opened "exclusively". This means that, if the file already exists,
531 this function will *fail* and errno will be set to EEXIST. If
532 BINARY is set, the file will be opened in binary mode, equivalent
535 If opening the file fails for any reason, including the file having
536 previously existed, this function returns NULL and sets errno
540 fopen_excl (const char *fname, bool binary)
544 int flags = O_WRONLY | O_CREAT | O_EXCL;
549 fd = open (fname, flags, 0666);
552 return fdopen (fd, binary ? "wb" : "w");
553 #else /* not O_EXCL */
554 /* Manually check whether the file exists. This is prone to race
555 conditions, but systems without O_EXCL haven't deserved
557 if (file_exists_p (fname))
562 return fopen (fname, binary ? "wb" : "w");
563 #endif /* not O_EXCL */
566 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
567 are missing, create them first. In case any mkdir() call fails,
568 return its error status. Returns 0 on successful completion.
570 The behaviour of this function should be identical to the behaviour
571 of `mkdir -p' on systems where mkdir supports the `-p' option. */
573 make_directory (const char *directory)
575 int i, ret, quit = 0;
578 /* Make a copy of dir, to be able to write to it. Otherwise, the
579 function is unsafe if called with a read-only char *argument. */
580 STRDUP_ALLOCA (dir, directory);
582 /* If the first character of dir is '/', skip it (and thus enable
583 creation of absolute-pathname directories. */
584 for (i = (*dir == '/'); 1; ++i)
586 for (; dir[i] && dir[i] != '/'; i++)
591 /* Check whether the directory already exists. Allow creation of
592 of intermediate directories to fail, as the initial path components
593 are not necessarily directories! */
594 if (!file_exists_p (dir))
595 ret = mkdir (dir, 0777);
606 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
607 should be a file name.
609 file_merge("/foo/bar", "baz") => "/foo/baz"
610 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
611 file_merge("foo", "bar") => "bar"
613 In other words, it's a simpler and gentler version of uri_merge. */
616 file_merge (const char *base, const char *file)
619 const char *cut = (const char *)strrchr (base, '/');
622 return xstrdup (file);
624 result = xmalloc (cut - base + 1 + strlen (file) + 1);
625 memcpy (result, base, cut - base);
626 result[cut - base] = '/';
627 strcpy (result + (cut - base) + 1, file);
632 /* Like fnmatch, but performs a case-insensitive match. */
635 fnmatch_nocase (const char *pattern, const char *string, int flags)
638 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
639 also present on *BSD platforms, and possibly elsewhere. */
640 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
642 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
643 char *patcopy = (char *) alloca (strlen (pattern) + 1);
644 char *strcopy = (char *) alloca (strlen (string) + 1);
646 for (p = patcopy; *pattern; pattern++, p++)
647 *p = c_tolower (*pattern);
649 for (p = strcopy; *string; string++, p++)
650 *p = c_tolower (*string);
652 return fnmatch (patcopy, strcopy, flags);
656 static bool in_acclist (const char *const *, const char *, bool);
658 /* Determine whether a file is acceptable to be followed, according to
659 lists of patterns to accept/reject. */
661 acceptable (const char *s)
665 while (l && s[l] != '/')
672 return (in_acclist ((const char *const *)opt.accepts, s, true)
673 && !in_acclist ((const char *const *)opt.rejects, s, true));
675 return in_acclist ((const char *const *)opt.accepts, s, true);
677 else if (opt.rejects)
678 return !in_acclist ((const char *const *)opt.rejects, s, true);
682 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
683 will return true if and only if D2 begins with `/something/' or is exactly
686 subdir_p (const char *d1, const char *d2)
690 if (!opt.ignore_case)
691 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
694 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
697 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
700 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
701 first element that matches DIR, through wildcards or front comparison (as
704 dir_matches_p (char **dirlist, const char *dir)
707 int (*matcher) (const char *, const char *, int)
708 = opt.ignore_case ? fnmatch_nocase : fnmatch;
710 for (x = dirlist; *x; x++)
712 /* Remove leading '/' */
713 char *p = *x + (**x == '/');
714 if (has_wildcards_p (p))
716 if (matcher (p, dir, FNM_PATHNAME) == 0)
721 if (subdir_p (p, dir))
726 return *x ? true : false;
729 /* Returns whether DIRECTORY is acceptable for download, wrt the
730 include/exclude lists.
732 The leading `/' is ignored in paths; relative and absolute paths
733 may be freely intermixed. */
736 accdir (const char *directory)
738 /* Remove starting '/'. */
739 if (*directory == '/')
743 if (!dir_matches_p (opt.includes, directory))
748 if (dir_matches_p (opt.excludes, directory))
754 /* Return true if STRING ends with TAIL. For instance:
756 match_tail ("abc", "bc", false) -> 1
757 match_tail ("abc", "ab", false) -> 0
758 match_tail ("abc", "abc", false) -> 1
760 If FOLD_CASE is true, the comparison will be case-insensitive. */
763 match_tail (const char *string, const char *tail, bool fold_case)
767 /* We want this to be fast, so we code two loops, one with
768 case-folding, one without. */
772 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
773 if (string[i] != tail[j])
778 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
779 if (c_tolower (string[i]) != c_tolower (tail[j]))
783 /* If the tail was exhausted, the match was succesful. */
790 /* Checks whether string S matches each element of ACCEPTS. A list
791 element are matched either with fnmatch() or match_tail(),
792 according to whether the element contains wildcards or not.
794 If the BACKWARD is false, don't do backward comparison -- just compare
797 in_acclist (const char *const *accepts, const char *s, bool backward)
799 for (; *accepts; accepts++)
801 if (has_wildcards_p (*accepts))
803 int res = opt.ignore_case
804 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
805 /* fnmatch returns 0 if the pattern *does* match the string. */
813 if (match_tail (s, *accepts, opt.ignore_case))
818 int cmp = opt.ignore_case
819 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
828 /* Return the location of STR's suffix (file extension). Examples:
829 suffix ("foo.bar") -> "bar"
830 suffix ("foo.bar.baz") -> "baz"
831 suffix ("/foo/bar") -> NULL
832 suffix ("/foo.bar/baz") -> NULL */
834 suffix (const char *str)
838 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
842 return (char *)str + i;
847 /* Return true if S contains globbing wildcards (`*', `?', `[' or
851 has_wildcards_p (const char *s)
854 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
859 /* Return true if FNAME ends with a typical HTML suffix. The
860 following (case-insensitive) suffixes are presumed to be HTML
865 ?html (`?' matches one character)
867 #### CAVEAT. This is not necessarily a good indication that FNAME
868 refers to a file that contains HTML! */
870 has_html_suffix_p (const char *fname)
874 if ((suf = suffix (fname)) == NULL)
876 if (!strcasecmp (suf, "html"))
878 if (!strcasecmp (suf, "htm"))
880 if (suf[0] && !strcasecmp (suf + 1, "html"))
885 /* Read a line from FP and return the pointer to freshly allocated
886 storage. The storage space is obtained through malloc() and should
887 be freed with free() when it is no longer needed.
889 The length of the line is not limited, except by available memory.
890 The newline character at the end of line is retained. The line is
891 terminated with a zero character.
893 After end-of-file is encountered without anything being read, NULL
894 is returned. NULL is also returned on error. To distinguish
895 between these two cases, use the stdio function ferror(). */
898 read_whole_line (FILE *fp)
902 char *line = xmalloc (bufsize);
904 while (fgets (line + length, bufsize - length, fp))
906 length += strlen (line + length);
908 /* Possible for example when reading from a binary file where
909 a line begins with \0. */
912 if (line[length - 1] == '\n')
915 /* fgets() guarantees to read the whole line, or to use up the
916 space we've given it. We can double the buffer
919 line = xrealloc (line, bufsize);
921 if (length == 0 || ferror (fp))
926 if (length + 1 < bufsize)
927 /* Relieve the memory from our exponential greediness. We say
928 `length + 1' because the terminating \0 is not included in
929 LENGTH. We don't need to zero-terminate the string ourselves,
930 though, because fgets() does that. */
931 line = xrealloc (line, length + 1);
935 /* Read FILE into memory. A pointer to `struct file_memory' are
936 returned; use struct element `content' to access file contents, and
937 the element `length' to know the file length. `content' is *not*
938 zero-terminated, and you should *not* read or write beyond the [0,
939 length) range of characters.
941 After you are done with the file contents, call read_file_free to
944 Depending on the operating system and the type of file that is
945 being read, read_file() either mmap's the file into memory, or
946 reads the file into the core using read().
948 If file is named "-", fileno(stdin) is used for reading instead.
949 If you want to read from a real file named "-", use "./-" instead. */
952 read_file (const char *file)
955 struct file_memory *fm;
957 bool inhibit_close = false;
959 /* Some magic in the finest tradition of Perl and its kin: if FILE
960 is "-", just use stdin. */
964 inhibit_close = true;
965 /* Note that we don't inhibit mmap() in this case. If stdin is
966 redirected from a regular file, mmap() will still work. */
969 fd = open (file, O_RDONLY);
972 fm = xnew (struct file_memory);
977 if (fstat (fd, &buf) < 0)
979 fm->length = buf.st_size;
980 /* NOTE: As far as I know, the callers of this function never
981 modify the file text. Relying on this would enable us to
982 specify PROT_READ and MAP_SHARED for a marginal gain in
983 efficiency, but at some cost to generality. */
984 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
986 if (fm->content == (char *)MAP_FAILED)
996 /* The most common reason why mmap() fails is that FD does not point
997 to a plain file. However, it's also possible that mmap() doesn't
998 work for a particular type of file. Therefore, whenever mmap()
999 fails, we just fall back to the regular method. */
1000 #endif /* HAVE_MMAP */
1003 size = 512; /* number of bytes fm->contents can
1004 hold at any given time. */
1005 fm->content = xmalloc (size);
1009 if (fm->length > size / 2)
1011 /* #### I'm not sure whether the whole exponential-growth
1012 thing makes sense with kernel read. On Linux at least,
1013 read() refuses to read more than 4K from a file at a
1014 single chunk anyway. But other Unixes might optimize it
1015 better, and it doesn't *hurt* anything, so I'm leaving
1018 /* Normally, we grow SIZE exponentially to make the number
1019 of calls to read() and realloc() logarithmic in relation
1020 to file size. However, read() can read an amount of data
1021 smaller than requested, and it would be unreasonable to
1022 double SIZE every time *something* was read. Therefore,
1023 we double SIZE only when the length exceeds half of the
1024 entire allocated size. */
1026 fm->content = xrealloc (fm->content, size);
1028 nread = read (fd, fm->content + fm->length, size - fm->length);
1030 /* Successful read. */
1031 fm->length += nread;
1041 if (size > fm->length && fm->length != 0)
1042 /* Due to exponential growth of fm->content, the allocated region
1043 might be much larger than what is actually needed. */
1044 fm->content = xrealloc (fm->content, fm->length);
1051 xfree (fm->content);
1056 /* Release the resources held by FM. Specifically, this calls
1057 munmap() or xfree() on fm->content, depending whether mmap or
1058 malloc/read were used to read in the file. It also frees the
1059 memory needed to hold the FM structure itself. */
1062 read_file_free (struct file_memory *fm)
1067 munmap (fm->content, fm->length);
1072 xfree (fm->content);
1077 /* Free the pointers in a NULL-terminated vector of pointers, then
1078 free the pointer itself. */
1080 free_vec (char **vec)
1091 /* Append vector V2 to vector V1. The function frees V2 and
1092 reallocates V1 (thus you may not use the contents of neither
1093 pointer after the call). If V1 is NULL, V2 is returned. */
1095 merge_vecs (char **v1, char **v2)
1105 /* To avoid j == 0 */
1110 for (i = 0; v1[i]; i++)
1113 for (j = 0; v2[j]; j++)
1115 /* Reallocate v1. */
1116 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1117 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1122 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1123 is allocated as needed. Return the new value of the vector. */
1126 vec_append (char **vec, const char *str)
1128 int cnt; /* count of vector elements, including
1129 the one we're about to append */
1132 for (cnt = 0; vec[cnt]; cnt++)
1138 /* Reallocate the array to fit the new element and the NULL. */
1139 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1140 /* Append a copy of STR to the vector. */
1141 vec[cnt - 1] = xstrdup (str);
1146 /* Sometimes it's useful to create "sets" of strings, i.e. special
1147 hash tables where you want to store strings as keys and merely
1148 query for their existence. Here is a set of utility routines that
1149 makes that transparent. */
1152 string_set_add (struct hash_table *ht, const char *s)
1154 /* First check whether the set element already exists. If it does,
1155 do nothing so that we don't have to free() the old element and
1156 then strdup() a new one. */
1157 if (hash_table_contains (ht, s))
1160 /* We use "1" as value. It provides us a useful and clear arbitrary
1161 value, and it consumes no memory -- the pointers to the same
1162 string "1" will be shared by all the key-value pairs in all `set'
1164 hash_table_put (ht, xstrdup (s), "1");
1167 /* Synonym for hash_table_contains... */
1170 string_set_contains (struct hash_table *ht, const char *s)
1172 return hash_table_contains (ht, s);
1175 /* Convert the specified string set to array. ARRAY should be large
1176 enough to hold hash_table_count(ht) char pointers. */
1178 void string_set_to_array (struct hash_table *ht, char **array)
1180 hash_table_iterator iter;
1181 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1182 *array++ = iter.key;
1185 /* Free the string set. This frees both the storage allocated for
1186 keys and the actual hash table. (hash_table_destroy would only
1187 destroy the hash table.) */
1190 string_set_free (struct hash_table *ht)
1192 hash_table_iterator iter;
1193 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1195 hash_table_destroy (ht);
1198 /* Utility function: simply call xfree() on all keys and values of HT. */
1201 free_keys_and_values (struct hash_table *ht)
1203 hash_table_iterator iter;
1204 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1211 /* Get digit grouping data for thousand separors by calling
1212 localeconv(). The data includes separator string and grouping info
1213 and is cached after the first call to the function.
1215 In locales that don't set a thousand separator (such as the "C"
1216 locale), this forces it to be ",". We are now only showing
1217 thousand separators in one place, so this shouldn't be a problem in
1221 get_grouping_data (const char **sep, const char **grouping)
1223 static const char *cached_sep;
1224 static const char *cached_grouping;
1225 static bool initialized;
1228 /* Get the grouping info from the locale. */
1229 struct lconv *lconv = localeconv ();
1230 cached_sep = lconv->thousands_sep;
1231 cached_grouping = lconv->grouping;
1232 #if ! USE_NLS_PROGRESS_BAR
1233 /* We can't count column widths, so ensure that the separator
1234 * is single-byte only (let check below determine what byte). */
1235 if (strlen(cached_sep) > 1)
1240 /* Many locales (such as "C" or "hr_HR") don't specify
1241 grouping, which we still want to use it for legibility.
1242 In those locales set the sep char to ',', unless that
1243 character is used for decimal point, in which case set it
1245 if (*lconv->decimal_point != ',')
1249 cached_grouping = "\x03";
1254 *grouping = cached_grouping;
1257 /* Return a printed representation of N with thousand separators.
1258 This should respect locale settings, with the exception of the "C"
1259 locale which mandates no separator, but we use one anyway.
1261 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1262 the separators because it's too non-portable, and it's hard to test
1263 for this feature at configure time. Besides, it wouldn't display
1264 separators in the "C" locale, still used by many Unix users. */
1267 with_thousand_seps (wgint n)
1269 static char outbuf[48];
1270 char *p = outbuf + sizeof outbuf;
1272 /* Info received from locale */
1273 const char *grouping, *sep;
1276 /* State information */
1277 int i = 0, groupsize;
1278 const char *atgroup;
1280 bool negative = n < 0;
1282 /* Initialize grouping data. */
1283 get_grouping_data (&sep, &grouping);
1284 seplen = strlen (sep);
1286 groupsize = *atgroup++;
1288 /* This would overflow on WGINT_MIN, but printing negative numbers
1289 is not an important goal of this fuinction. */
1293 /* Write the number into the buffer, backwards, inserting the
1294 separators as necessary. */
1298 *--p = n % 10 + '0';
1302 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1303 if (++i == groupsize)
1308 memcpy (p -= seplen, sep, seplen);
1311 groupsize = *atgroup++;
1320 /* N, a byte quantity, is converted to a human-readable abberviated
1321 form a la sizes printed by `ls -lh'. The result is written to a
1322 static buffer, a pointer to which is returned.
1324 Unlike `with_thousand_seps', this approximates to the nearest unit.
1325 Quoting GNU libit: "Most people visually process strings of 3-4
1326 digits effectively, but longer strings of digits are more prone to
1327 misinterpretation. Hence, converting to an abbreviated form
1328 usually improves readability."
1330 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1331 original computer-related meaning of "powers of 1024". We don't
1332 use the "*bibyte" names invented in 1998, and seldom used in
1333 practice. Wikipedia's entry on "binary prefix" discusses this in
1337 human_readable (HR_NUMTYPE n)
1339 /* These suffixes are compatible with those of GNU `ls -lh'. */
1340 static char powers[] =
1342 'K', /* kilobyte, 2^10 bytes */
1343 'M', /* megabyte, 2^20 bytes */
1344 'G', /* gigabyte, 2^30 bytes */
1345 'T', /* terabyte, 2^40 bytes */
1346 'P', /* petabyte, 2^50 bytes */
1347 'E', /* exabyte, 2^60 bytes */
1352 /* If the quantity is smaller than 1K, just print it. */
1355 snprintf (buf, sizeof (buf), "%d", (int) n);
1359 /* Loop over powers, dividing N with 1024 in each iteration. This
1360 works unchanged for all sizes of wgint, while still avoiding
1361 non-portable `long double' arithmetic. */
1362 for (i = 0; i < countof (powers); i++)
1364 /* At each iteration N is greater than the *subsequent* power.
1365 That way N/1024.0 produces a decimal number in the units of
1367 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1369 double val = n / 1024.0;
1370 /* Print values smaller than 10 with one decimal digits, and
1371 others without any decimals. */
1372 snprintf (buf, sizeof (buf), "%.*f%c",
1373 val < 10 ? 1 : 0, val, powers[i]);
1378 return NULL; /* unreached */
1381 /* Count the digits in the provided number. Used to allocate space
1382 when printing numbers. */
1385 numdigit (wgint number)
1389 ++cnt; /* accomodate '-' */
1390 while ((number /= 10) != 0)
1395 #define PR(mask) *p++ = n / (mask) + '0'
1397 /* DIGITS_<D> is used to print a D-digit number and should be called
1398 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1399 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1400 Recursively this continues until DIGITS_1 is invoked. */
1402 #define DIGITS_1(mask) PR (mask)
1403 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1404 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1405 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1406 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1407 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1408 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1409 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1410 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1411 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1413 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1415 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1416 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1417 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1418 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1419 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1420 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1421 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1422 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1423 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1425 /* Shorthand for casting to wgint. */
1428 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1429 `sprintf(buffer, "%lld", (long long) number)', only typically much
1430 faster and portable to machines without long long.
1432 The speedup may make a difference in programs that frequently
1433 convert numbers to strings. Some implementations of sprintf,
1434 particularly the one in some versions of GNU libc, have been known
1435 to be quite slow when converting integers to strings.
1437 Return the pointer to the location where the terminating zero was
1438 printed. (Equivalent to calling buffer+strlen(buffer) after the
1441 BUFFER should be large enough to accept as many bytes as you expect
1442 the number to take up. On machines with 64-bit wgints the maximum
1443 needed size is 24 bytes. That includes the digits needed for the
1444 largest 64-bit number, the `-' sign in case it's negative, and the
1445 terminating '\0'. */
1448 number_to_string (char *buffer, wgint number)
1453 int last_digit_char = 0;
1455 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1456 /* We are running in a very strange environment. Leave the correct
1457 printing to sprintf. */
1458 p += sprintf (buf, "%j", (intmax_t) (n));
1459 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1465 /* n = -n would overflow because -n would evaluate to a
1466 wgint value larger than WGINT_MAX. Need to make n
1467 smaller and handle the last digit separately. */
1468 int last_digit = n % 10;
1469 /* The sign of n%10 is implementation-defined. */
1471 last_digit_char = '0' - last_digit;
1473 last_digit_char = '0' + last_digit;
1474 /* After n is made smaller, -n will not overflow. */
1482 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1483 way printing any N is fully open-coded without a loop or jump.
1484 (Also see description of DIGITS_*.) */
1486 if (n < 10) DIGITS_1 (1);
1487 else if (n < 100) DIGITS_2 (10);
1488 else if (n < 1000) DIGITS_3 (100);
1489 else if (n < 10000) DIGITS_4 (1000);
1490 else if (n < 100000) DIGITS_5 (10000);
1491 else if (n < 1000000) DIGITS_6 (100000);
1492 else if (n < 10000000) DIGITS_7 (1000000);
1493 else if (n < 100000000) DIGITS_8 (10000000);
1494 else if (n < 1000000000) DIGITS_9 (100000000);
1495 #if SIZEOF_WGINT == 4
1496 /* wgint is 32 bits wide: no number has more than 10 digits. */
1497 else DIGITS_10 (1000000000);
1499 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1500 Constants are constructed by compile-time multiplication to avoid
1501 dealing with different notations for 64-bit constants
1502 (nL/nLL/nI64, depending on the compiler and architecture). */
1503 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1504 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1505 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1506 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1507 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1508 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1509 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1510 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1511 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1512 else DIGITS_19 (1000000000*(W)1000000000);
1515 if (last_digit_char)
1516 *p++ = last_digit_char;
1519 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1526 #undef SPRINTF_WGINT
1549 /* Print NUMBER to a statically allocated string and return a pointer
1550 to the printed representation.
1552 This function is intended to be used in conjunction with printf.
1553 It is hard to portably print wgint values:
1554 a) you cannot use printf("%ld", number) because wgint can be long
1555 long on 32-bit machines with LFS.
1556 b) you cannot use printf("%lld", number) because NUMBER could be
1557 long on 32-bit machines without LFS, or on 64-bit machines,
1558 which do not require LFS. Also, Windows doesn't support %lld.
1559 c) you cannot use printf("%j", (int_max_t) number) because not all
1560 versions of printf support "%j", the most notable being the one
1562 d) you cannot #define WGINT_FMT to the appropriate format and use
1563 printf(WGINT_FMT, number) because that would break translations
1564 for user-visible messages, such as printf("Downloaded: %d
1567 What you should use instead is printf("%s", number_to_static_string
1570 CAVEAT: since the function returns pointers to static data, you
1571 must be careful to copy its result before calling it again.
1572 However, to make it more useful with printf, the function maintains
1573 an internal ring of static buffers to return. That way things like
1574 printf("%s %s", number_to_static_string (num1),
1575 number_to_static_string (num2)) work as expected. Three buffers
1576 are currently used, which means that "%s %s %s" will work, but "%s
1577 %s %s %s" won't. If you need to print more than three wgints,
1578 bump the RING_SIZE (or rethink your message.) */
1581 number_to_static_string (wgint number)
1583 static char ring[RING_SIZE][24];
1585 char *buf = ring[ringpos];
1586 number_to_string (buf, number);
1587 ringpos = (ringpos + 1) % RING_SIZE;
1591 /* Determine the width of the terminal we're running on. If that's
1592 not possible, return 0. */
1595 determine_screen_width (void)
1597 /* If there's a way to get the terminal size using POSIX
1598 tcgetattr(), somebody please tell me. */
1603 if (opt.lfilename != NULL)
1606 fd = fileno (stderr);
1607 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1608 return 0; /* most likely ENOTTY */
1611 #elif defined(WINDOWS)
1612 CONSOLE_SCREEN_BUFFER_INFO csbi;
1613 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1615 return csbi.dwSize.X;
1616 #else /* neither TIOCGWINSZ nor WINDOWS */
1618 #endif /* neither TIOCGWINSZ nor WINDOWS */
1621 /* Whether the rnd system (either rand or [dl]rand48) has been
1623 static int rnd_seeded;
1625 /* Return a random number between 0 and MAX-1, inclusive.
1627 If the system does not support lrand48 and MAX is greater than the
1628 value of RAND_MAX+1 on the system, the returned value will be in
1629 the range [0, RAND_MAX]. This may be fixed in a future release.
1630 The random number generator is seeded automatically the first time
1633 This uses lrand48 where available, rand elsewhere. DO NOT use it
1634 for cryptography. It is only meant to be used in situations where
1635 quality of the random numbers returned doesn't really matter. */
1638 random_number (int max)
1643 srand48 ((long) time (NULL) ^ (long) getpid ());
1646 return lrand48 () % max;
1647 #else /* not HAVE_DRAND48 */
1653 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1658 /* Like rand() % max, but uses the high-order bits for better
1659 randomness on architectures where rand() is implemented using a
1660 simple congruential generator. */
1662 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1663 return (int) bounded;
1665 #endif /* not HAVE_DRAND48 */
1668 /* Return a random uniformly distributed floating point number in the
1669 [0, 1) range. Uses drand48 where available, and a really lame
1670 kludge elsewhere. */
1678 srand48 ((long) time (NULL) ^ (long) getpid ());
1682 #else /* not HAVE_DRAND48 */
1683 return ( random_number (10000) / 10000.0
1684 + random_number (10000) / (10000.0 * 10000.0)
1685 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1686 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1687 #endif /* not HAVE_DRAND48 */
1690 /* Implementation of run_with_timeout, a generic timeout-forcing
1691 routine for systems with Unix-like signal handling. */
1693 #ifdef USE_SIGNAL_TIMEOUT
1694 # ifdef HAVE_SIGSETJMP
1695 # define SETJMP(env) sigsetjmp (env, 1)
1697 static sigjmp_buf run_with_timeout_env;
1700 abort_run_with_timeout (int sig)
1702 assert (sig == SIGALRM);
1703 siglongjmp (run_with_timeout_env, -1);
1705 # else /* not HAVE_SIGSETJMP */
1706 # define SETJMP(env) setjmp (env)
1708 static jmp_buf run_with_timeout_env;
1711 abort_run_with_timeout (int sig)
1713 assert (sig == SIGALRM);
1714 /* We don't have siglongjmp to preserve the set of blocked signals;
1715 if we longjumped out of the handler at this point, SIGALRM would
1716 remain blocked. We must unblock it manually. */
1717 int mask = siggetmask ();
1718 mask &= ~sigmask (SIGALRM);
1721 /* Now it's safe to longjump. */
1722 longjmp (run_with_timeout_env, -1);
1724 # endif /* not HAVE_SIGSETJMP */
1726 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1727 setitimer where available, alarm otherwise.
1729 TIMEOUT should be non-zero. If the timeout value is so small that
1730 it would be rounded to zero, it is rounded to the least legal value
1731 instead (1us for setitimer, 1s for alarm). That ensures that
1732 SIGALRM will be delivered in all cases. */
1735 alarm_set (double timeout)
1738 /* Use the modern itimer interface. */
1739 struct itimerval itv;
1741 itv.it_value.tv_sec = (long) timeout;
1742 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1743 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1744 /* Ensure that we wait for at least the minimum interval.
1745 Specifying zero would mean "wait forever". */
1746 itv.it_value.tv_usec = 1;
1747 setitimer (ITIMER_REAL, &itv, NULL);
1748 #else /* not ITIMER_REAL */
1749 /* Use the old alarm() interface. */
1750 int secs = (int) timeout;
1752 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1753 because alarm(0) means "never deliver the alarm", i.e. "wait
1754 forever", which is not what someone who specifies a 0.5s
1755 timeout would expect. */
1758 #endif /* not ITIMER_REAL */
1761 /* Cancel the alarm set with alarm_set. */
1767 struct itimerval disable;
1769 setitimer (ITIMER_REAL, &disable, NULL);
1770 #else /* not ITIMER_REAL */
1772 #endif /* not ITIMER_REAL */
1775 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1776 seconds. Returns true if the function was interrupted with a
1777 timeout, false otherwise.
1779 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1780 using setitimer() or alarm(). The timeout is enforced by
1781 longjumping out of the SIGALRM handler. This has several
1782 advantages compared to the traditional approach of relying on
1783 signals causing system calls to exit with EINTR:
1785 * The callback function is *forcibly* interrupted after the
1786 timeout expires, (almost) regardless of what it was doing and
1787 whether it was in a syscall. For example, a calculation that
1788 takes a long time is interrupted as reliably as an IO
1791 * It works with both SYSV and BSD signals because it doesn't
1792 depend on the default setting of SA_RESTART.
1794 * It doesn't require special handler setup beyond a simple call
1795 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1798 The only downside is that, if FUN allocates internal resources that
1799 are normally freed prior to exit from the functions, they will be
1800 lost in case of timeout. */
1803 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1813 signal (SIGALRM, abort_run_with_timeout);
1814 if (SETJMP (run_with_timeout_env) != 0)
1816 /* Longjumped out of FUN with a timeout. */
1817 signal (SIGALRM, SIG_DFL);
1820 alarm_set (timeout);
1823 /* Preserve errno in case alarm() or signal() modifies it. */
1824 saved_errno = errno;
1826 signal (SIGALRM, SIG_DFL);
1827 errno = saved_errno;
1832 #else /* not USE_SIGNAL_TIMEOUT */
1835 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1836 define it under Windows, because Windows has its own version of
1837 run_with_timeout that uses threads. */
1840 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1845 #endif /* not WINDOWS */
1846 #endif /* not USE_SIGNAL_TIMEOUT */
1850 /* Sleep the specified amount of seconds. On machines without
1851 nanosleep(), this may sleep shorter if interrupted by signals. */
1854 xsleep (double seconds)
1856 #ifdef HAVE_NANOSLEEP
1857 /* nanosleep is the preferred interface because it offers high
1858 accuracy and, more importantly, because it allows us to reliably
1859 restart receiving a signal such as SIGWINCH. (There was an
1860 actual Debian bug report about --limit-rate malfunctioning while
1861 the terminal was being resized.) */
1862 struct timespec sleep, remaining;
1863 sleep.tv_sec = (long) seconds;
1864 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1865 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1866 /* If nanosleep has been interrupted by a signal, adjust the
1867 sleeping period and return to sleep. */
1869 #elif defined(HAVE_USLEEP)
1870 /* If usleep is available, use it in preference to select. */
1873 /* On some systems, usleep cannot handle values larger than
1874 1,000,000. If the period is larger than that, use sleep
1875 first, then add usleep for subsecond accuracy. */
1877 seconds -= (long) seconds;
1879 usleep (seconds * 1000000);
1880 #else /* fall back select */
1881 /* Note that, although Windows supports select, it can't be used to
1882 implement sleeping because Winsock's select doesn't implement
1883 timeout when it is passed NULL pointers for all fd sets. (But it
1884 does under Cygwin, which implements Unix-compatible select.) */
1885 struct timeval sleep;
1886 sleep.tv_sec = (long) seconds;
1887 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1888 select (0, NULL, NULL, NULL, &sleep);
1889 /* If select returns -1 and errno is EINTR, it means we were
1890 interrupted by a signal. But without knowing how long we've
1891 actually slept, we can't return to sleep. Using gettimeofday to
1892 track sleeps is slow and unreliable due to clock skew. */
1896 #endif /* not WINDOWS */
1898 /* Encode the octets in DATA of length LENGTH to base64 format,
1899 storing the result to DEST. The output will be zero-terminated,
1900 and must point to a writable buffer of at least
1901 1+BASE64_LENGTH(length) bytes. The function returns the length of
1902 the resulting base64 data, not counting the terminating zero.
1904 This implementation does not emit newlines after 76 characters of
1908 base64_encode (const void *data, int length, char *dest)
1910 /* Conversion table. */
1911 static const char tbl[64] = {
1912 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
1913 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
1914 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
1915 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
1917 /* Access bytes in DATA as unsigned char, otherwise the shifts below
1918 don't work for data with MSB set. */
1919 const unsigned char *s = data;
1920 /* Theoretical ANSI violation when length < 3. */
1921 const unsigned char *end = (const unsigned char *) data + length - 2;
1924 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1925 for (; s < end; s += 3)
1927 *p++ = tbl[s[0] >> 2];
1928 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1929 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1930 *p++ = tbl[s[2] & 0x3f];
1933 /* Pad the result if necessary... */
1937 *p++ = tbl[s[0] >> 2];
1938 *p++ = tbl[(s[0] & 3) << 4];
1943 *p++ = tbl[s[0] >> 2];
1944 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1945 *p++ = tbl[((s[1] & 0xf) << 2)];
1949 /* ...and zero-terminate it. */
1955 /* Store in C the next non-whitespace character from the string, or \0
1956 when end of string is reached. */
1957 #define NEXT_CHAR(c, p) do { \
1958 c = (unsigned char) *p++; \
1959 } while (c_isspace (c))
1961 #define IS_ASCII(c) (((c) & 0x80) == 0)
1963 /* Decode data from BASE64 (a null-terminated string) into memory
1964 pointed to by DEST. DEST is assumed to be large enough to
1965 accomodate the decoded data, which is guaranteed to be no more than
1968 Since DEST is assumed to contain binary data, it is not
1969 NUL-terminated. The function returns the length of the data
1970 written to TO. -1 is returned in case of error caused by malformed
1973 This function originates from Free Recode. */
1976 base64_decode (const char *base64, void *dest)
1978 /* Table of base64 values for first 128 characters. Note that this
1979 assumes ASCII (but so does Wget in other places). */
1980 static const signed char base64_char_to_value[128] =
1982 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1983 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1984 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1985 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1986 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1987 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1988 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1989 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1990 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1991 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1992 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1993 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1994 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1996 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
1997 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
1999 const char *p = base64;
2005 unsigned long value;
2007 /* Process first byte of a quadruplet. */
2011 if (c == '=' || !IS_BASE64 (c))
2012 return -1; /* illegal char while decoding base64 */
2013 value = BASE64_CHAR_TO_VALUE (c) << 18;
2015 /* Process second byte of a quadruplet. */
2018 return -1; /* premature EOF while decoding base64 */
2019 if (c == '=' || !IS_BASE64 (c))
2020 return -1; /* illegal char while decoding base64 */
2021 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2024 /* Process third byte of a quadruplet. */
2027 return -1; /* premature EOF while decoding base64 */
2029 return -1; /* illegal char while decoding base64 */
2035 return -1; /* premature EOF while decoding base64 */
2037 return -1; /* padding `=' expected but not found */
2041 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2042 *q++ = 0xff & value >> 8;
2044 /* Process fourth byte of a quadruplet. */
2047 return -1; /* premature EOF while decoding base64 */
2051 return -1; /* illegal char while decoding base64 */
2053 value |= BASE64_CHAR_TO_VALUE (c);
2054 *q++ = 0xff & value;
2057 #undef BASE64_CHAR_TO_VALUE
2059 return q - (char *) dest;
2065 /* Simple merge sort for use by stable_sort. Implementation courtesy
2066 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2069 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2070 int (*cmpfun) (const void *, const void *))
2072 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2076 size_t mid = (to + from) / 2;
2077 mergesort_internal (base, temp, size, from, mid, cmpfun);
2078 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2081 for (k = from; (i <= mid) && (j <= to); k++)
2082 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2083 memcpy (ELT (temp, k), ELT (base, i++), size);
2085 memcpy (ELT (temp, k), ELT (base, j++), size);
2087 memcpy (ELT (temp, k++), ELT (base, i++), size);
2089 memcpy (ELT (temp, k++), ELT (base, j++), size);
2090 for (k = from; k <= to; k++)
2091 memcpy (ELT (base, k), ELT (temp, k), size);
2096 /* Stable sort with interface exactly like standard library's qsort.
2097 Uses mergesort internally, allocating temporary storage with
2101 stable_sort (void *base, size_t nmemb, size_t size,
2102 int (*cmpfun) (const void *, const void *))
2106 void *temp = alloca (nmemb * size * sizeof (void *));
2107 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2111 /* Print a decimal number. If it is equal to or larger than ten, the
2112 number is rounded. Otherwise it is printed with one significant
2113 digit without trailing zeros and with no more than three fractional
2114 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2115 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2117 This is useful for displaying durations because it provides
2118 order-of-magnitude information without unnecessary clutter --
2119 long-running downloads are shown without the fractional part, and
2120 short ones still retain one significant digit. */
2123 print_decimal (double number)
2125 static char buf[32];
2126 double n = number >= 0 ? number : -number;
2129 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2130 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2131 snprintf (buf, sizeof buf, "%.0f", number);
2133 snprintf (buf, sizeof buf, "%.1f", number);
2134 else if (n >= 0.001)
2135 snprintf (buf, sizeof buf, "%.1g", number);
2136 else if (n >= 0.0005)
2137 /* round [0.0005, 0.001) to 0.001 */
2138 snprintf (buf, sizeof buf, "%.3f", number);
2140 /* print numbers close to 0 as 0, not 0.000 */
2157 { "/somedir", "/somedir", true },
2158 { "/somedir", "/somedir/d2", true },
2159 { "/somedir/d1", "/somedir", false },
2162 for (i = 0; i < countof(test_array); ++i)
2164 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2166 mu_assert ("test_subdir_p: wrong result",
2167 res == test_array[i].result);
2174 test_dir_matches_p()
2182 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2183 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2184 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2185 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2186 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2187 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2188 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2189 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2190 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2191 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2192 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2193 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2194 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2197 for (i = 0; i < countof(test_array); ++i)
2199 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2201 mu_assert ("test_dir_matches_p: wrong result",
2202 res == test_array[i].result);
2208 #endif /* TESTING */