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
92 memfatal (const char *context, long attempted_size)
94 /* Make sure we don't try to store part of the log line, and thus
96 log_set_save_context (false);
98 /* We have different log outputs in different situations:
99 1) output without bytes information
100 2) output with bytes information */
101 if (attempted_size == UNKNOWN_ATTEMPTED_SIZE)
103 logprintf (LOG_ALWAYS,
104 _("%s: %s: Failed to allocate enough memory; memory exhausted.\n"),
109 logprintf (LOG_ALWAYS,
110 _("%s: %s: Failed to allocate %ld bytes; memory exhausted.\n"),
111 exec_name, context, attempted_size);
117 /* Utility function: like xstrdup(), but also lowercases S. */
120 xstrdup_lower (const char *s)
122 char *copy = xstrdup (s);
129 /* Copy the string formed by two pointers (one on the beginning, other
130 on the char after the last char) to a new, malloc-ed location.
133 strdupdelim (const char *beg, const char *end)
135 char *res = xmalloc (end - beg + 1);
136 memcpy (res, beg, end - beg);
137 res[end - beg] = '\0';
141 /* Parse a string containing comma-separated elements, and return a
142 vector of char pointers with the elements. Spaces following the
143 commas are ignored. */
145 sepstring (const char *s)
159 res = xrealloc (res, (i + 2) * sizeof (char *));
160 res[i] = strdupdelim (p, s);
163 /* Skip the blanks following the ','. */
164 while (c_isspace (*s))
171 res = xrealloc (res, (i + 2) * sizeof (char *));
172 res[i] = strdupdelim (p, s);
177 /* Like sprintf, but prints into a string of sufficient size freshly
178 allocated with malloc, which is returned. If unable to print due
179 to invalid format, returns NULL. Inability to allocate needed
180 memory results in abort, as with xmalloc. This is in spirit
181 similar to the GNU/BSD extension asprintf, but somewhat easier to
184 Internally the function either calls vasprintf or loops around
185 vsnprintf until the correct size is found. Since Wget also ships a
186 fallback implementation of vsnprintf, this should be portable. */
188 /* Constant is using for limits memory allocation for text buffer.
189 Applicable in situation when: vasprintf is not available in the system
190 and vsnprintf return -1 when long line is truncated (in old versions of
191 glibc and in other system where C99 doesn`t support) */
193 #define FMT_MAX_LENGTH 1048576
196 aprintf (const char *fmt, ...)
198 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
203 va_start (args, fmt);
204 ret = vasprintf (&str, fmt, args);
206 if (ret < 0 && errno == ENOMEM)
207 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency
208 with xmalloc/xrealloc */
212 #else /* not HAVE_VASPRINTF */
214 /* vasprintf is unavailable. snprintf into a small buffer and
215 resize it as necessary. */
217 char *str = xmalloc (size);
219 /* #### This code will infloop and eventually abort in xrealloc if
220 passed a FMT that causes snprintf to consistently return -1. */
227 va_start (args, fmt);
228 n = vsnprintf (str, size, fmt, args);
231 /* If the printing worked, return the string. */
232 if (n > -1 && n < size)
235 /* Else try again with a larger buffer. */
236 if (n > -1) /* C99 */
237 size = n + 1; /* precisely what is needed */
238 else if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
239 { /* maybe we have some wrong
241 logprintf (LOG_ALWAYS,
242 _("%s: aprintf: text buffer is too big (%ld bytes), "
244 exec_name, size); /* printout a log message */
245 abort (); /* and abort... */
249 /* else, we continue to grow our
250 * buffer: Twice the old size. */
253 str = xrealloc (str, size);
255 #endif /* not HAVE_VASPRINTF */
258 /* Concatenate the NULL-terminated list of string arguments into
259 freshly allocated space. */
262 concat_strings (const char *str0, ...)
265 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
268 const char *next_str;
269 int total_length = 0;
272 /* Calculate the length of and allocate the resulting string. */
275 va_start (args, str0);
276 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
278 int len = strlen (next_str);
279 if (argcount < countof (saved_lengths))
280 saved_lengths[argcount++] = len;
284 p = ret = xmalloc (total_length + 1);
286 /* Copy the strings into the allocated space. */
289 va_start (args, str0);
290 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
293 if (argcount < countof (saved_lengths))
294 len = saved_lengths[argcount++];
296 len = strlen (next_str);
297 memcpy (p, next_str, len);
306 /* Format the provided time according to the specified format. The
307 format is a string with format elements supported by strftime. */
310 fmttime (time_t t, const char *fmt)
312 static char output[32];
313 struct tm *tm = localtime(&t);
316 if (!strftime(output, sizeof(output), fmt, tm))
321 /* Return pointer to a static char[] buffer in which zero-terminated
322 string-representation of TM (in form hh:mm:ss) is printed.
324 If TM is NULL, the current time will be used. */
329 return fmttime(t, "%H:%M:%S");
332 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
335 datetime_str (time_t t)
337 return fmttime(t, "%Y-%m-%d %H:%M:%S");
340 /* The Windows versions of the following two functions are defined in
341 mswindows.c. On MSDOS this function should never be called. */
343 #if !defined(WINDOWS) && !defined(MSDOS)
345 fork_to_background (void)
348 /* Whether we arrange our own version of opt.lfilename here. */
349 bool logfile_changed = false;
351 if (!opt.lfilename && (!opt.quiet || opt.server_response))
353 /* We must create the file immediately to avoid either a race
354 condition (which arises from using unique_name and failing to
355 use fopen_excl) or lying to the user about the log file name
356 (which arises from using unique_name, printing the name, and
357 using fopen_excl later on.) */
358 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
361 logfile_changed = true;
374 /* parent, no error */
375 printf (_("Continuing in background, pid %d.\n"), (int) pid);
377 printf (_("Output will be written to %s.\n"), quote (opt.lfilename));
378 exit (0); /* #### should we use _exit()? */
381 /* child: give up the privileges and keep running. */
383 freopen ("/dev/null", "r", stdin);
384 freopen ("/dev/null", "w", stdout);
385 freopen ("/dev/null", "w", stderr);
387 #endif /* !WINDOWS && !MSDOS */
389 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
390 specified with TM. The atime ("access time") is set to the current
394 touch (const char *file, time_t tm)
396 #ifdef HAVE_STRUCT_UTIMBUF
397 struct utimbuf times;
405 times.actime = time (NULL);
406 if (utime (file, ×) == -1)
407 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
410 /* Checks if FILE is a symbolic link, and removes it if it is. Does
411 nothing under MS-Windows. */
413 remove_link (const char *file)
418 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
420 DEBUGP (("Unlinking %s (symlink).\n", file));
423 logprintf (LOG_VERBOSE, _("Failed to unlink symlink %s: %s\n"),
424 quote (file), strerror (errno));
429 /* Does FILENAME exist? This is quite a lousy implementation, since
430 it supplies no error codes -- only a yes-or-no answer. Thus it
431 will return that a file does not exist if, e.g., the directory is
432 unreadable. I don't mind it too much currently, though. The
433 proper way should, of course, be to have a third, error state,
434 other than true/false, but that would introduce uncalled-for
435 additional complexity to the callers. */
437 file_exists_p (const char *filename)
440 return access (filename, F_OK) >= 0;
443 return stat (filename, &buf) >= 0;
447 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
448 Returns 0 on error. */
450 file_non_directory_p (const char *path)
453 /* Use lstat() rather than stat() so that symbolic links pointing to
454 directories can be identified correctly. */
455 if (lstat (path, &buf) != 0)
457 return S_ISDIR (buf.st_mode) ? false : true;
460 /* Return the size of file named by FILENAME, or -1 if it cannot be
461 opened or seeked into. */
463 file_size (const char *filename)
465 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
467 /* We use fseek rather than stat to determine the file size because
468 that way we can also verify that the file is readable without
469 explicitly checking for permissions. Inspired by the POST patch
471 FILE *fp = fopen (filename, "rb");
474 fseeko (fp, 0, SEEK_END);
480 if (stat (filename, &st) < 0)
486 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
487 doesn't exist is found. Return a freshly allocated copy of the
491 unique_name_1 (const char *prefix)
494 int plen = strlen (prefix);
495 char *template = (char *)alloca (plen + 1 + 24);
496 char *template_tail = template + plen;
498 memcpy (template, prefix, plen);
499 *template_tail++ = '.';
502 number_to_string (template_tail, count++);
503 while (file_exists_p (template));
505 return xstrdup (template);
508 /* Return a unique file name, based on FILE.
510 More precisely, if FILE doesn't exist, it is returned unmodified.
511 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
512 file name that doesn't exist is returned.
514 The resulting file is not created, only verified that it didn't
515 exist at the point in time when the function was called.
516 Therefore, where security matters, don't rely that the file created
517 by this function exists until you open it with O_EXCL or
520 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
521 string. Otherwise, it may return FILE if the file doesn't exist
522 (and therefore doesn't need changing). */
525 unique_name (const char *file, bool allow_passthrough)
527 /* If the FILE itself doesn't exist, return it without
529 if (!file_exists_p (file))
530 return allow_passthrough ? (char *)file : xstrdup (file);
532 /* Otherwise, find a numeric suffix that results in unused file name
534 return unique_name_1 (file);
537 /* Create a file based on NAME, except without overwriting an existing
538 file with that name. Providing O_EXCL is correctly implemented,
539 this function does not have the race condition associated with
540 opening the file returned by unique_name. */
543 unique_create (const char *name, bool binary, char **opened_name)
545 /* unique file name, based on NAME */
546 char *uname = unique_name (name, false);
548 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
551 uname = unique_name (name, false);
553 if (opened_name && fp != NULL)
556 *opened_name = uname;
568 /* Open the file for writing, with the addition that the file is
569 opened "exclusively". This means that, if the file already exists,
570 this function will *fail* and errno will be set to EEXIST. If
571 BINARY is set, the file will be opened in binary mode, equivalent
574 If opening the file fails for any reason, including the file having
575 previously existed, this function returns NULL and sets errno
579 fopen_excl (const char *fname, bool binary)
583 int flags = O_WRONLY | O_CREAT | O_EXCL;
588 fd = open (fname, flags, 0666);
591 return fdopen (fd, binary ? "wb" : "w");
592 #else /* not O_EXCL */
593 /* Manually check whether the file exists. This is prone to race
594 conditions, but systems without O_EXCL haven't deserved
596 if (file_exists_p (fname))
601 return fopen (fname, binary ? "wb" : "w");
602 #endif /* not O_EXCL */
605 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
606 are missing, create them first. In case any mkdir() call fails,
607 return its error status. Returns 0 on successful completion.
609 The behaviour of this function should be identical to the behaviour
610 of `mkdir -p' on systems where mkdir supports the `-p' option. */
612 make_directory (const char *directory)
614 int i, ret, quit = 0;
617 /* Make a copy of dir, to be able to write to it. Otherwise, the
618 function is unsafe if called with a read-only char *argument. */
619 STRDUP_ALLOCA (dir, directory);
621 /* If the first character of dir is '/', skip it (and thus enable
622 creation of absolute-pathname directories. */
623 for (i = (*dir == '/'); 1; ++i)
625 for (; dir[i] && dir[i] != '/'; i++)
630 /* Check whether the directory already exists. Allow creation of
631 of intermediate directories to fail, as the initial path components
632 are not necessarily directories! */
633 if (!file_exists_p (dir))
634 ret = mkdir (dir, 0777);
645 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
646 should be a file name.
648 file_merge("/foo/bar", "baz") => "/foo/baz"
649 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
650 file_merge("foo", "bar") => "bar"
652 In other words, it's a simpler and gentler version of uri_merge. */
655 file_merge (const char *base, const char *file)
658 const char *cut = (const char *)strrchr (base, '/');
661 return xstrdup (file);
663 result = xmalloc (cut - base + 1 + strlen (file) + 1);
664 memcpy (result, base, cut - base);
665 result[cut - base] = '/';
666 strcpy (result + (cut - base) + 1, file);
671 /* Like fnmatch, but performs a case-insensitive match. */
674 fnmatch_nocase (const char *pattern, const char *string, int flags)
677 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
678 also present on *BSD platforms, and possibly elsewhere. */
679 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
681 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
682 char *patcopy = (char *) alloca (strlen (pattern) + 1);
683 char *strcopy = (char *) alloca (strlen (string) + 1);
685 for (p = patcopy; *pattern; pattern++, p++)
686 *p = c_tolower (*pattern);
688 for (p = strcopy; *string; string++, p++)
689 *p = c_tolower (*string);
691 return fnmatch (patcopy, strcopy, flags);
695 static bool in_acclist (const char *const *, const char *, bool);
697 /* Determine whether a file is acceptable to be followed, according to
698 lists of patterns to accept/reject. */
700 acceptable (const char *s)
704 while (l && s[l] != '/')
711 return (in_acclist ((const char *const *)opt.accepts, s, true)
712 && !in_acclist ((const char *const *)opt.rejects, s, true));
714 return in_acclist ((const char *const *)opt.accepts, s, true);
716 else if (opt.rejects)
717 return !in_acclist ((const char *const *)opt.rejects, s, true);
721 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
722 will return true if and only if D2 begins with `/something/' or is exactly
725 subdir_p (const char *d1, const char *d2)
729 if (!opt.ignore_case)
730 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
733 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
736 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
739 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
740 first element that matches DIR, through wildcards or front comparison (as
743 dir_matches_p (char **dirlist, const char *dir)
746 int (*matcher) (const char *, const char *, int)
747 = opt.ignore_case ? fnmatch_nocase : fnmatch;
749 for (x = dirlist; *x; x++)
751 /* Remove leading '/' */
752 char *p = *x + (**x == '/');
753 if (has_wildcards_p (p))
755 if (matcher (p, dir, FNM_PATHNAME) == 0)
760 if (subdir_p (p, dir))
765 return *x ? true : false;
768 /* Returns whether DIRECTORY is acceptable for download, wrt the
769 include/exclude lists.
771 The leading `/' is ignored in paths; relative and absolute paths
772 may be freely intermixed. */
775 accdir (const char *directory)
777 /* Remove starting '/'. */
778 if (*directory == '/')
782 if (!dir_matches_p (opt.includes, directory))
787 if (dir_matches_p (opt.excludes, directory))
793 /* Return true if STRING ends with TAIL. For instance:
795 match_tail ("abc", "bc", false) -> 1
796 match_tail ("abc", "ab", false) -> 0
797 match_tail ("abc", "abc", false) -> 1
799 If FOLD_CASE is true, the comparison will be case-insensitive. */
802 match_tail (const char *string, const char *tail, bool fold_case)
806 /* We want this to be fast, so we code two loops, one with
807 case-folding, one without. */
811 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
812 if (string[i] != tail[j])
817 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
818 if (c_tolower (string[i]) != c_tolower (tail[j]))
822 /* If the tail was exhausted, the match was succesful. */
829 /* Checks whether string S matches each element of ACCEPTS. A list
830 element are matched either with fnmatch() or match_tail(),
831 according to whether the element contains wildcards or not.
833 If the BACKWARD is false, don't do backward comparison -- just compare
836 in_acclist (const char *const *accepts, const char *s, bool backward)
838 for (; *accepts; accepts++)
840 if (has_wildcards_p (*accepts))
842 int res = opt.ignore_case
843 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
844 /* fnmatch returns 0 if the pattern *does* match the string. */
852 if (match_tail (s, *accepts, opt.ignore_case))
857 int cmp = opt.ignore_case
858 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
867 /* Return the location of STR's suffix (file extension). Examples:
868 suffix ("foo.bar") -> "bar"
869 suffix ("foo.bar.baz") -> "baz"
870 suffix ("/foo/bar") -> NULL
871 suffix ("/foo.bar/baz") -> NULL */
873 suffix (const char *str)
877 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
881 return (char *)str + i;
886 /* Return true if S contains globbing wildcards (`*', `?', `[' or
890 has_wildcards_p (const char *s)
893 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
898 /* Return true if FNAME ends with a typical HTML suffix. The
899 following (case-insensitive) suffixes are presumed to be HTML
904 ?html (`?' matches one character)
906 #### CAVEAT. This is not necessarily a good indication that FNAME
907 refers to a file that contains HTML! */
909 has_html_suffix_p (const char *fname)
913 if ((suf = suffix (fname)) == NULL)
915 if (!strcasecmp (suf, "html"))
917 if (!strcasecmp (suf, "htm"))
919 if (suf[0] && !strcasecmp (suf + 1, "html"))
924 /* Read a line from FP and return the pointer to freshly allocated
925 storage. The storage space is obtained through malloc() and should
926 be freed with free() when it is no longer needed.
928 The length of the line is not limited, except by available memory.
929 The newline character at the end of line is retained. The line is
930 terminated with a zero character.
932 After end-of-file is encountered without anything being read, NULL
933 is returned. NULL is also returned on error. To distinguish
934 between these two cases, use the stdio function ferror(). */
937 read_whole_line (FILE *fp)
941 char *line = xmalloc (bufsize);
943 while (fgets (line + length, bufsize - length, fp))
945 length += strlen (line + length);
947 /* Possible for example when reading from a binary file where
948 a line begins with \0. */
951 if (line[length - 1] == '\n')
954 /* fgets() guarantees to read the whole line, or to use up the
955 space we've given it. We can double the buffer
958 line = xrealloc (line, bufsize);
960 if (length == 0 || ferror (fp))
965 if (length + 1 < bufsize)
966 /* Relieve the memory from our exponential greediness. We say
967 `length + 1' because the terminating \0 is not included in
968 LENGTH. We don't need to zero-terminate the string ourselves,
969 though, because fgets() does that. */
970 line = xrealloc (line, length + 1);
974 /* Read FILE into memory. A pointer to `struct file_memory' are
975 returned; use struct element `content' to access file contents, and
976 the element `length' to know the file length. `content' is *not*
977 zero-terminated, and you should *not* read or write beyond the [0,
978 length) range of characters.
980 After you are done with the file contents, call read_file_free to
983 Depending on the operating system and the type of file that is
984 being read, read_file() either mmap's the file into memory, or
985 reads the file into the core using read().
987 If file is named "-", fileno(stdin) is used for reading instead.
988 If you want to read from a real file named "-", use "./-" instead. */
991 read_file (const char *file)
994 struct file_memory *fm;
996 bool inhibit_close = false;
998 /* Some magic in the finest tradition of Perl and its kin: if FILE
999 is "-", just use stdin. */
1002 fd = fileno (stdin);
1003 inhibit_close = true;
1004 /* Note that we don't inhibit mmap() in this case. If stdin is
1005 redirected from a regular file, mmap() will still work. */
1008 fd = open (file, O_RDONLY);
1011 fm = xnew (struct file_memory);
1016 if (fstat (fd, &buf) < 0)
1018 fm->length = buf.st_size;
1019 /* NOTE: As far as I know, the callers of this function never
1020 modify the file text. Relying on this would enable us to
1021 specify PROT_READ and MAP_SHARED for a marginal gain in
1022 efficiency, but at some cost to generality. */
1023 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1024 MAP_PRIVATE, fd, 0);
1025 if (fm->content == (char *)MAP_FAILED)
1035 /* The most common reason why mmap() fails is that FD does not point
1036 to a plain file. However, it's also possible that mmap() doesn't
1037 work for a particular type of file. Therefore, whenever mmap()
1038 fails, we just fall back to the regular method. */
1039 #endif /* HAVE_MMAP */
1042 size = 512; /* number of bytes fm->contents can
1043 hold at any given time. */
1044 fm->content = xmalloc (size);
1048 if (fm->length > size / 2)
1050 /* #### I'm not sure whether the whole exponential-growth
1051 thing makes sense with kernel read. On Linux at least,
1052 read() refuses to read more than 4K from a file at a
1053 single chunk anyway. But other Unixes might optimize it
1054 better, and it doesn't *hurt* anything, so I'm leaving
1057 /* Normally, we grow SIZE exponentially to make the number
1058 of calls to read() and realloc() logarithmic in relation
1059 to file size. However, read() can read an amount of data
1060 smaller than requested, and it would be unreasonable to
1061 double SIZE every time *something* was read. Therefore,
1062 we double SIZE only when the length exceeds half of the
1063 entire allocated size. */
1065 fm->content = xrealloc (fm->content, size);
1067 nread = read (fd, fm->content + fm->length, size - fm->length);
1069 /* Successful read. */
1070 fm->length += nread;
1080 if (size > fm->length && fm->length != 0)
1081 /* Due to exponential growth of fm->content, the allocated region
1082 might be much larger than what is actually needed. */
1083 fm->content = xrealloc (fm->content, fm->length);
1090 xfree (fm->content);
1095 /* Release the resources held by FM. Specifically, this calls
1096 munmap() or xfree() on fm->content, depending whether mmap or
1097 malloc/read were used to read in the file. It also frees the
1098 memory needed to hold the FM structure itself. */
1101 read_file_free (struct file_memory *fm)
1106 munmap (fm->content, fm->length);
1111 xfree (fm->content);
1116 /* Free the pointers in a NULL-terminated vector of pointers, then
1117 free the pointer itself. */
1119 free_vec (char **vec)
1130 /* Append vector V2 to vector V1. The function frees V2 and
1131 reallocates V1 (thus you may not use the contents of neither
1132 pointer after the call). If V1 is NULL, V2 is returned. */
1134 merge_vecs (char **v1, char **v2)
1144 /* To avoid j == 0 */
1149 for (i = 0; v1[i]; i++)
1152 for (j = 0; v2[j]; j++)
1154 /* Reallocate v1. */
1155 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1156 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1161 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1162 is allocated as needed. Return the new value of the vector. */
1165 vec_append (char **vec, const char *str)
1167 int cnt; /* count of vector elements, including
1168 the one we're about to append */
1171 for (cnt = 0; vec[cnt]; cnt++)
1177 /* Reallocate the array to fit the new element and the NULL. */
1178 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1179 /* Append a copy of STR to the vector. */
1180 vec[cnt - 1] = xstrdup (str);
1185 /* Sometimes it's useful to create "sets" of strings, i.e. special
1186 hash tables where you want to store strings as keys and merely
1187 query for their existence. Here is a set of utility routines that
1188 makes that transparent. */
1191 string_set_add (struct hash_table *ht, const char *s)
1193 /* First check whether the set element already exists. If it does,
1194 do nothing so that we don't have to free() the old element and
1195 then strdup() a new one. */
1196 if (hash_table_contains (ht, s))
1199 /* We use "1" as value. It provides us a useful and clear arbitrary
1200 value, and it consumes no memory -- the pointers to the same
1201 string "1" will be shared by all the key-value pairs in all `set'
1203 hash_table_put (ht, xstrdup (s), "1");
1206 /* Synonym for hash_table_contains... */
1209 string_set_contains (struct hash_table *ht, const char *s)
1211 return hash_table_contains (ht, s);
1214 /* Convert the specified string set to array. ARRAY should be large
1215 enough to hold hash_table_count(ht) char pointers. */
1217 void string_set_to_array (struct hash_table *ht, char **array)
1219 hash_table_iterator iter;
1220 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1221 *array++ = iter.key;
1224 /* Free the string set. This frees both the storage allocated for
1225 keys and the actual hash table. (hash_table_destroy would only
1226 destroy the hash table.) */
1229 string_set_free (struct hash_table *ht)
1231 hash_table_iterator iter;
1232 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1234 hash_table_destroy (ht);
1237 /* Utility function: simply call xfree() on all keys and values of HT. */
1240 free_keys_and_values (struct hash_table *ht)
1242 hash_table_iterator iter;
1243 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1250 /* Get digit grouping data for thousand separors by calling
1251 localeconv(). The data includes separator string and grouping info
1252 and is cached after the first call to the function.
1254 In locales that don't set a thousand separator (such as the "C"
1255 locale), this forces it to be ",". We are now only showing
1256 thousand separators in one place, so this shouldn't be a problem in
1260 get_grouping_data (const char **sep, const char **grouping)
1262 static const char *cached_sep;
1263 static const char *cached_grouping;
1264 static bool initialized;
1267 /* Get the grouping info from the locale. */
1268 struct lconv *lconv = localeconv ();
1269 cached_sep = lconv->thousands_sep;
1270 cached_grouping = lconv->grouping;
1271 #if ! USE_NLS_PROGRESS_BAR
1272 /* We can't count column widths, so ensure that the separator
1273 * is single-byte only (let check below determine what byte). */
1274 if (strlen(cached_sep) > 1)
1279 /* Many locales (such as "C" or "hr_HR") don't specify
1280 grouping, which we still want to use it for legibility.
1281 In those locales set the sep char to ',', unless that
1282 character is used for decimal point, in which case set it
1284 if (*lconv->decimal_point != ',')
1288 cached_grouping = "\x03";
1293 *grouping = cached_grouping;
1296 /* Return a printed representation of N with thousand separators.
1297 This should respect locale settings, with the exception of the "C"
1298 locale which mandates no separator, but we use one anyway.
1300 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1301 the separators because it's too non-portable, and it's hard to test
1302 for this feature at configure time. Besides, it wouldn't display
1303 separators in the "C" locale, still used by many Unix users. */
1306 with_thousand_seps (wgint n)
1308 static char outbuf[48];
1309 char *p = outbuf + sizeof outbuf;
1311 /* Info received from locale */
1312 const char *grouping, *sep;
1315 /* State information */
1316 int i = 0, groupsize;
1317 const char *atgroup;
1319 bool negative = n < 0;
1321 /* Initialize grouping data. */
1322 get_grouping_data (&sep, &grouping);
1323 seplen = strlen (sep);
1325 groupsize = *atgroup++;
1327 /* This would overflow on WGINT_MIN, but printing negative numbers
1328 is not an important goal of this fuinction. */
1332 /* Write the number into the buffer, backwards, inserting the
1333 separators as necessary. */
1337 *--p = n % 10 + '0';
1341 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1342 if (++i == groupsize)
1347 memcpy (p -= seplen, sep, seplen);
1350 groupsize = *atgroup++;
1359 /* N, a byte quantity, is converted to a human-readable abberviated
1360 form a la sizes printed by `ls -lh'. The result is written to a
1361 static buffer, a pointer to which is returned.
1363 Unlike `with_thousand_seps', this approximates to the nearest unit.
1364 Quoting GNU libit: "Most people visually process strings of 3-4
1365 digits effectively, but longer strings of digits are more prone to
1366 misinterpretation. Hence, converting to an abbreviated form
1367 usually improves readability."
1369 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1370 original computer-related meaning of "powers of 1024". We don't
1371 use the "*bibyte" names invented in 1998, and seldom used in
1372 practice. Wikipedia's entry on "binary prefix" discusses this in
1376 human_readable (HR_NUMTYPE n)
1378 /* These suffixes are compatible with those of GNU `ls -lh'. */
1379 static char powers[] =
1381 'K', /* kilobyte, 2^10 bytes */
1382 'M', /* megabyte, 2^20 bytes */
1383 'G', /* gigabyte, 2^30 bytes */
1384 'T', /* terabyte, 2^40 bytes */
1385 'P', /* petabyte, 2^50 bytes */
1386 'E', /* exabyte, 2^60 bytes */
1391 /* If the quantity is smaller than 1K, just print it. */
1394 snprintf (buf, sizeof (buf), "%d", (int) n);
1398 /* Loop over powers, dividing N with 1024 in each iteration. This
1399 works unchanged for all sizes of wgint, while still avoiding
1400 non-portable `long double' arithmetic. */
1401 for (i = 0; i < countof (powers); i++)
1403 /* At each iteration N is greater than the *subsequent* power.
1404 That way N/1024.0 produces a decimal number in the units of
1406 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1408 double val = n / 1024.0;
1409 /* Print values smaller than 10 with one decimal digits, and
1410 others without any decimals. */
1411 snprintf (buf, sizeof (buf), "%.*f%c",
1412 val < 10 ? 1 : 0, val, powers[i]);
1417 return NULL; /* unreached */
1420 /* Count the digits in the provided number. Used to allocate space
1421 when printing numbers. */
1424 numdigit (wgint number)
1428 ++cnt; /* accomodate '-' */
1429 while ((number /= 10) != 0)
1434 #define PR(mask) *p++ = n / (mask) + '0'
1436 /* DIGITS_<D> is used to print a D-digit number and should be called
1437 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1438 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1439 Recursively this continues until DIGITS_1 is invoked. */
1441 #define DIGITS_1(mask) PR (mask)
1442 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1443 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1444 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1445 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1446 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1447 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1448 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1449 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1450 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1452 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1454 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1455 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1456 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1457 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1458 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1459 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1460 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1461 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1462 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1464 /* Shorthand for casting to wgint. */
1467 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1468 `sprintf(buffer, "%lld", (long long) number)', only typically much
1469 faster and portable to machines without long long.
1471 The speedup may make a difference in programs that frequently
1472 convert numbers to strings. Some implementations of sprintf,
1473 particularly the one in some versions of GNU libc, have been known
1474 to be quite slow when converting integers to strings.
1476 Return the pointer to the location where the terminating zero was
1477 printed. (Equivalent to calling buffer+strlen(buffer) after the
1480 BUFFER should be large enough to accept as many bytes as you expect
1481 the number to take up. On machines with 64-bit wgints the maximum
1482 needed size is 24 bytes. That includes the digits needed for the
1483 largest 64-bit number, the `-' sign in case it's negative, and the
1484 terminating '\0'. */
1487 number_to_string (char *buffer, wgint number)
1492 int last_digit_char = 0;
1494 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1495 /* We are running in a very strange environment. Leave the correct
1496 printing to sprintf. */
1497 p += sprintf (buf, "%j", (intmax_t) (n));
1498 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1504 /* n = -n would overflow because -n would evaluate to a
1505 wgint value larger than WGINT_MAX. Need to make n
1506 smaller and handle the last digit separately. */
1507 int last_digit = n % 10;
1508 /* The sign of n%10 is implementation-defined. */
1510 last_digit_char = '0' - last_digit;
1512 last_digit_char = '0' + last_digit;
1513 /* After n is made smaller, -n will not overflow. */
1521 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1522 way printing any N is fully open-coded without a loop or jump.
1523 (Also see description of DIGITS_*.) */
1525 if (n < 10) DIGITS_1 (1);
1526 else if (n < 100) DIGITS_2 (10);
1527 else if (n < 1000) DIGITS_3 (100);
1528 else if (n < 10000) DIGITS_4 (1000);
1529 else if (n < 100000) DIGITS_5 (10000);
1530 else if (n < 1000000) DIGITS_6 (100000);
1531 else if (n < 10000000) DIGITS_7 (1000000);
1532 else if (n < 100000000) DIGITS_8 (10000000);
1533 else if (n < 1000000000) DIGITS_9 (100000000);
1534 #if SIZEOF_WGINT == 4
1535 /* wgint is 32 bits wide: no number has more than 10 digits. */
1536 else DIGITS_10 (1000000000);
1538 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1539 Constants are constructed by compile-time multiplication to avoid
1540 dealing with different notations for 64-bit constants
1541 (nL/nLL/nI64, depending on the compiler and architecture). */
1542 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1543 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1544 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1545 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1546 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1547 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1548 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1549 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1550 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1551 else DIGITS_19 (1000000000*(W)1000000000);
1554 if (last_digit_char)
1555 *p++ = last_digit_char;
1558 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1565 #undef SPRINTF_WGINT
1588 /* Print NUMBER to a statically allocated string and return a pointer
1589 to the printed representation.
1591 This function is intended to be used in conjunction with printf.
1592 It is hard to portably print wgint values:
1593 a) you cannot use printf("%ld", number) because wgint can be long
1594 long on 32-bit machines with LFS.
1595 b) you cannot use printf("%lld", number) because NUMBER could be
1596 long on 32-bit machines without LFS, or on 64-bit machines,
1597 which do not require LFS. Also, Windows doesn't support %lld.
1598 c) you cannot use printf("%j", (int_max_t) number) because not all
1599 versions of printf support "%j", the most notable being the one
1601 d) you cannot #define WGINT_FMT to the appropriate format and use
1602 printf(WGINT_FMT, number) because that would break translations
1603 for user-visible messages, such as printf("Downloaded: %d
1606 What you should use instead is printf("%s", number_to_static_string
1609 CAVEAT: since the function returns pointers to static data, you
1610 must be careful to copy its result before calling it again.
1611 However, to make it more useful with printf, the function maintains
1612 an internal ring of static buffers to return. That way things like
1613 printf("%s %s", number_to_static_string (num1),
1614 number_to_static_string (num2)) work as expected. Three buffers
1615 are currently used, which means that "%s %s %s" will work, but "%s
1616 %s %s %s" won't. If you need to print more than three wgints,
1617 bump the RING_SIZE (or rethink your message.) */
1620 number_to_static_string (wgint number)
1622 static char ring[RING_SIZE][24];
1624 char *buf = ring[ringpos];
1625 number_to_string (buf, number);
1626 ringpos = (ringpos + 1) % RING_SIZE;
1630 /* Determine the width of the terminal we're running on. If that's
1631 not possible, return 0. */
1634 determine_screen_width (void)
1636 /* If there's a way to get the terminal size using POSIX
1637 tcgetattr(), somebody please tell me. */
1642 if (opt.lfilename != NULL)
1645 fd = fileno (stderr);
1646 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1647 return 0; /* most likely ENOTTY */
1650 #elif defined(WINDOWS)
1651 CONSOLE_SCREEN_BUFFER_INFO csbi;
1652 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1654 return csbi.dwSize.X;
1655 #else /* neither TIOCGWINSZ nor WINDOWS */
1657 #endif /* neither TIOCGWINSZ nor WINDOWS */
1660 /* Whether the rnd system (either rand or [dl]rand48) has been
1662 static int rnd_seeded;
1664 /* Return a random number between 0 and MAX-1, inclusive.
1666 If the system does not support lrand48 and MAX is greater than the
1667 value of RAND_MAX+1 on the system, the returned value will be in
1668 the range [0, RAND_MAX]. This may be fixed in a future release.
1669 The random number generator is seeded automatically the first time
1672 This uses lrand48 where available, rand elsewhere. DO NOT use it
1673 for cryptography. It is only meant to be used in situations where
1674 quality of the random numbers returned doesn't really matter. */
1677 random_number (int max)
1682 srand48 ((long) time (NULL) ^ (long) getpid ());
1685 return lrand48 () % max;
1686 #else /* not HAVE_DRAND48 */
1692 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1697 /* Like rand() % max, but uses the high-order bits for better
1698 randomness on architectures where rand() is implemented using a
1699 simple congruential generator. */
1701 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1702 return (int) bounded;
1704 #endif /* not HAVE_DRAND48 */
1707 /* Return a random uniformly distributed floating point number in the
1708 [0, 1) range. Uses drand48 where available, and a really lame
1709 kludge elsewhere. */
1717 srand48 ((long) time (NULL) ^ (long) getpid ());
1721 #else /* not HAVE_DRAND48 */
1722 return ( random_number (10000) / 10000.0
1723 + random_number (10000) / (10000.0 * 10000.0)
1724 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1725 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1726 #endif /* not HAVE_DRAND48 */
1729 /* Implementation of run_with_timeout, a generic timeout-forcing
1730 routine for systems with Unix-like signal handling. */
1732 #ifdef USE_SIGNAL_TIMEOUT
1733 # ifdef HAVE_SIGSETJMP
1734 # define SETJMP(env) sigsetjmp (env, 1)
1736 static sigjmp_buf run_with_timeout_env;
1739 abort_run_with_timeout (int sig)
1741 assert (sig == SIGALRM);
1742 siglongjmp (run_with_timeout_env, -1);
1744 # else /* not HAVE_SIGSETJMP */
1745 # define SETJMP(env) setjmp (env)
1747 static jmp_buf run_with_timeout_env;
1750 abort_run_with_timeout (int sig)
1752 assert (sig == SIGALRM);
1753 /* We don't have siglongjmp to preserve the set of blocked signals;
1754 if we longjumped out of the handler at this point, SIGALRM would
1755 remain blocked. We must unblock it manually. */
1756 int mask = siggetmask ();
1757 mask &= ~sigmask (SIGALRM);
1760 /* Now it's safe to longjump. */
1761 longjmp (run_with_timeout_env, -1);
1763 # endif /* not HAVE_SIGSETJMP */
1765 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1766 setitimer where available, alarm otherwise.
1768 TIMEOUT should be non-zero. If the timeout value is so small that
1769 it would be rounded to zero, it is rounded to the least legal value
1770 instead (1us for setitimer, 1s for alarm). That ensures that
1771 SIGALRM will be delivered in all cases. */
1774 alarm_set (double timeout)
1777 /* Use the modern itimer interface. */
1778 struct itimerval itv;
1780 itv.it_value.tv_sec = (long) timeout;
1781 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1782 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1783 /* Ensure that we wait for at least the minimum interval.
1784 Specifying zero would mean "wait forever". */
1785 itv.it_value.tv_usec = 1;
1786 setitimer (ITIMER_REAL, &itv, NULL);
1787 #else /* not ITIMER_REAL */
1788 /* Use the old alarm() interface. */
1789 int secs = (int) timeout;
1791 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1792 because alarm(0) means "never deliver the alarm", i.e. "wait
1793 forever", which is not what someone who specifies a 0.5s
1794 timeout would expect. */
1797 #endif /* not ITIMER_REAL */
1800 /* Cancel the alarm set with alarm_set. */
1806 struct itimerval disable;
1808 setitimer (ITIMER_REAL, &disable, NULL);
1809 #else /* not ITIMER_REAL */
1811 #endif /* not ITIMER_REAL */
1814 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1815 seconds. Returns true if the function was interrupted with a
1816 timeout, false otherwise.
1818 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1819 using setitimer() or alarm(). The timeout is enforced by
1820 longjumping out of the SIGALRM handler. This has several
1821 advantages compared to the traditional approach of relying on
1822 signals causing system calls to exit with EINTR:
1824 * The callback function is *forcibly* interrupted after the
1825 timeout expires, (almost) regardless of what it was doing and
1826 whether it was in a syscall. For example, a calculation that
1827 takes a long time is interrupted as reliably as an IO
1830 * It works with both SYSV and BSD signals because it doesn't
1831 depend on the default setting of SA_RESTART.
1833 * It doesn't require special handler setup beyond a simple call
1834 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1837 The only downside is that, if FUN allocates internal resources that
1838 are normally freed prior to exit from the functions, they will be
1839 lost in case of timeout. */
1842 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1852 signal (SIGALRM, abort_run_with_timeout);
1853 if (SETJMP (run_with_timeout_env) != 0)
1855 /* Longjumped out of FUN with a timeout. */
1856 signal (SIGALRM, SIG_DFL);
1859 alarm_set (timeout);
1862 /* Preserve errno in case alarm() or signal() modifies it. */
1863 saved_errno = errno;
1865 signal (SIGALRM, SIG_DFL);
1866 errno = saved_errno;
1871 #else /* not USE_SIGNAL_TIMEOUT */
1874 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1875 define it under Windows, because Windows has its own version of
1876 run_with_timeout that uses threads. */
1879 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1884 #endif /* not WINDOWS */
1885 #endif /* not USE_SIGNAL_TIMEOUT */
1889 /* Sleep the specified amount of seconds. On machines without
1890 nanosleep(), this may sleep shorter if interrupted by signals. */
1893 xsleep (double seconds)
1895 #ifdef HAVE_NANOSLEEP
1896 /* nanosleep is the preferred interface because it offers high
1897 accuracy and, more importantly, because it allows us to reliably
1898 restart receiving a signal such as SIGWINCH. (There was an
1899 actual Debian bug report about --limit-rate malfunctioning while
1900 the terminal was being resized.) */
1901 struct timespec sleep, remaining;
1902 sleep.tv_sec = (long) seconds;
1903 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1904 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1905 /* If nanosleep has been interrupted by a signal, adjust the
1906 sleeping period and return to sleep. */
1908 #elif defined(HAVE_USLEEP)
1909 /* If usleep is available, use it in preference to select. */
1912 /* On some systems, usleep cannot handle values larger than
1913 1,000,000. If the period is larger than that, use sleep
1914 first, then add usleep for subsecond accuracy. */
1916 seconds -= (long) seconds;
1918 usleep (seconds * 1000000);
1919 #else /* fall back select */
1920 /* Note that, although Windows supports select, it can't be used to
1921 implement sleeping because Winsock's select doesn't implement
1922 timeout when it is passed NULL pointers for all fd sets. (But it
1923 does under Cygwin, which implements Unix-compatible select.) */
1924 struct timeval sleep;
1925 sleep.tv_sec = (long) seconds;
1926 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1927 select (0, NULL, NULL, NULL, &sleep);
1928 /* If select returns -1 and errno is EINTR, it means we were
1929 interrupted by a signal. But without knowing how long we've
1930 actually slept, we can't return to sleep. Using gettimeofday to
1931 track sleeps is slow and unreliable due to clock skew. */
1935 #endif /* not WINDOWS */
1937 /* Encode the octets in DATA of length LENGTH to base64 format,
1938 storing the result to DEST. The output will be zero-terminated,
1939 and must point to a writable buffer of at least
1940 1+BASE64_LENGTH(length) bytes. The function returns the length of
1941 the resulting base64 data, not counting the terminating zero.
1943 This implementation does not emit newlines after 76 characters of
1947 base64_encode (const void *data, int length, char *dest)
1949 /* Conversion table. */
1950 static const char tbl[64] = {
1951 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
1952 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
1953 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
1954 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
1956 /* Access bytes in DATA as unsigned char, otherwise the shifts below
1957 don't work for data with MSB set. */
1958 const unsigned char *s = data;
1959 /* Theoretical ANSI violation when length < 3. */
1960 const unsigned char *end = (const unsigned char *) data + length - 2;
1963 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1964 for (; s < end; s += 3)
1966 *p++ = tbl[s[0] >> 2];
1967 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1968 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1969 *p++ = tbl[s[2] & 0x3f];
1972 /* Pad the result if necessary... */
1976 *p++ = tbl[s[0] >> 2];
1977 *p++ = tbl[(s[0] & 3) << 4];
1982 *p++ = tbl[s[0] >> 2];
1983 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1984 *p++ = tbl[((s[1] & 0xf) << 2)];
1988 /* ...and zero-terminate it. */
1994 /* Store in C the next non-whitespace character from the string, or \0
1995 when end of string is reached. */
1996 #define NEXT_CHAR(c, p) do { \
1997 c = (unsigned char) *p++; \
1998 } while (c_isspace (c))
2000 #define IS_ASCII(c) (((c) & 0x80) == 0)
2002 /* Decode data from BASE64 (a null-terminated string) into memory
2003 pointed to by DEST. DEST is assumed to be large enough to
2004 accomodate the decoded data, which is guaranteed to be no more than
2007 Since DEST is assumed to contain binary data, it is not
2008 NUL-terminated. The function returns the length of the data
2009 written to TO. -1 is returned in case of error caused by malformed
2012 This function originates from Free Recode. */
2015 base64_decode (const char *base64, void *dest)
2017 /* Table of base64 values for first 128 characters. Note that this
2018 assumes ASCII (but so does Wget in other places). */
2019 static const signed char base64_char_to_value[128] =
2021 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
2022 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
2023 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2024 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2025 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2026 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2027 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2028 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2029 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2030 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2031 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2032 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2033 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2035 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2036 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2038 const char *p = base64;
2044 unsigned long value;
2046 /* Process first byte of a quadruplet. */
2050 if (c == '=' || !IS_BASE64 (c))
2051 return -1; /* illegal char while decoding base64 */
2052 value = BASE64_CHAR_TO_VALUE (c) << 18;
2054 /* Process second byte of a quadruplet. */
2057 return -1; /* premature EOF while decoding base64 */
2058 if (c == '=' || !IS_BASE64 (c))
2059 return -1; /* illegal char while decoding base64 */
2060 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2063 /* Process third byte of a quadruplet. */
2066 return -1; /* premature EOF while decoding base64 */
2068 return -1; /* illegal char while decoding base64 */
2074 return -1; /* premature EOF while decoding base64 */
2076 return -1; /* padding `=' expected but not found */
2080 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2081 *q++ = 0xff & value >> 8;
2083 /* Process fourth byte of a quadruplet. */
2086 return -1; /* premature EOF while decoding base64 */
2090 return -1; /* illegal char while decoding base64 */
2092 value |= BASE64_CHAR_TO_VALUE (c);
2093 *q++ = 0xff & value;
2096 #undef BASE64_CHAR_TO_VALUE
2098 return q - (char *) dest;
2104 /* Simple merge sort for use by stable_sort. Implementation courtesy
2105 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2108 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2109 int (*cmpfun) (const void *, const void *))
2111 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2115 size_t mid = (to + from) / 2;
2116 mergesort_internal (base, temp, size, from, mid, cmpfun);
2117 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2120 for (k = from; (i <= mid) && (j <= to); k++)
2121 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2122 memcpy (ELT (temp, k), ELT (base, i++), size);
2124 memcpy (ELT (temp, k), ELT (base, j++), size);
2126 memcpy (ELT (temp, k++), ELT (base, i++), size);
2128 memcpy (ELT (temp, k++), ELT (base, j++), size);
2129 for (k = from; k <= to; k++)
2130 memcpy (ELT (base, k), ELT (temp, k), size);
2135 /* Stable sort with interface exactly like standard library's qsort.
2136 Uses mergesort internally, allocating temporary storage with
2140 stable_sort (void *base, size_t nmemb, size_t size,
2141 int (*cmpfun) (const void *, const void *))
2145 void *temp = alloca (nmemb * size * sizeof (void *));
2146 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2150 /* Print a decimal number. If it is equal to or larger than ten, the
2151 number is rounded. Otherwise it is printed with one significant
2152 digit without trailing zeros and with no more than three fractional
2153 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2154 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2156 This is useful for displaying durations because it provides
2157 order-of-magnitude information without unnecessary clutter --
2158 long-running downloads are shown without the fractional part, and
2159 short ones still retain one significant digit. */
2162 print_decimal (double number)
2164 static char buf[32];
2165 double n = number >= 0 ? number : -number;
2168 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2169 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2170 snprintf (buf, sizeof buf, "%.0f", number);
2172 snprintf (buf, sizeof buf, "%.1f", number);
2173 else if (n >= 0.001)
2174 snprintf (buf, sizeof buf, "%.1g", number);
2175 else if (n >= 0.0005)
2176 /* round [0.0005, 0.001) to 0.001 */
2177 snprintf (buf, sizeof buf, "%.3f", number);
2179 /* print numbers close to 0 as 0, not 0.000 */
2196 { "/somedir", "/somedir", true },
2197 { "/somedir", "/somedir/d2", true },
2198 { "/somedir/d1", "/somedir", false },
2201 for (i = 0; i < countof(test_array); ++i)
2203 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2205 mu_assert ("test_subdir_p: wrong result",
2206 res == test_array[i].result);
2213 test_dir_matches_p()
2221 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2222 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2223 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2224 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2225 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2226 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2227 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2228 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2229 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2230 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2231 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2232 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2233 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2234 { { "/Tmp/has", NULL, NULL }, "/Tmp/has space", false },
2235 { { "/Tmp/has", NULL, NULL }, "/Tmp/has,comma", false },
2238 for (i = 0; i < countof(test_array); ++i)
2240 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2242 mu_assert ("test_dir_matches_p: wrong result",
2243 res == test_array[i].result);
2249 #endif /* TESTING */