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;
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 = c_tolower (*pattern);
641 for (p = strcopy; *string; string++, p++)
642 *p = c_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)
682 if (!opt.ignore_case)
683 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
686 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
689 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
692 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
693 first element that matches DIR, through wildcards or front comparison (as
696 dir_matches_p (char **dirlist, const char *dir)
699 int (*matcher) (const char *, const char *, int)
700 = opt.ignore_case ? fnmatch_nocase : fnmatch;
702 for (x = dirlist; *x; x++)
704 /* Remove leading '/' */
705 char *p = *x + (**x == '/');
706 if (has_wildcards_p (p))
708 if (matcher (p, dir, FNM_PATHNAME) == 0)
713 if (subdir_p (p, dir))
718 return *x ? true : false;
721 /* Returns whether DIRECTORY is acceptable for download, wrt the
722 include/exclude lists.
724 The leading `/' is ignored in paths; relative and absolute paths
725 may be freely intermixed. */
728 accdir (const char *directory)
730 /* Remove starting '/'. */
731 if (*directory == '/')
735 if (!dir_matches_p (opt.includes, directory))
740 if (dir_matches_p (opt.excludes, directory))
746 /* Return true if STRING ends with TAIL. For instance:
748 match_tail ("abc", "bc", false) -> 1
749 match_tail ("abc", "ab", false) -> 0
750 match_tail ("abc", "abc", false) -> 1
752 If FOLD_CASE is true, the comparison will be case-insensitive. */
755 match_tail (const char *string, const char *tail, bool fold_case)
759 /* We want this to be fast, so we code two loops, one with
760 case-folding, one without. */
764 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
765 if (string[i] != tail[j])
770 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
771 if (c_tolower (string[i]) != c_tolower (tail[j]))
775 /* If the tail was exhausted, the match was succesful. */
782 /* Checks whether string S matches each element of ACCEPTS. A list
783 element are matched either with fnmatch() or match_tail(),
784 according to whether the element contains wildcards or not.
786 If the BACKWARD is false, don't do backward comparison -- just compare
789 in_acclist (const char *const *accepts, const char *s, bool backward)
791 for (; *accepts; accepts++)
793 if (has_wildcards_p (*accepts))
795 int res = opt.ignore_case
796 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
797 /* fnmatch returns 0 if the pattern *does* match the string. */
805 if (match_tail (s, *accepts, opt.ignore_case))
810 int cmp = opt.ignore_case
811 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
820 /* Return the location of STR's suffix (file extension). Examples:
821 suffix ("foo.bar") -> "bar"
822 suffix ("foo.bar.baz") -> "baz"
823 suffix ("/foo/bar") -> NULL
824 suffix ("/foo.bar/baz") -> NULL */
826 suffix (const char *str)
830 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
834 return (char *)str + i;
839 /* Return true if S contains globbing wildcards (`*', `?', `[' or
843 has_wildcards_p (const char *s)
846 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
851 /* Return true if FNAME ends with a typical HTML suffix. The
852 following (case-insensitive) suffixes are presumed to be HTML
857 ?html (`?' matches one character)
859 #### CAVEAT. This is not necessarily a good indication that FNAME
860 refers to a file that contains HTML! */
862 has_html_suffix_p (const char *fname)
866 if ((suf = suffix (fname)) == NULL)
868 if (!strcasecmp (suf, "html"))
870 if (!strcasecmp (suf, "htm"))
872 if (suf[0] && !strcasecmp (suf + 1, "html"))
877 /* Read a line from FP and return the pointer to freshly allocated
878 storage. The storage space is obtained through malloc() and should
879 be freed with free() when it is no longer needed.
881 The length of the line is not limited, except by available memory.
882 The newline character at the end of line is retained. The line is
883 terminated with a zero character.
885 After end-of-file is encountered without anything being read, NULL
886 is returned. NULL is also returned on error. To distinguish
887 between these two cases, use the stdio function ferror(). */
890 read_whole_line (FILE *fp)
894 char *line = xmalloc (bufsize);
896 while (fgets (line + length, bufsize - length, fp))
898 length += strlen (line + length);
900 /* Possible for example when reading from a binary file where
901 a line begins with \0. */
904 if (line[length - 1] == '\n')
907 /* fgets() guarantees to read the whole line, or to use up the
908 space we've given it. We can double the buffer
911 line = xrealloc (line, bufsize);
913 if (length == 0 || ferror (fp))
918 if (length + 1 < bufsize)
919 /* Relieve the memory from our exponential greediness. We say
920 `length + 1' because the terminating \0 is not included in
921 LENGTH. We don't need to zero-terminate the string ourselves,
922 though, because fgets() does that. */
923 line = xrealloc (line, length + 1);
927 /* Read FILE into memory. A pointer to `struct file_memory' are
928 returned; use struct element `content' to access file contents, and
929 the element `length' to know the file length. `content' is *not*
930 zero-terminated, and you should *not* read or write beyond the [0,
931 length) range of characters.
933 After you are done with the file contents, call read_file_free to
936 Depending on the operating system and the type of file that is
937 being read, read_file() either mmap's the file into memory, or
938 reads the file into the core using read().
940 If file is named "-", fileno(stdin) is used for reading instead.
941 If you want to read from a real file named "-", use "./-" instead. */
944 read_file (const char *file)
947 struct file_memory *fm;
949 bool inhibit_close = false;
951 /* Some magic in the finest tradition of Perl and its kin: if FILE
952 is "-", just use stdin. */
956 inhibit_close = true;
957 /* Note that we don't inhibit mmap() in this case. If stdin is
958 redirected from a regular file, mmap() will still work. */
961 fd = open (file, O_RDONLY);
964 fm = xnew (struct file_memory);
969 if (fstat (fd, &buf) < 0)
971 fm->length = buf.st_size;
972 /* NOTE: As far as I know, the callers of this function never
973 modify the file text. Relying on this would enable us to
974 specify PROT_READ and MAP_SHARED for a marginal gain in
975 efficiency, but at some cost to generality. */
976 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
978 if (fm->content == (char *)MAP_FAILED)
988 /* The most common reason why mmap() fails is that FD does not point
989 to a plain file. However, it's also possible that mmap() doesn't
990 work for a particular type of file. Therefore, whenever mmap()
991 fails, we just fall back to the regular method. */
992 #endif /* HAVE_MMAP */
995 size = 512; /* number of bytes fm->contents can
996 hold at any given time. */
997 fm->content = xmalloc (size);
1001 if (fm->length > size / 2)
1003 /* #### I'm not sure whether the whole exponential-growth
1004 thing makes sense with kernel read. On Linux at least,
1005 read() refuses to read more than 4K from a file at a
1006 single chunk anyway. But other Unixes might optimize it
1007 better, and it doesn't *hurt* anything, so I'm leaving
1010 /* Normally, we grow SIZE exponentially to make the number
1011 of calls to read() and realloc() logarithmic in relation
1012 to file size. However, read() can read an amount of data
1013 smaller than requested, and it would be unreasonable to
1014 double SIZE every time *something* was read. Therefore,
1015 we double SIZE only when the length exceeds half of the
1016 entire allocated size. */
1018 fm->content = xrealloc (fm->content, size);
1020 nread = read (fd, fm->content + fm->length, size - fm->length);
1022 /* Successful read. */
1023 fm->length += nread;
1033 if (size > fm->length && fm->length != 0)
1034 /* Due to exponential growth of fm->content, the allocated region
1035 might be much larger than what is actually needed. */
1036 fm->content = xrealloc (fm->content, fm->length);
1043 xfree (fm->content);
1048 /* Release the resources held by FM. Specifically, this calls
1049 munmap() or xfree() on fm->content, depending whether mmap or
1050 malloc/read were used to read in the file. It also frees the
1051 memory needed to hold the FM structure itself. */
1054 read_file_free (struct file_memory *fm)
1059 munmap (fm->content, fm->length);
1064 xfree (fm->content);
1069 /* Free the pointers in a NULL-terminated vector of pointers, then
1070 free the pointer itself. */
1072 free_vec (char **vec)
1083 /* Append vector V2 to vector V1. The function frees V2 and
1084 reallocates V1 (thus you may not use the contents of neither
1085 pointer after the call). If V1 is NULL, V2 is returned. */
1087 merge_vecs (char **v1, char **v2)
1097 /* To avoid j == 0 */
1102 for (i = 0; v1[i]; i++)
1105 for (j = 0; v2[j]; j++)
1107 /* Reallocate v1. */
1108 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1109 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1114 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1115 is allocated as needed. Return the new value of the vector. */
1118 vec_append (char **vec, const char *str)
1120 int cnt; /* count of vector elements, including
1121 the one we're about to append */
1124 for (cnt = 0; vec[cnt]; cnt++)
1130 /* Reallocate the array to fit the new element and the NULL. */
1131 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1132 /* Append a copy of STR to the vector. */
1133 vec[cnt - 1] = xstrdup (str);
1138 /* Sometimes it's useful to create "sets" of strings, i.e. special
1139 hash tables where you want to store strings as keys and merely
1140 query for their existence. Here is a set of utility routines that
1141 makes that transparent. */
1144 string_set_add (struct hash_table *ht, const char *s)
1146 /* First check whether the set element already exists. If it does,
1147 do nothing so that we don't have to free() the old element and
1148 then strdup() a new one. */
1149 if (hash_table_contains (ht, s))
1152 /* We use "1" as value. It provides us a useful and clear arbitrary
1153 value, and it consumes no memory -- the pointers to the same
1154 string "1" will be shared by all the key-value pairs in all `set'
1156 hash_table_put (ht, xstrdup (s), "1");
1159 /* Synonym for hash_table_contains... */
1162 string_set_contains (struct hash_table *ht, const char *s)
1164 return hash_table_contains (ht, s);
1167 /* Convert the specified string set to array. ARRAY should be large
1168 enough to hold hash_table_count(ht) char pointers. */
1170 void string_set_to_array (struct hash_table *ht, char **array)
1172 hash_table_iterator iter;
1173 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1174 *array++ = iter.key;
1177 /* Free the string set. This frees both the storage allocated for
1178 keys and the actual hash table. (hash_table_destroy would only
1179 destroy the hash table.) */
1182 string_set_free (struct hash_table *ht)
1184 hash_table_iterator iter;
1185 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1187 hash_table_destroy (ht);
1190 /* Utility function: simply call xfree() on all keys and values of HT. */
1193 free_keys_and_values (struct hash_table *ht)
1195 hash_table_iterator iter;
1196 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1203 /* Get digit grouping data for thousand separors by calling
1204 localeconv(). The data includes separator string and grouping info
1205 and is cached after the first call to the function.
1207 In locales that don't set a thousand separator (such as the "C"
1208 locale), this forces it to be ",". We are now only showing
1209 thousand separators in one place, so this shouldn't be a problem in
1213 get_grouping_data (const char **sep, const char **grouping)
1215 static const char *cached_sep;
1216 static const char *cached_grouping;
1217 static bool initialized;
1220 /* Get the grouping info from the locale. */
1221 struct lconv *lconv = localeconv ();
1222 cached_sep = lconv->thousands_sep;
1223 cached_grouping = lconv->grouping;
1224 #if ! USE_NLS_PROGRESS_BAR
1225 /* We can't count column widths, so ensure that the separator
1226 * is single-byte only (let check below determine what byte). */
1227 if (strlen(cached_sep) > 1)
1232 /* Many locales (such as "C" or "hr_HR") don't specify
1233 grouping, which we still want to use it for legibility.
1234 In those locales set the sep char to ',', unless that
1235 character is used for decimal point, in which case set it
1237 if (*lconv->decimal_point != ',')
1241 cached_grouping = "\x03";
1246 *grouping = cached_grouping;
1249 /* Return a printed representation of N with thousand separators.
1250 This should respect locale settings, with the exception of the "C"
1251 locale which mandates no separator, but we use one anyway.
1253 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1254 the separators because it's too non-portable, and it's hard to test
1255 for this feature at configure time. Besides, it wouldn't display
1256 separators in the "C" locale, still used by many Unix users. */
1259 with_thousand_seps (wgint n)
1261 static char outbuf[48];
1262 char *p = outbuf + sizeof outbuf;
1264 /* Info received from locale */
1265 const char *grouping, *sep;
1268 /* State information */
1269 int i = 0, groupsize;
1270 const char *atgroup;
1272 bool negative = n < 0;
1274 /* Initialize grouping data. */
1275 get_grouping_data (&sep, &grouping);
1276 seplen = strlen (sep);
1278 groupsize = *atgroup++;
1280 /* This would overflow on WGINT_MIN, but printing negative numbers
1281 is not an important goal of this fuinction. */
1285 /* Write the number into the buffer, backwards, inserting the
1286 separators as necessary. */
1290 *--p = n % 10 + '0';
1294 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1295 if (++i == groupsize)
1300 memcpy (p -= seplen, sep, seplen);
1303 groupsize = *atgroup++;
1312 /* N, a byte quantity, is converted to a human-readable abberviated
1313 form a la sizes printed by `ls -lh'. The result is written to a
1314 static buffer, a pointer to which is returned.
1316 Unlike `with_thousand_seps', this approximates to the nearest unit.
1317 Quoting GNU libit: "Most people visually process strings of 3-4
1318 digits effectively, but longer strings of digits are more prone to
1319 misinterpretation. Hence, converting to an abbreviated form
1320 usually improves readability."
1322 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1323 original computer-related meaning of "powers of 1024". We don't
1324 use the "*bibyte" names invented in 1998, and seldom used in
1325 practice. Wikipedia's entry on "binary prefix" discusses this in
1329 human_readable (HR_NUMTYPE n)
1331 /* These suffixes are compatible with those of GNU `ls -lh'. */
1332 static char powers[] =
1334 'K', /* kilobyte, 2^10 bytes */
1335 'M', /* megabyte, 2^20 bytes */
1336 'G', /* gigabyte, 2^30 bytes */
1337 'T', /* terabyte, 2^40 bytes */
1338 'P', /* petabyte, 2^50 bytes */
1339 'E', /* exabyte, 2^60 bytes */
1344 /* If the quantity is smaller than 1K, just print it. */
1347 snprintf (buf, sizeof (buf), "%d", (int) n);
1351 /* Loop over powers, dividing N with 1024 in each iteration. This
1352 works unchanged for all sizes of wgint, while still avoiding
1353 non-portable `long double' arithmetic. */
1354 for (i = 0; i < countof (powers); i++)
1356 /* At each iteration N is greater than the *subsequent* power.
1357 That way N/1024.0 produces a decimal number in the units of
1359 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1361 double val = n / 1024.0;
1362 /* Print values smaller than 10 with one decimal digits, and
1363 others without any decimals. */
1364 snprintf (buf, sizeof (buf), "%.*f%c",
1365 val < 10 ? 1 : 0, val, powers[i]);
1370 return NULL; /* unreached */
1373 /* Count the digits in the provided number. Used to allocate space
1374 when printing numbers. */
1377 numdigit (wgint number)
1381 ++cnt; /* accomodate '-' */
1382 while ((number /= 10) != 0)
1387 #define PR(mask) *p++ = n / (mask) + '0'
1389 /* DIGITS_<D> is used to print a D-digit number and should be called
1390 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1391 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1392 Recursively this continues until DIGITS_1 is invoked. */
1394 #define DIGITS_1(mask) PR (mask)
1395 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1396 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1397 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1398 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1399 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1400 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1401 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1402 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1403 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1405 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1407 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1408 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1409 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1410 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1411 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1412 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1413 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1414 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1415 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1417 /* Shorthand for casting to wgint. */
1420 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1421 `sprintf(buffer, "%lld", (long long) number)', only typically much
1422 faster and portable to machines without long long.
1424 The speedup may make a difference in programs that frequently
1425 convert numbers to strings. Some implementations of sprintf,
1426 particularly the one in some versions of GNU libc, have been known
1427 to be quite slow when converting integers to strings.
1429 Return the pointer to the location where the terminating zero was
1430 printed. (Equivalent to calling buffer+strlen(buffer) after the
1433 BUFFER should be large enough to accept as many bytes as you expect
1434 the number to take up. On machines with 64-bit wgints the maximum
1435 needed size is 24 bytes. That includes the digits needed for the
1436 largest 64-bit number, the `-' sign in case it's negative, and the
1437 terminating '\0'. */
1440 number_to_string (char *buffer, wgint number)
1445 int last_digit_char = 0;
1447 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1448 /* We are running in a very strange environment. Leave the correct
1449 printing to sprintf. */
1450 p += sprintf (buf, "%j", (intmax_t) (n));
1451 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1457 /* n = -n would overflow because -n would evaluate to a
1458 wgint value larger than WGINT_MAX. Need to make n
1459 smaller and handle the last digit separately. */
1460 int last_digit = n % 10;
1461 /* The sign of n%10 is implementation-defined. */
1463 last_digit_char = '0' - last_digit;
1465 last_digit_char = '0' + last_digit;
1466 /* After n is made smaller, -n will not overflow. */
1474 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1475 way printing any N is fully open-coded without a loop or jump.
1476 (Also see description of DIGITS_*.) */
1478 if (n < 10) DIGITS_1 (1);
1479 else if (n < 100) DIGITS_2 (10);
1480 else if (n < 1000) DIGITS_3 (100);
1481 else if (n < 10000) DIGITS_4 (1000);
1482 else if (n < 100000) DIGITS_5 (10000);
1483 else if (n < 1000000) DIGITS_6 (100000);
1484 else if (n < 10000000) DIGITS_7 (1000000);
1485 else if (n < 100000000) DIGITS_8 (10000000);
1486 else if (n < 1000000000) DIGITS_9 (100000000);
1487 #if SIZEOF_WGINT == 4
1488 /* wgint is 32 bits wide: no number has more than 10 digits. */
1489 else DIGITS_10 (1000000000);
1491 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1492 Constants are constructed by compile-time multiplication to avoid
1493 dealing with different notations for 64-bit constants
1494 (nL/nLL/nI64, depending on the compiler and architecture). */
1495 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1496 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1497 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1498 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1499 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1500 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1501 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1502 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1503 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1504 else DIGITS_19 (1000000000*(W)1000000000);
1507 if (last_digit_char)
1508 *p++ = last_digit_char;
1511 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1518 #undef SPRINTF_WGINT
1541 /* Print NUMBER to a statically allocated string and return a pointer
1542 to the printed representation.
1544 This function is intended to be used in conjunction with printf.
1545 It is hard to portably print wgint values:
1546 a) you cannot use printf("%ld", number) because wgint can be long
1547 long on 32-bit machines with LFS.
1548 b) you cannot use printf("%lld", number) because NUMBER could be
1549 long on 32-bit machines without LFS, or on 64-bit machines,
1550 which do not require LFS. Also, Windows doesn't support %lld.
1551 c) you cannot use printf("%j", (int_max_t) number) because not all
1552 versions of printf support "%j", the most notable being the one
1554 d) you cannot #define WGINT_FMT to the appropriate format and use
1555 printf(WGINT_FMT, number) because that would break translations
1556 for user-visible messages, such as printf("Downloaded: %d
1559 What you should use instead is printf("%s", number_to_static_string
1562 CAVEAT: since the function returns pointers to static data, you
1563 must be careful to copy its result before calling it again.
1564 However, to make it more useful with printf, the function maintains
1565 an internal ring of static buffers to return. That way things like
1566 printf("%s %s", number_to_static_string (num1),
1567 number_to_static_string (num2)) work as expected. Three buffers
1568 are currently used, which means that "%s %s %s" will work, but "%s
1569 %s %s %s" won't. If you need to print more than three wgints,
1570 bump the RING_SIZE (or rethink your message.) */
1573 number_to_static_string (wgint number)
1575 static char ring[RING_SIZE][24];
1577 char *buf = ring[ringpos];
1578 number_to_string (buf, number);
1579 ringpos = (ringpos + 1) % RING_SIZE;
1583 /* Determine the width of the terminal we're running on. If that's
1584 not possible, return 0. */
1587 determine_screen_width (void)
1589 /* If there's a way to get the terminal size using POSIX
1590 tcgetattr(), somebody please tell me. */
1595 if (opt.lfilename != NULL)
1598 fd = fileno (stderr);
1599 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1600 return 0; /* most likely ENOTTY */
1603 #elif defined(WINDOWS)
1604 CONSOLE_SCREEN_BUFFER_INFO csbi;
1605 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1607 return csbi.dwSize.X;
1608 #else /* neither TIOCGWINSZ nor WINDOWS */
1610 #endif /* neither TIOCGWINSZ nor WINDOWS */
1613 /* Whether the rnd system (either rand or [dl]rand48) has been
1615 static int rnd_seeded;
1617 /* Return a random number between 0 and MAX-1, inclusive.
1619 If the system does not support lrand48 and MAX is greater than the
1620 value of RAND_MAX+1 on the system, the returned value will be in
1621 the range [0, RAND_MAX]. This may be fixed in a future release.
1622 The random number generator is seeded automatically the first time
1625 This uses lrand48 where available, rand elsewhere. DO NOT use it
1626 for cryptography. It is only meant to be used in situations where
1627 quality of the random numbers returned doesn't really matter. */
1630 random_number (int max)
1635 srand48 ((long) time (NULL) ^ (long) getpid ());
1638 return lrand48 () % max;
1639 #else /* not HAVE_DRAND48 */
1645 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1650 /* Like rand() % max, but uses the high-order bits for better
1651 randomness on architectures where rand() is implemented using a
1652 simple congruential generator. */
1654 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1655 return (int) bounded;
1657 #endif /* not HAVE_DRAND48 */
1660 /* Return a random uniformly distributed floating point number in the
1661 [0, 1) range. Uses drand48 where available, and a really lame
1662 kludge elsewhere. */
1670 srand48 ((long) time (NULL) ^ (long) getpid ());
1674 #else /* not HAVE_DRAND48 */
1675 return ( random_number (10000) / 10000.0
1676 + random_number (10000) / (10000.0 * 10000.0)
1677 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1678 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1679 #endif /* not HAVE_DRAND48 */
1682 /* Implementation of run_with_timeout, a generic timeout-forcing
1683 routine for systems with Unix-like signal handling. */
1685 #ifdef USE_SIGNAL_TIMEOUT
1686 # ifdef HAVE_SIGSETJMP
1687 # define SETJMP(env) sigsetjmp (env, 1)
1689 static sigjmp_buf run_with_timeout_env;
1692 abort_run_with_timeout (int sig)
1694 assert (sig == SIGALRM);
1695 siglongjmp (run_with_timeout_env, -1);
1697 # else /* not HAVE_SIGSETJMP */
1698 # define SETJMP(env) setjmp (env)
1700 static jmp_buf run_with_timeout_env;
1703 abort_run_with_timeout (int sig)
1705 assert (sig == SIGALRM);
1706 /* We don't have siglongjmp to preserve the set of blocked signals;
1707 if we longjumped out of the handler at this point, SIGALRM would
1708 remain blocked. We must unblock it manually. */
1709 int mask = siggetmask ();
1710 mask &= ~sigmask (SIGALRM);
1713 /* Now it's safe to longjump. */
1714 longjmp (run_with_timeout_env, -1);
1716 # endif /* not HAVE_SIGSETJMP */
1718 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1719 setitimer where available, alarm otherwise.
1721 TIMEOUT should be non-zero. If the timeout value is so small that
1722 it would be rounded to zero, it is rounded to the least legal value
1723 instead (1us for setitimer, 1s for alarm). That ensures that
1724 SIGALRM will be delivered in all cases. */
1727 alarm_set (double timeout)
1730 /* Use the modern itimer interface. */
1731 struct itimerval itv;
1733 itv.it_value.tv_sec = (long) timeout;
1734 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1735 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1736 /* Ensure that we wait for at least the minimum interval.
1737 Specifying zero would mean "wait forever". */
1738 itv.it_value.tv_usec = 1;
1739 setitimer (ITIMER_REAL, &itv, NULL);
1740 #else /* not ITIMER_REAL */
1741 /* Use the old alarm() interface. */
1742 int secs = (int) timeout;
1744 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1745 because alarm(0) means "never deliver the alarm", i.e. "wait
1746 forever", which is not what someone who specifies a 0.5s
1747 timeout would expect. */
1750 #endif /* not ITIMER_REAL */
1753 /* Cancel the alarm set with alarm_set. */
1759 struct itimerval disable;
1761 setitimer (ITIMER_REAL, &disable, NULL);
1762 #else /* not ITIMER_REAL */
1764 #endif /* not ITIMER_REAL */
1767 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1768 seconds. Returns true if the function was interrupted with a
1769 timeout, false otherwise.
1771 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1772 using setitimer() or alarm(). The timeout is enforced by
1773 longjumping out of the SIGALRM handler. This has several
1774 advantages compared to the traditional approach of relying on
1775 signals causing system calls to exit with EINTR:
1777 * The callback function is *forcibly* interrupted after the
1778 timeout expires, (almost) regardless of what it was doing and
1779 whether it was in a syscall. For example, a calculation that
1780 takes a long time is interrupted as reliably as an IO
1783 * It works with both SYSV and BSD signals because it doesn't
1784 depend on the default setting of SA_RESTART.
1786 * It doesn't require special handler setup beyond a simple call
1787 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1790 The only downside is that, if FUN allocates internal resources that
1791 are normally freed prior to exit from the functions, they will be
1792 lost in case of timeout. */
1795 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1805 signal (SIGALRM, abort_run_with_timeout);
1806 if (SETJMP (run_with_timeout_env) != 0)
1808 /* Longjumped out of FUN with a timeout. */
1809 signal (SIGALRM, SIG_DFL);
1812 alarm_set (timeout);
1815 /* Preserve errno in case alarm() or signal() modifies it. */
1816 saved_errno = errno;
1818 signal (SIGALRM, SIG_DFL);
1819 errno = saved_errno;
1824 #else /* not USE_SIGNAL_TIMEOUT */
1827 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1828 define it under Windows, because Windows has its own version of
1829 run_with_timeout that uses threads. */
1832 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1837 #endif /* not WINDOWS */
1838 #endif /* not USE_SIGNAL_TIMEOUT */
1842 /* Sleep the specified amount of seconds. On machines without
1843 nanosleep(), this may sleep shorter if interrupted by signals. */
1846 xsleep (double seconds)
1848 #ifdef HAVE_NANOSLEEP
1849 /* nanosleep is the preferred interface because it offers high
1850 accuracy and, more importantly, because it allows us to reliably
1851 restart receiving a signal such as SIGWINCH. (There was an
1852 actual Debian bug report about --limit-rate malfunctioning while
1853 the terminal was being resized.) */
1854 struct timespec sleep, remaining;
1855 sleep.tv_sec = (long) seconds;
1856 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1857 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1858 /* If nanosleep has been interrupted by a signal, adjust the
1859 sleeping period and return to sleep. */
1861 #elif defined(HAVE_USLEEP)
1862 /* If usleep is available, use it in preference to select. */
1865 /* On some systems, usleep cannot handle values larger than
1866 1,000,000. If the period is larger than that, use sleep
1867 first, then add usleep for subsecond accuracy. */
1869 seconds -= (long) seconds;
1871 usleep (seconds * 1000000);
1872 #else /* fall back select */
1873 /* Note that, although Windows supports select, it can't be used to
1874 implement sleeping because Winsock's select doesn't implement
1875 timeout when it is passed NULL pointers for all fd sets. (But it
1876 does under Cygwin, which implements Unix-compatible select.) */
1877 struct timeval sleep;
1878 sleep.tv_sec = (long) seconds;
1879 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1880 select (0, NULL, NULL, NULL, &sleep);
1881 /* If select returns -1 and errno is EINTR, it means we were
1882 interrupted by a signal. But without knowing how long we've
1883 actually slept, we can't return to sleep. Using gettimeofday to
1884 track sleeps is slow and unreliable due to clock skew. */
1888 #endif /* not WINDOWS */
1890 /* Encode the octets in DATA of length LENGTH to base64 format,
1891 storing the result to DEST. The output will be zero-terminated,
1892 and must point to a writable buffer of at least
1893 1+BASE64_LENGTH(length) bytes. The function returns the length of
1894 the resulting base64 data, not counting the terminating zero.
1896 This implementation does not emit newlines after 76 characters of
1900 base64_encode (const void *data, int length, char *dest)
1902 /* Conversion table. */
1903 static const char tbl[64] = {
1904 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
1905 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
1906 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
1907 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
1909 /* Access bytes in DATA as unsigned char, otherwise the shifts below
1910 don't work for data with MSB set. */
1911 const unsigned char *s = data;
1912 /* Theoretical ANSI violation when length < 3. */
1913 const unsigned char *end = (const unsigned char *) data + length - 2;
1916 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1917 for (; s < end; s += 3)
1919 *p++ = tbl[s[0] >> 2];
1920 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1921 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1922 *p++ = tbl[s[2] & 0x3f];
1925 /* Pad the result if necessary... */
1929 *p++ = tbl[s[0] >> 2];
1930 *p++ = tbl[(s[0] & 3) << 4];
1935 *p++ = tbl[s[0] >> 2];
1936 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1937 *p++ = tbl[((s[1] & 0xf) << 2)];
1941 /* ...and zero-terminate it. */
1947 /* Store in C the next non-whitespace character from the string, or \0
1948 when end of string is reached. */
1949 #define NEXT_CHAR(c, p) do { \
1950 c = (unsigned char) *p++; \
1951 } while (c_isspace (c))
1953 #define IS_ASCII(c) (((c) & 0x80) == 0)
1955 /* Decode data from BASE64 (a null-terminated string) into memory
1956 pointed to by DEST. DEST is assumed to be large enough to
1957 accomodate the decoded data, which is guaranteed to be no more than
1960 Since DEST is assumed to contain binary data, it is not
1961 NUL-terminated. The function returns the length of the data
1962 written to TO. -1 is returned in case of error caused by malformed
1965 This function originates from Free Recode. */
1968 base64_decode (const char *base64, void *dest)
1970 /* Table of base64 values for first 128 characters. Note that this
1971 assumes ASCII (but so does Wget in other places). */
1972 static const signed char base64_char_to_value[128] =
1974 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1975 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1976 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1977 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1978 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1979 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1980 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1981 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1982 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1983 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1984 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1985 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1986 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1988 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
1989 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
1991 const char *p = base64;
1997 unsigned long value;
1999 /* Process first byte of a quadruplet. */
2003 if (c == '=' || !IS_BASE64 (c))
2004 return -1; /* illegal char while decoding base64 */
2005 value = BASE64_CHAR_TO_VALUE (c) << 18;
2007 /* Process second byte of a quadruplet. */
2010 return -1; /* premature EOF while decoding base64 */
2011 if (c == '=' || !IS_BASE64 (c))
2012 return -1; /* illegal char while decoding base64 */
2013 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2016 /* Process third byte of a quadruplet. */
2019 return -1; /* premature EOF while decoding base64 */
2021 return -1; /* illegal char while decoding base64 */
2027 return -1; /* premature EOF while decoding base64 */
2029 return -1; /* padding `=' expected but not found */
2033 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2034 *q++ = 0xff & value >> 8;
2036 /* Process fourth byte of a quadruplet. */
2039 return -1; /* premature EOF while decoding base64 */
2043 return -1; /* illegal char while decoding base64 */
2045 value |= BASE64_CHAR_TO_VALUE (c);
2046 *q++ = 0xff & value;
2049 #undef BASE64_CHAR_TO_VALUE
2051 return q - (char *) dest;
2057 /* Simple merge sort for use by stable_sort. Implementation courtesy
2058 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2061 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2062 int (*cmpfun) (const void *, const void *))
2064 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2068 size_t mid = (to + from) / 2;
2069 mergesort_internal (base, temp, size, from, mid, cmpfun);
2070 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2073 for (k = from; (i <= mid) && (j <= to); k++)
2074 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2075 memcpy (ELT (temp, k), ELT (base, i++), size);
2077 memcpy (ELT (temp, k), ELT (base, j++), size);
2079 memcpy (ELT (temp, k++), ELT (base, i++), size);
2081 memcpy (ELT (temp, k++), ELT (base, j++), size);
2082 for (k = from; k <= to; k++)
2083 memcpy (ELT (base, k), ELT (temp, k), size);
2088 /* Stable sort with interface exactly like standard library's qsort.
2089 Uses mergesort internally, allocating temporary storage with
2093 stable_sort (void *base, size_t nmemb, size_t size,
2094 int (*cmpfun) (const void *, const void *))
2098 void *temp = alloca (nmemb * size * sizeof (void *));
2099 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2103 /* Print a decimal number. If it is equal to or larger than ten, the
2104 number is rounded. Otherwise it is printed with one significant
2105 digit without trailing zeros and with no more than three fractional
2106 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2107 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2109 This is useful for displaying durations because it provides
2110 order-of-magnitude information without unnecessary clutter --
2111 long-running downloads are shown without the fractional part, and
2112 short ones still retain one significant digit. */
2115 print_decimal (double number)
2117 static char buf[32];
2118 double n = number >= 0 ? number : -number;
2121 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2122 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2123 snprintf (buf, sizeof buf, "%.0f", number);
2125 snprintf (buf, sizeof buf, "%.1f", number);
2126 else if (n >= 0.001)
2127 snprintf (buf, sizeof buf, "%.1g", number);
2128 else if (n >= 0.0005)
2129 /* round [0.0005, 0.001) to 0.001 */
2130 snprintf (buf, sizeof buf, "%.3f", number);
2132 /* print numbers close to 0 as 0, not 0.000 */
2149 { "/somedir", "/somedir", true },
2150 { "/somedir", "/somedir/d2", true },
2151 { "/somedir/d1", "/somedir", false },
2154 for (i = 0; i < countof(test_array); ++i)
2156 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2158 mu_assert ("test_subdir_p: wrong result",
2159 res == test_array[i].result);
2166 test_dir_matches_p()
2174 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2175 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2176 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2177 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2178 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2179 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2180 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2181 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2182 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2183 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2184 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2185 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2186 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2189 for (i = 0; i < countof(test_array); ++i)
2191 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2193 mu_assert ("test_dir_matches_p: wrong result",
2194 res == test_array[i].result);
2200 #endif /* TESTING */