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. */
31 #define USE_GNULIB_ALLOC
39 #ifdef HAVE_SYS_TIME_H
40 # include <sys/time.h>
46 # include <sys/mman.h>
49 # include <process.h> /* getpid() */
54 #ifdef HAVE_SYS_UTIME_H
55 # include <sys/utime.h>
63 /* For TIOCGWINSZ and friends: */
64 #ifdef HAVE_SYS_IOCTL_H
65 # include <sys/ioctl.h>
71 /* Needed for Unix version of run_with_timeout. */
75 #ifndef HAVE_SIGSETJMP
76 /* If sigsetjmp is a macro, configure won't pick it up. */
78 # define HAVE_SIGSETJMP
82 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
83 # define USE_SIGNAL_TIMEOUT
93 /* Utility function: like xstrdup(), but also lowercases S. */
96 xstrdup_lower (const char *s)
98 char *copy = xstrdup (s);
105 /* Copy the string formed by two pointers (one on the beginning, other
106 on the char after the last char) to a new, malloc-ed location.
109 strdupdelim (const char *beg, const char *end)
111 char *res = xmalloc (end - beg + 1);
112 memcpy (res, beg, end - beg);
113 res[end - beg] = '\0';
117 /* Parse a string containing comma-separated elements, and return a
118 vector of char pointers with the elements. Spaces following the
119 commas are ignored. */
121 sepstring (const char *s)
135 res = xrealloc (res, (i + 2) * sizeof (char *));
136 res[i] = strdupdelim (p, s);
139 /* Skip the blanks following the ','. */
140 while (c_isspace (*s))
147 res = xrealloc (res, (i + 2) * sizeof (char *));
148 res[i] = strdupdelim (p, s);
153 /* Like sprintf, but prints into a string of sufficient size freshly
154 allocated with malloc, which is returned. If unable to print due
155 to invalid format, returns NULL. Inability to allocate needed
156 memory results in abort, as with xmalloc. This is in spirit
157 similar to the GNU/BSD extension asprintf, but somewhat easier to
160 Internally the function either calls vasprintf or loops around
161 vsnprintf until the correct size is found. Since Wget also ships a
162 fallback implementation of vsnprintf, this should be portable. */
164 /* Constant is using for limits memory allocation for text buffer.
165 Applicable in situation when: vasprintf is not available in the system
166 and vsnprintf return -1 when long line is truncated (in old versions of
167 glibc and in other system where C99 doesn`t support) */
169 #define FMT_MAX_LENGTH 1048576
172 aprintf (const char *fmt, ...)
174 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
179 va_start (args, fmt);
180 ret = vasprintf (&str, fmt, args);
182 if (ret < 0 && errno == ENOMEM)
183 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency
184 with xmalloc/xrealloc */
188 #else /* not HAVE_VASPRINTF */
190 /* vasprintf is unavailable. snprintf into a small buffer and
191 resize it as necessary. */
193 char *str = xmalloc (size);
195 /* #### This code will infloop and eventually abort in xrealloc if
196 passed a FMT that causes snprintf to consistently return -1. */
203 va_start (args, fmt);
204 n = vsnprintf (str, size, fmt, args);
207 /* If the printing worked, return the string. */
208 if (n > -1 && n < size)
211 /* Else try again with a larger buffer. */
212 if (n > -1) /* C99 */
213 size = n + 1; /* precisely what is needed */
214 else if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
215 { /* maybe we have some wrong
217 logprintf (LOG_ALWAYS,
218 _("%s: aprintf: text buffer is too big (%ld bytes), "
220 exec_name, size); /* printout a log message */
221 abort (); /* and abort... */
225 /* else, we continue to grow our
226 * buffer: Twice the old size. */
229 str = xrealloc (str, size);
231 #endif /* not HAVE_VASPRINTF */
234 /* Concatenate the NULL-terminated list of string arguments into
235 freshly allocated space. */
238 concat_strings (const char *str0, ...)
241 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
244 const char *next_str;
245 int total_length = 0;
248 /* Calculate the length of and allocate the resulting string. */
251 va_start (args, str0);
252 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
254 int len = strlen (next_str);
255 if (argcount < countof (saved_lengths))
256 saved_lengths[argcount++] = len;
260 p = ret = xmalloc (total_length + 1);
262 /* Copy the strings into the allocated space. */
265 va_start (args, str0);
266 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
269 if (argcount < countof (saved_lengths))
270 len = saved_lengths[argcount++];
272 len = strlen (next_str);
273 memcpy (p, next_str, len);
282 /* Format the provided time according to the specified format. The
283 format is a string with format elements supported by strftime. */
286 fmttime (time_t t, const char *fmt)
288 static char output[32];
289 struct tm *tm = localtime(&t);
292 if (!strftime(output, sizeof(output), fmt, tm))
297 /* Return pointer to a static char[] buffer in which zero-terminated
298 string-representation of TM (in form hh:mm:ss) is printed.
300 If TM is NULL, the current time will be used. */
305 return fmttime(t, "%H:%M:%S");
308 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
311 datetime_str (time_t t)
313 return fmttime(t, "%Y-%m-%d %H:%M:%S");
316 /* The Windows versions of the following two functions are defined in
317 mswindows.c. On MSDOS this function should never be called. */
319 #if !defined(WINDOWS) && !defined(MSDOS)
321 fork_to_background (void)
324 /* Whether we arrange our own version of opt.lfilename here. */
325 bool logfile_changed = false;
327 if (!opt.lfilename && (!opt.quiet || opt.server_response))
329 /* We must create the file immediately to avoid either a race
330 condition (which arises from using unique_name and failing to
331 use fopen_excl) or lying to the user about the log file name
332 (which arises from using unique_name, printing the name, and
333 using fopen_excl later on.) */
334 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
337 logfile_changed = true;
350 /* parent, no error */
351 printf (_("Continuing in background, pid %d.\n"), (int) pid);
353 printf (_("Output will be written to %s.\n"), quote (opt.lfilename));
354 exit (0); /* #### should we use _exit()? */
357 /* child: give up the privileges and keep running. */
359 freopen ("/dev/null", "r", stdin);
360 freopen ("/dev/null", "w", stdout);
361 freopen ("/dev/null", "w", stderr);
363 #endif /* !WINDOWS && !MSDOS */
365 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
366 specified with TM. The atime ("access time") is set to the current
370 touch (const char *file, time_t tm)
372 #ifdef HAVE_STRUCT_UTIMBUF
373 struct utimbuf times;
381 times.actime = time (NULL);
382 if (utime (file, ×) == -1)
383 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
386 /* Checks if FILE is a symbolic link, and removes it if it is. Does
387 nothing under MS-Windows. */
389 remove_link (const char *file)
394 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
396 DEBUGP (("Unlinking %s (symlink).\n", file));
399 logprintf (LOG_VERBOSE, _("Failed to unlink symlink %s: %s\n"),
400 quote (file), strerror (errno));
405 /* Does FILENAME exist? This is quite a lousy implementation, since
406 it supplies no error codes -- only a yes-or-no answer. Thus it
407 will return that a file does not exist if, e.g., the directory is
408 unreadable. I don't mind it too much currently, though. The
409 proper way should, of course, be to have a third, error state,
410 other than true/false, but that would introduce uncalled-for
411 additional complexity to the callers. */
413 file_exists_p (const char *filename)
416 return access (filename, F_OK) >= 0;
419 return stat (filename, &buf) >= 0;
423 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
424 Returns 0 on error. */
426 file_non_directory_p (const char *path)
429 /* Use lstat() rather than stat() so that symbolic links pointing to
430 directories can be identified correctly. */
431 if (lstat (path, &buf) != 0)
433 return S_ISDIR (buf.st_mode) ? false : true;
436 /* Return the size of file named by FILENAME, or -1 if it cannot be
437 opened or seeked into. */
439 file_size (const char *filename)
441 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
443 /* We use fseek rather than stat to determine the file size because
444 that way we can also verify that the file is readable without
445 explicitly checking for permissions. Inspired by the POST patch
447 FILE *fp = fopen (filename, "rb");
450 fseeko (fp, 0, SEEK_END);
456 if (stat (filename, &st) < 0)
462 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
463 doesn't exist is found. Return a freshly allocated copy of the
467 unique_name_1 (const char *prefix)
470 int plen = strlen (prefix);
471 char *template = (char *)alloca (plen + 1 + 24);
472 char *template_tail = template + plen;
474 memcpy (template, prefix, plen);
475 *template_tail++ = '.';
478 number_to_string (template_tail, count++);
479 while (file_exists_p (template));
481 return xstrdup (template);
484 /* Return a unique file name, based on FILE.
486 More precisely, if FILE doesn't exist, it is returned unmodified.
487 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
488 file name that doesn't exist is returned.
490 The resulting file is not created, only verified that it didn't
491 exist at the point in time when the function was called.
492 Therefore, where security matters, don't rely that the file created
493 by this function exists until you open it with O_EXCL or
496 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
497 string. Otherwise, it may return FILE if the file doesn't exist
498 (and therefore doesn't need changing). */
501 unique_name (const char *file, bool allow_passthrough)
503 /* If the FILE itself doesn't exist, return it without
505 if (!file_exists_p (file))
506 return allow_passthrough ? (char *)file : xstrdup (file);
508 /* Otherwise, find a numeric suffix that results in unused file name
510 return unique_name_1 (file);
513 /* Create a file based on NAME, except without overwriting an existing
514 file with that name. Providing O_EXCL is correctly implemented,
515 this function does not have the race condition associated with
516 opening the file returned by unique_name. */
519 unique_create (const char *name, bool binary, char **opened_name)
521 /* unique file name, based on NAME */
522 char *uname = unique_name (name, false);
524 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
527 uname = unique_name (name, false);
529 if (opened_name && fp != NULL)
532 *opened_name = uname;
544 /* Open the file for writing, with the addition that the file is
545 opened "exclusively". This means that, if the file already exists,
546 this function will *fail* and errno will be set to EEXIST. If
547 BINARY is set, the file will be opened in binary mode, equivalent
550 If opening the file fails for any reason, including the file having
551 previously existed, this function returns NULL and sets errno
555 fopen_excl (const char *fname, bool binary)
559 int flags = O_WRONLY | O_CREAT | O_EXCL;
564 fd = open (fname, flags, 0666);
567 return fdopen (fd, binary ? "wb" : "w");
568 #else /* not O_EXCL */
569 /* Manually check whether the file exists. This is prone to race
570 conditions, but systems without O_EXCL haven't deserved
572 if (file_exists_p (fname))
577 return fopen (fname, binary ? "wb" : "w");
578 #endif /* not O_EXCL */
581 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
582 are missing, create them first. In case any mkdir() call fails,
583 return its error status. Returns 0 on successful completion.
585 The behaviour of this function should be identical to the behaviour
586 of `mkdir -p' on systems where mkdir supports the `-p' option. */
588 make_directory (const char *directory)
590 int i, ret, quit = 0;
593 /* Make a copy of dir, to be able to write to it. Otherwise, the
594 function is unsafe if called with a read-only char *argument. */
595 STRDUP_ALLOCA (dir, directory);
597 /* If the first character of dir is '/', skip it (and thus enable
598 creation of absolute-pathname directories. */
599 for (i = (*dir == '/'); 1; ++i)
601 for (; dir[i] && dir[i] != '/'; i++)
606 /* Check whether the directory already exists. Allow creation of
607 of intermediate directories to fail, as the initial path components
608 are not necessarily directories! */
609 if (!file_exists_p (dir))
610 ret = mkdir (dir, 0777);
621 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
622 should be a file name.
624 file_merge("/foo/bar", "baz") => "/foo/baz"
625 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
626 file_merge("foo", "bar") => "bar"
628 In other words, it's a simpler and gentler version of uri_merge. */
631 file_merge (const char *base, const char *file)
634 const char *cut = (const char *)strrchr (base, '/');
637 return xstrdup (file);
639 result = xmalloc (cut - base + 1 + strlen (file) + 1);
640 memcpy (result, base, cut - base);
641 result[cut - base] = '/';
642 strcpy (result + (cut - base) + 1, file);
647 /* Like fnmatch, but performs a case-insensitive match. */
650 fnmatch_nocase (const char *pattern, const char *string, int flags)
653 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
654 also present on *BSD platforms, and possibly elsewhere. */
655 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
657 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
658 char *patcopy = (char *) alloca (strlen (pattern) + 1);
659 char *strcopy = (char *) alloca (strlen (string) + 1);
661 for (p = patcopy; *pattern; pattern++, p++)
662 *p = c_tolower (*pattern);
664 for (p = strcopy; *string; string++, p++)
665 *p = c_tolower (*string);
667 return fnmatch (patcopy, strcopy, flags);
671 static bool in_acclist (const char *const *, const char *, bool);
673 /* Determine whether a file is acceptable to be followed, according to
674 lists of patterns to accept/reject. */
676 acceptable (const char *s)
680 while (l && s[l] != '/')
687 return (in_acclist ((const char *const *)opt.accepts, s, true)
688 && !in_acclist ((const char *const *)opt.rejects, s, true));
690 return in_acclist ((const char *const *)opt.accepts, s, true);
692 else if (opt.rejects)
693 return !in_acclist ((const char *const *)opt.rejects, s, true);
697 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
698 will return true if and only if D2 begins with `/something/' or is exactly
701 subdir_p (const char *d1, const char *d2)
705 if (!opt.ignore_case)
706 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
709 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
712 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
715 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
716 first element that matches DIR, through wildcards or front comparison (as
719 dir_matches_p (char **dirlist, const char *dir)
722 int (*matcher) (const char *, const char *, int)
723 = opt.ignore_case ? fnmatch_nocase : fnmatch;
725 for (x = dirlist; *x; x++)
727 /* Remove leading '/' */
728 char *p = *x + (**x == '/');
729 if (has_wildcards_p (p))
731 if (matcher (p, dir, FNM_PATHNAME) == 0)
736 if (subdir_p (p, dir))
741 return *x ? true : false;
744 /* Returns whether DIRECTORY is acceptable for download, wrt the
745 include/exclude lists.
747 The leading `/' is ignored in paths; relative and absolute paths
748 may be freely intermixed. */
751 accdir (const char *directory)
753 /* Remove starting '/'. */
754 if (*directory == '/')
758 if (!dir_matches_p (opt.includes, directory))
763 if (dir_matches_p (opt.excludes, directory))
769 /* Return true if STRING ends with TAIL. For instance:
771 match_tail ("abc", "bc", false) -> 1
772 match_tail ("abc", "ab", false) -> 0
773 match_tail ("abc", "abc", false) -> 1
775 If FOLD_CASE is true, the comparison will be case-insensitive. */
778 match_tail (const char *string, const char *tail, bool fold_case)
782 /* We want this to be fast, so we code two loops, one with
783 case-folding, one without. */
787 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
788 if (string[i] != tail[j])
793 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
794 if (c_tolower (string[i]) != c_tolower (tail[j]))
798 /* If the tail was exhausted, the match was succesful. */
805 /* Checks whether string S matches each element of ACCEPTS. A list
806 element are matched either with fnmatch() or match_tail(),
807 according to whether the element contains wildcards or not.
809 If the BACKWARD is false, don't do backward comparison -- just compare
812 in_acclist (const char *const *accepts, const char *s, bool backward)
814 for (; *accepts; accepts++)
816 if (has_wildcards_p (*accepts))
818 int res = opt.ignore_case
819 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
820 /* fnmatch returns 0 if the pattern *does* match the string. */
828 if (match_tail (s, *accepts, opt.ignore_case))
833 int cmp = opt.ignore_case
834 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
843 /* Return the location of STR's suffix (file extension). Examples:
844 suffix ("foo.bar") -> "bar"
845 suffix ("foo.bar.baz") -> "baz"
846 suffix ("/foo/bar") -> NULL
847 suffix ("/foo.bar/baz") -> NULL */
849 suffix (const char *str)
853 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
857 return (char *)str + i;
862 /* Return true if S contains globbing wildcards (`*', `?', `[' or
866 has_wildcards_p (const char *s)
869 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
874 /* Return true if FNAME ends with a typical HTML suffix. The
875 following (case-insensitive) suffixes are presumed to be HTML
880 ?html (`?' matches one character)
882 #### CAVEAT. This is not necessarily a good indication that FNAME
883 refers to a file that contains HTML! */
885 has_html_suffix_p (const char *fname)
889 if ((suf = suffix (fname)) == NULL)
891 if (!strcasecmp (suf, "html"))
893 if (!strcasecmp (suf, "htm"))
895 if (suf[0] && !strcasecmp (suf + 1, "html"))
900 /* Read a line from FP and return the pointer to freshly allocated
901 storage. The storage space is obtained through malloc() and should
902 be freed with free() when it is no longer needed.
904 The length of the line is not limited, except by available memory.
905 The newline character at the end of line is retained. The line is
906 terminated with a zero character.
908 After end-of-file is encountered without anything being read, NULL
909 is returned. NULL is also returned on error. To distinguish
910 between these two cases, use the stdio function ferror(). */
913 read_whole_line (FILE *fp)
917 char *line = xmalloc (bufsize);
919 while (fgets (line + length, bufsize - length, fp))
921 length += strlen (line + length);
923 /* Possible for example when reading from a binary file where
924 a line begins with \0. */
927 if (line[length - 1] == '\n')
930 /* fgets() guarantees to read the whole line, or to use up the
931 space we've given it. We can double the buffer
934 line = xrealloc (line, bufsize);
936 if (length == 0 || ferror (fp))
941 if (length + 1 < bufsize)
942 /* Relieve the memory from our exponential greediness. We say
943 `length + 1' because the terminating \0 is not included in
944 LENGTH. We don't need to zero-terminate the string ourselves,
945 though, because fgets() does that. */
946 line = xrealloc (line, length + 1);
950 /* Read FILE into memory. A pointer to `struct file_memory' are
951 returned; use struct element `content' to access file contents, and
952 the element `length' to know the file length. `content' is *not*
953 zero-terminated, and you should *not* read or write beyond the [0,
954 length) range of characters.
956 After you are done with the file contents, call read_file_free to
959 Depending on the operating system and the type of file that is
960 being read, read_file() either mmap's the file into memory, or
961 reads the file into the core using read().
963 If file is named "-", fileno(stdin) is used for reading instead.
964 If you want to read from a real file named "-", use "./-" instead. */
967 read_file (const char *file)
970 struct file_memory *fm;
972 bool inhibit_close = false;
974 /* Some magic in the finest tradition of Perl and its kin: if FILE
975 is "-", just use stdin. */
979 inhibit_close = true;
980 /* Note that we don't inhibit mmap() in this case. If stdin is
981 redirected from a regular file, mmap() will still work. */
984 fd = open (file, O_RDONLY);
987 fm = xnew (struct file_memory);
992 if (fstat (fd, &buf) < 0)
994 fm->length = buf.st_size;
995 /* NOTE: As far as I know, the callers of this function never
996 modify the file text. Relying on this would enable us to
997 specify PROT_READ and MAP_SHARED for a marginal gain in
998 efficiency, but at some cost to generality. */
999 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1000 MAP_PRIVATE, fd, 0);
1001 if (fm->content == (char *)MAP_FAILED)
1011 /* The most common reason why mmap() fails is that FD does not point
1012 to a plain file. However, it's also possible that mmap() doesn't
1013 work for a particular type of file. Therefore, whenever mmap()
1014 fails, we just fall back to the regular method. */
1015 #endif /* HAVE_MMAP */
1018 size = 512; /* number of bytes fm->contents can
1019 hold at any given time. */
1020 fm->content = xmalloc (size);
1024 if (fm->length > size / 2)
1026 /* #### I'm not sure whether the whole exponential-growth
1027 thing makes sense with kernel read. On Linux at least,
1028 read() refuses to read more than 4K from a file at a
1029 single chunk anyway. But other Unixes might optimize it
1030 better, and it doesn't *hurt* anything, so I'm leaving
1033 /* Normally, we grow SIZE exponentially to make the number
1034 of calls to read() and realloc() logarithmic in relation
1035 to file size. However, read() can read an amount of data
1036 smaller than requested, and it would be unreasonable to
1037 double SIZE every time *something* was read. Therefore,
1038 we double SIZE only when the length exceeds half of the
1039 entire allocated size. */
1041 fm->content = xrealloc (fm->content, size);
1043 nread = read (fd, fm->content + fm->length, size - fm->length);
1045 /* Successful read. */
1046 fm->length += nread;
1056 if (size > fm->length && fm->length != 0)
1057 /* Due to exponential growth of fm->content, the allocated region
1058 might be much larger than what is actually needed. */
1059 fm->content = xrealloc (fm->content, fm->length);
1066 xfree (fm->content);
1071 /* Release the resources held by FM. Specifically, this calls
1072 munmap() or xfree() on fm->content, depending whether mmap or
1073 malloc/read were used to read in the file. It also frees the
1074 memory needed to hold the FM structure itself. */
1077 read_file_free (struct file_memory *fm)
1082 munmap (fm->content, fm->length);
1087 xfree (fm->content);
1092 /* Free the pointers in a NULL-terminated vector of pointers, then
1093 free the pointer itself. */
1095 free_vec (char **vec)
1106 /* Append vector V2 to vector V1. The function frees V2 and
1107 reallocates V1 (thus you may not use the contents of neither
1108 pointer after the call). If V1 is NULL, V2 is returned. */
1110 merge_vecs (char **v1, char **v2)
1120 /* To avoid j == 0 */
1125 for (i = 0; v1[i]; i++)
1128 for (j = 0; v2[j]; j++)
1130 /* Reallocate v1. */
1131 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1132 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1137 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1138 is allocated as needed. Return the new value of the vector. */
1141 vec_append (char **vec, const char *str)
1143 int cnt; /* count of vector elements, including
1144 the one we're about to append */
1147 for (cnt = 0; vec[cnt]; cnt++)
1153 /* Reallocate the array to fit the new element and the NULL. */
1154 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1155 /* Append a copy of STR to the vector. */
1156 vec[cnt - 1] = xstrdup (str);
1161 /* Sometimes it's useful to create "sets" of strings, i.e. special
1162 hash tables where you want to store strings as keys and merely
1163 query for their existence. Here is a set of utility routines that
1164 makes that transparent. */
1167 string_set_add (struct hash_table *ht, const char *s)
1169 /* First check whether the set element already exists. If it does,
1170 do nothing so that we don't have to free() the old element and
1171 then strdup() a new one. */
1172 if (hash_table_contains (ht, s))
1175 /* We use "1" as value. It provides us a useful and clear arbitrary
1176 value, and it consumes no memory -- the pointers to the same
1177 string "1" will be shared by all the key-value pairs in all `set'
1179 hash_table_put (ht, xstrdup (s), "1");
1182 /* Synonym for hash_table_contains... */
1185 string_set_contains (struct hash_table *ht, const char *s)
1187 return hash_table_contains (ht, s);
1190 /* Convert the specified string set to array. ARRAY should be large
1191 enough to hold hash_table_count(ht) char pointers. */
1193 void string_set_to_array (struct hash_table *ht, char **array)
1195 hash_table_iterator iter;
1196 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1197 *array++ = iter.key;
1200 /* Free the string set. This frees both the storage allocated for
1201 keys and the actual hash table. (hash_table_destroy would only
1202 destroy the hash table.) */
1205 string_set_free (struct hash_table *ht)
1207 hash_table_iterator iter;
1208 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1210 hash_table_destroy (ht);
1213 /* Utility function: simply call xfree() on all keys and values of HT. */
1216 free_keys_and_values (struct hash_table *ht)
1218 hash_table_iterator iter;
1219 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1226 /* Get digit grouping data for thousand separors by calling
1227 localeconv(). The data includes separator string and grouping info
1228 and is cached after the first call to the function.
1230 In locales that don't set a thousand separator (such as the "C"
1231 locale), this forces it to be ",". We are now only showing
1232 thousand separators in one place, so this shouldn't be a problem in
1236 get_grouping_data (const char **sep, const char **grouping)
1238 static const char *cached_sep;
1239 static const char *cached_grouping;
1240 static bool initialized;
1243 /* Get the grouping info from the locale. */
1244 struct lconv *lconv = localeconv ();
1245 cached_sep = lconv->thousands_sep;
1246 cached_grouping = lconv->grouping;
1247 #if ! USE_NLS_PROGRESS_BAR
1248 /* We can't count column widths, so ensure that the separator
1249 * is single-byte only (let check below determine what byte). */
1250 if (strlen(cached_sep) > 1)
1255 /* Many locales (such as "C" or "hr_HR") don't specify
1256 grouping, which we still want to use it for legibility.
1257 In those locales set the sep char to ',', unless that
1258 character is used for decimal point, in which case set it
1260 if (*lconv->decimal_point != ',')
1264 cached_grouping = "\x03";
1269 *grouping = cached_grouping;
1272 /* Return a printed representation of N with thousand separators.
1273 This should respect locale settings, with the exception of the "C"
1274 locale which mandates no separator, but we use one anyway.
1276 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1277 the separators because it's too non-portable, and it's hard to test
1278 for this feature at configure time. Besides, it wouldn't display
1279 separators in the "C" locale, still used by many Unix users. */
1282 with_thousand_seps (wgint n)
1284 static char outbuf[48];
1285 char *p = outbuf + sizeof outbuf;
1287 /* Info received from locale */
1288 const char *grouping, *sep;
1291 /* State information */
1292 int i = 0, groupsize;
1293 const char *atgroup;
1295 bool negative = n < 0;
1297 /* Initialize grouping data. */
1298 get_grouping_data (&sep, &grouping);
1299 seplen = strlen (sep);
1301 groupsize = *atgroup++;
1303 /* This would overflow on WGINT_MIN, but printing negative numbers
1304 is not an important goal of this fuinction. */
1308 /* Write the number into the buffer, backwards, inserting the
1309 separators as necessary. */
1313 *--p = n % 10 + '0';
1317 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1318 if (++i == groupsize)
1323 memcpy (p -= seplen, sep, seplen);
1326 groupsize = *atgroup++;
1335 /* N, a byte quantity, is converted to a human-readable abberviated
1336 form a la sizes printed by `ls -lh'. The result is written to a
1337 static buffer, a pointer to which is returned.
1339 Unlike `with_thousand_seps', this approximates to the nearest unit.
1340 Quoting GNU libit: "Most people visually process strings of 3-4
1341 digits effectively, but longer strings of digits are more prone to
1342 misinterpretation. Hence, converting to an abbreviated form
1343 usually improves readability."
1345 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1346 original computer-related meaning of "powers of 1024". We don't
1347 use the "*bibyte" names invented in 1998, and seldom used in
1348 practice. Wikipedia's entry on "binary prefix" discusses this in
1352 human_readable (HR_NUMTYPE n)
1354 /* These suffixes are compatible with those of GNU `ls -lh'. */
1355 static char powers[] =
1357 'K', /* kilobyte, 2^10 bytes */
1358 'M', /* megabyte, 2^20 bytes */
1359 'G', /* gigabyte, 2^30 bytes */
1360 'T', /* terabyte, 2^40 bytes */
1361 'P', /* petabyte, 2^50 bytes */
1362 'E', /* exabyte, 2^60 bytes */
1367 /* If the quantity is smaller than 1K, just print it. */
1370 snprintf (buf, sizeof (buf), "%d", (int) n);
1374 /* Loop over powers, dividing N with 1024 in each iteration. This
1375 works unchanged for all sizes of wgint, while still avoiding
1376 non-portable `long double' arithmetic. */
1377 for (i = 0; i < countof (powers); i++)
1379 /* At each iteration N is greater than the *subsequent* power.
1380 That way N/1024.0 produces a decimal number in the units of
1382 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1384 double val = n / 1024.0;
1385 /* Print values smaller than 10 with one decimal digits, and
1386 others without any decimals. */
1387 snprintf (buf, sizeof (buf), "%.*f%c",
1388 val < 10 ? 1 : 0, val, powers[i]);
1393 return NULL; /* unreached */
1396 /* Count the digits in the provided number. Used to allocate space
1397 when printing numbers. */
1400 numdigit (wgint number)
1404 ++cnt; /* accomodate '-' */
1405 while ((number /= 10) != 0)
1410 #define PR(mask) *p++ = n / (mask) + '0'
1412 /* DIGITS_<D> is used to print a D-digit number and should be called
1413 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1414 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1415 Recursively this continues until DIGITS_1 is invoked. */
1417 #define DIGITS_1(mask) PR (mask)
1418 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1419 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1420 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1421 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1422 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1423 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1424 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1425 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1426 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1428 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1430 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1431 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1432 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1433 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1434 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1435 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1436 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1437 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1438 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1440 /* Shorthand for casting to wgint. */
1443 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1444 `sprintf(buffer, "%lld", (long long) number)', only typically much
1445 faster and portable to machines without long long.
1447 The speedup may make a difference in programs that frequently
1448 convert numbers to strings. Some implementations of sprintf,
1449 particularly the one in some versions of GNU libc, have been known
1450 to be quite slow when converting integers to strings.
1452 Return the pointer to the location where the terminating zero was
1453 printed. (Equivalent to calling buffer+strlen(buffer) after the
1456 BUFFER should be large enough to accept as many bytes as you expect
1457 the number to take up. On machines with 64-bit wgints the maximum
1458 needed size is 24 bytes. That includes the digits needed for the
1459 largest 64-bit number, the `-' sign in case it's negative, and the
1460 terminating '\0'. */
1463 number_to_string (char *buffer, wgint number)
1468 int last_digit_char = 0;
1470 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1471 /* We are running in a very strange environment. Leave the correct
1472 printing to sprintf. */
1473 p += sprintf (buf, "%j", (intmax_t) (n));
1474 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1480 /* n = -n would overflow because -n would evaluate to a
1481 wgint value larger than WGINT_MAX. Need to make n
1482 smaller and handle the last digit separately. */
1483 int last_digit = n % 10;
1484 /* The sign of n%10 is implementation-defined. */
1486 last_digit_char = '0' - last_digit;
1488 last_digit_char = '0' + last_digit;
1489 /* After n is made smaller, -n will not overflow. */
1497 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1498 way printing any N is fully open-coded without a loop or jump.
1499 (Also see description of DIGITS_*.) */
1501 if (n < 10) DIGITS_1 (1);
1502 else if (n < 100) DIGITS_2 (10);
1503 else if (n < 1000) DIGITS_3 (100);
1504 else if (n < 10000) DIGITS_4 (1000);
1505 else if (n < 100000) DIGITS_5 (10000);
1506 else if (n < 1000000) DIGITS_6 (100000);
1507 else if (n < 10000000) DIGITS_7 (1000000);
1508 else if (n < 100000000) DIGITS_8 (10000000);
1509 else if (n < 1000000000) DIGITS_9 (100000000);
1510 #if SIZEOF_WGINT == 4
1511 /* wgint is 32 bits wide: no number has more than 10 digits. */
1512 else DIGITS_10 (1000000000);
1514 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1515 Constants are constructed by compile-time multiplication to avoid
1516 dealing with different notations for 64-bit constants
1517 (nL/nLL/nI64, depending on the compiler and architecture). */
1518 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1519 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1520 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1521 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1522 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1523 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1524 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1525 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1526 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1527 else DIGITS_19 (1000000000*(W)1000000000);
1530 if (last_digit_char)
1531 *p++ = last_digit_char;
1534 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1541 #undef SPRINTF_WGINT
1564 /* Print NUMBER to a statically allocated string and return a pointer
1565 to the printed representation.
1567 This function is intended to be used in conjunction with printf.
1568 It is hard to portably print wgint values:
1569 a) you cannot use printf("%ld", number) because wgint can be long
1570 long on 32-bit machines with LFS.
1571 b) you cannot use printf("%lld", number) because NUMBER could be
1572 long on 32-bit machines without LFS, or on 64-bit machines,
1573 which do not require LFS. Also, Windows doesn't support %lld.
1574 c) you cannot use printf("%j", (int_max_t) number) because not all
1575 versions of printf support "%j", the most notable being the one
1577 d) you cannot #define WGINT_FMT to the appropriate format and use
1578 printf(WGINT_FMT, number) because that would break translations
1579 for user-visible messages, such as printf("Downloaded: %d
1582 What you should use instead is printf("%s", number_to_static_string
1585 CAVEAT: since the function returns pointers to static data, you
1586 must be careful to copy its result before calling it again.
1587 However, to make it more useful with printf, the function maintains
1588 an internal ring of static buffers to return. That way things like
1589 printf("%s %s", number_to_static_string (num1),
1590 number_to_static_string (num2)) work as expected. Three buffers
1591 are currently used, which means that "%s %s %s" will work, but "%s
1592 %s %s %s" won't. If you need to print more than three wgints,
1593 bump the RING_SIZE (or rethink your message.) */
1596 number_to_static_string (wgint number)
1598 static char ring[RING_SIZE][24];
1600 char *buf = ring[ringpos];
1601 number_to_string (buf, number);
1602 ringpos = (ringpos + 1) % RING_SIZE;
1606 /* Determine the width of the terminal we're running on. If that's
1607 not possible, return 0. */
1610 determine_screen_width (void)
1612 /* If there's a way to get the terminal size using POSIX
1613 tcgetattr(), somebody please tell me. */
1618 if (opt.lfilename != NULL)
1621 fd = fileno (stderr);
1622 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1623 return 0; /* most likely ENOTTY */
1626 #elif defined(WINDOWS)
1627 CONSOLE_SCREEN_BUFFER_INFO csbi;
1628 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1630 return csbi.dwSize.X;
1631 #else /* neither TIOCGWINSZ nor WINDOWS */
1633 #endif /* neither TIOCGWINSZ nor WINDOWS */
1636 /* Whether the rnd system (either rand or [dl]rand48) has been
1638 static int rnd_seeded;
1640 /* Return a random number between 0 and MAX-1, inclusive.
1642 If the system does not support lrand48 and MAX is greater than the
1643 value of RAND_MAX+1 on the system, the returned value will be in
1644 the range [0, RAND_MAX]. This may be fixed in a future release.
1645 The random number generator is seeded automatically the first time
1648 This uses lrand48 where available, rand elsewhere. DO NOT use it
1649 for cryptography. It is only meant to be used in situations where
1650 quality of the random numbers returned doesn't really matter. */
1653 random_number (int max)
1658 srand48 ((long) time (NULL) ^ (long) getpid ());
1661 return lrand48 () % max;
1662 #else /* not HAVE_DRAND48 */
1668 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1673 /* Like rand() % max, but uses the high-order bits for better
1674 randomness on architectures where rand() is implemented using a
1675 simple congruential generator. */
1677 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1678 return (int) bounded;
1680 #endif /* not HAVE_DRAND48 */
1683 /* Return a random uniformly distributed floating point number in the
1684 [0, 1) range. Uses drand48 where available, and a really lame
1685 kludge elsewhere. */
1693 srand48 ((long) time (NULL) ^ (long) getpid ());
1697 #else /* not HAVE_DRAND48 */
1698 return ( random_number (10000) / 10000.0
1699 + random_number (10000) / (10000.0 * 10000.0)
1700 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1701 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1702 #endif /* not HAVE_DRAND48 */
1705 /* Implementation of run_with_timeout, a generic timeout-forcing
1706 routine for systems with Unix-like signal handling. */
1708 #ifdef USE_SIGNAL_TIMEOUT
1709 # ifdef HAVE_SIGSETJMP
1710 # define SETJMP(env) sigsetjmp (env, 1)
1712 static sigjmp_buf run_with_timeout_env;
1715 abort_run_with_timeout (int sig)
1717 assert (sig == SIGALRM);
1718 siglongjmp (run_with_timeout_env, -1);
1720 # else /* not HAVE_SIGSETJMP */
1721 # define SETJMP(env) setjmp (env)
1723 static jmp_buf run_with_timeout_env;
1726 abort_run_with_timeout (int sig)
1728 assert (sig == SIGALRM);
1729 /* We don't have siglongjmp to preserve the set of blocked signals;
1730 if we longjumped out of the handler at this point, SIGALRM would
1731 remain blocked. We must unblock it manually. */
1732 int mask = siggetmask ();
1733 mask &= ~sigmask (SIGALRM);
1736 /* Now it's safe to longjump. */
1737 longjmp (run_with_timeout_env, -1);
1739 # endif /* not HAVE_SIGSETJMP */
1741 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1742 setitimer where available, alarm otherwise.
1744 TIMEOUT should be non-zero. If the timeout value is so small that
1745 it would be rounded to zero, it is rounded to the least legal value
1746 instead (1us for setitimer, 1s for alarm). That ensures that
1747 SIGALRM will be delivered in all cases. */
1750 alarm_set (double timeout)
1753 /* Use the modern itimer interface. */
1754 struct itimerval itv;
1756 itv.it_value.tv_sec = (long) timeout;
1757 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1758 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1759 /* Ensure that we wait for at least the minimum interval.
1760 Specifying zero would mean "wait forever". */
1761 itv.it_value.tv_usec = 1;
1762 setitimer (ITIMER_REAL, &itv, NULL);
1763 #else /* not ITIMER_REAL */
1764 /* Use the old alarm() interface. */
1765 int secs = (int) timeout;
1767 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1768 because alarm(0) means "never deliver the alarm", i.e. "wait
1769 forever", which is not what someone who specifies a 0.5s
1770 timeout would expect. */
1773 #endif /* not ITIMER_REAL */
1776 /* Cancel the alarm set with alarm_set. */
1782 struct itimerval disable;
1784 setitimer (ITIMER_REAL, &disable, NULL);
1785 #else /* not ITIMER_REAL */
1787 #endif /* not ITIMER_REAL */
1790 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1791 seconds. Returns true if the function was interrupted with a
1792 timeout, false otherwise.
1794 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1795 using setitimer() or alarm(). The timeout is enforced by
1796 longjumping out of the SIGALRM handler. This has several
1797 advantages compared to the traditional approach of relying on
1798 signals causing system calls to exit with EINTR:
1800 * The callback function is *forcibly* interrupted after the
1801 timeout expires, (almost) regardless of what it was doing and
1802 whether it was in a syscall. For example, a calculation that
1803 takes a long time is interrupted as reliably as an IO
1806 * It works with both SYSV and BSD signals because it doesn't
1807 depend on the default setting of SA_RESTART.
1809 * It doesn't require special handler setup beyond a simple call
1810 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1813 The only downside is that, if FUN allocates internal resources that
1814 are normally freed prior to exit from the functions, they will be
1815 lost in case of timeout. */
1818 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1828 signal (SIGALRM, abort_run_with_timeout);
1829 if (SETJMP (run_with_timeout_env) != 0)
1831 /* Longjumped out of FUN with a timeout. */
1832 signal (SIGALRM, SIG_DFL);
1835 alarm_set (timeout);
1838 /* Preserve errno in case alarm() or signal() modifies it. */
1839 saved_errno = errno;
1841 signal (SIGALRM, SIG_DFL);
1842 errno = saved_errno;
1847 #else /* not USE_SIGNAL_TIMEOUT */
1850 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1851 define it under Windows, because Windows has its own version of
1852 run_with_timeout that uses threads. */
1855 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1860 #endif /* not WINDOWS */
1861 #endif /* not USE_SIGNAL_TIMEOUT */
1865 /* Sleep the specified amount of seconds. On machines without
1866 nanosleep(), this may sleep shorter if interrupted by signals. */
1869 xsleep (double seconds)
1871 #ifdef HAVE_NANOSLEEP
1872 /* nanosleep is the preferred interface because it offers high
1873 accuracy and, more importantly, because it allows us to reliably
1874 restart receiving a signal such as SIGWINCH. (There was an
1875 actual Debian bug report about --limit-rate malfunctioning while
1876 the terminal was being resized.) */
1877 struct timespec sleep, remaining;
1878 sleep.tv_sec = (long) seconds;
1879 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1880 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1881 /* If nanosleep has been interrupted by a signal, adjust the
1882 sleeping period and return to sleep. */
1884 #elif defined(HAVE_USLEEP)
1885 /* If usleep is available, use it in preference to select. */
1888 /* On some systems, usleep cannot handle values larger than
1889 1,000,000. If the period is larger than that, use sleep
1890 first, then add usleep for subsecond accuracy. */
1892 seconds -= (long) seconds;
1894 usleep (seconds * 1000000);
1895 #else /* fall back select */
1896 /* Note that, although Windows supports select, it can't be used to
1897 implement sleeping because Winsock's select doesn't implement
1898 timeout when it is passed NULL pointers for all fd sets. (But it
1899 does under Cygwin, which implements Unix-compatible select.) */
1900 struct timeval sleep;
1901 sleep.tv_sec = (long) seconds;
1902 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1903 select (0, NULL, NULL, NULL, &sleep);
1904 /* If select returns -1 and errno is EINTR, it means we were
1905 interrupted by a signal. But without knowing how long we've
1906 actually slept, we can't return to sleep. Using gettimeofday to
1907 track sleeps is slow and unreliable due to clock skew. */
1911 #endif /* not WINDOWS */
1913 /* Encode the octets in DATA of length LENGTH to base64 format,
1914 storing the result to DEST. The output will be zero-terminated,
1915 and must point to a writable buffer of at least
1916 1+BASE64_LENGTH(length) bytes. The function returns the length of
1917 the resulting base64 data, not counting the terminating zero.
1919 This implementation does not emit newlines after 76 characters of
1923 base64_encode (const void *data, int length, char *dest)
1925 /* Conversion table. */
1926 static const char tbl[64] = {
1927 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
1928 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
1929 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
1930 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
1932 /* Access bytes in DATA as unsigned char, otherwise the shifts below
1933 don't work for data with MSB set. */
1934 const unsigned char *s = data;
1935 /* Theoretical ANSI violation when length < 3. */
1936 const unsigned char *end = (const unsigned char *) data + length - 2;
1939 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1940 for (; s < end; s += 3)
1942 *p++ = tbl[s[0] >> 2];
1943 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1944 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1945 *p++ = tbl[s[2] & 0x3f];
1948 /* Pad the result if necessary... */
1952 *p++ = tbl[s[0] >> 2];
1953 *p++ = tbl[(s[0] & 3) << 4];
1958 *p++ = tbl[s[0] >> 2];
1959 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1960 *p++ = tbl[((s[1] & 0xf) << 2)];
1964 /* ...and zero-terminate it. */
1970 /* Store in C the next non-whitespace character from the string, or \0
1971 when end of string is reached. */
1972 #define NEXT_CHAR(c, p) do { \
1973 c = (unsigned char) *p++; \
1974 } while (c_isspace (c))
1976 #define IS_ASCII(c) (((c) & 0x80) == 0)
1978 /* Decode data from BASE64 (a null-terminated string) into memory
1979 pointed to by DEST. DEST is assumed to be large enough to
1980 accomodate the decoded data, which is guaranteed to be no more than
1983 Since DEST is assumed to contain binary data, it is not
1984 NUL-terminated. The function returns the length of the data
1985 written to TO. -1 is returned in case of error caused by malformed
1988 This function originates from Free Recode. */
1991 base64_decode (const char *base64, void *dest)
1993 /* Table of base64 values for first 128 characters. Note that this
1994 assumes ASCII (but so does Wget in other places). */
1995 static const signed char base64_char_to_value[128] =
1997 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1998 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1999 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2000 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2001 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2002 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2003 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2004 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2005 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2006 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2007 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2008 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2009 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2011 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2012 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2014 const char *p = base64;
2020 unsigned long value;
2022 /* Process first byte of a quadruplet. */
2026 if (c == '=' || !IS_BASE64 (c))
2027 return -1; /* illegal char while decoding base64 */
2028 value = BASE64_CHAR_TO_VALUE (c) << 18;
2030 /* Process second byte of a quadruplet. */
2033 return -1; /* premature EOF while decoding base64 */
2034 if (c == '=' || !IS_BASE64 (c))
2035 return -1; /* illegal char while decoding base64 */
2036 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2039 /* Process third byte of a quadruplet. */
2042 return -1; /* premature EOF while decoding base64 */
2044 return -1; /* illegal char while decoding base64 */
2050 return -1; /* premature EOF while decoding base64 */
2052 return -1; /* padding `=' expected but not found */
2056 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2057 *q++ = 0xff & value >> 8;
2059 /* Process fourth byte of a quadruplet. */
2062 return -1; /* premature EOF while decoding base64 */
2066 return -1; /* illegal char while decoding base64 */
2068 value |= BASE64_CHAR_TO_VALUE (c);
2069 *q++ = 0xff & value;
2072 #undef BASE64_CHAR_TO_VALUE
2074 return q - (char *) dest;
2080 /* Simple merge sort for use by stable_sort. Implementation courtesy
2081 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2084 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2085 int (*cmpfun) (const void *, const void *))
2087 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2091 size_t mid = (to + from) / 2;
2092 mergesort_internal (base, temp, size, from, mid, cmpfun);
2093 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2096 for (k = from; (i <= mid) && (j <= to); k++)
2097 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2098 memcpy (ELT (temp, k), ELT (base, i++), size);
2100 memcpy (ELT (temp, k), ELT (base, j++), size);
2102 memcpy (ELT (temp, k++), ELT (base, i++), size);
2104 memcpy (ELT (temp, k++), ELT (base, j++), size);
2105 for (k = from; k <= to; k++)
2106 memcpy (ELT (base, k), ELT (temp, k), size);
2111 /* Stable sort with interface exactly like standard library's qsort.
2112 Uses mergesort internally, allocating temporary storage with
2116 stable_sort (void *base, size_t nmemb, size_t size,
2117 int (*cmpfun) (const void *, const void *))
2121 void *temp = alloca (nmemb * size * sizeof (void *));
2122 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2126 /* Print a decimal number. If it is equal to or larger than ten, the
2127 number is rounded. Otherwise it is printed with one significant
2128 digit without trailing zeros and with no more than three fractional
2129 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2130 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2132 This is useful for displaying durations because it provides
2133 order-of-magnitude information without unnecessary clutter --
2134 long-running downloads are shown without the fractional part, and
2135 short ones still retain one significant digit. */
2138 print_decimal (double number)
2140 static char buf[32];
2141 double n = number >= 0 ? number : -number;
2144 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2145 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2146 snprintf (buf, sizeof buf, "%.0f", number);
2148 snprintf (buf, sizeof buf, "%.1f", number);
2149 else if (n >= 0.001)
2150 snprintf (buf, sizeof buf, "%.1g", number);
2151 else if (n >= 0.0005)
2152 /* round [0.0005, 0.001) to 0.001 */
2153 snprintf (buf, sizeof buf, "%.3f", number);
2155 /* print numbers close to 0 as 0, not 0.000 */
2172 { "/somedir", "/somedir", true },
2173 { "/somedir", "/somedir/d2", true },
2174 { "/somedir/d1", "/somedir", false },
2177 for (i = 0; i < countof(test_array); ++i)
2179 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2181 mu_assert ("test_subdir_p: wrong result",
2182 res == test_array[i].result);
2189 test_dir_matches_p()
2197 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2198 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2199 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2200 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2201 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2202 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2203 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2204 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2205 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2206 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2207 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2208 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2209 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2212 for (i = 0; i < countof(test_array); ++i)
2214 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2216 mu_assert ("test_dir_matches_p: wrong result",
2217 res == test_array[i].result);
2223 #endif /* TESTING */