1 /* Various utility functions.
2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
37 #ifdef HAVE_SYS_TIME_H
38 # include <sys/time.h>
44 # include <sys/mman.h>
47 # include <process.h> /* getpid() */
52 #ifdef HAVE_SYS_UTIME_H
53 # include <sys/utime.h>
61 /* For TIOCGWINSZ and friends: */
62 #ifdef HAVE_SYS_IOCTL_H
63 # include <sys/ioctl.h>
69 /* Needed for Unix version of run_with_timeout. */
73 #ifndef HAVE_SIGSETJMP
74 /* If sigsetjmp is a macro, configure won't pick it up. */
76 # define HAVE_SIGSETJMP
80 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
81 # define USE_SIGNAL_TIMEOUT
91 /* Utility function: like xstrdup(), but also lowercases S. */
94 xstrdup_lower (const char *s)
96 char *copy = xstrdup (s);
103 /* Copy the string formed by two pointers (one on the beginning, other
104 on the char after the last char) to a new, malloc-ed location.
107 strdupdelim (const char *beg, const char *end)
109 char *res = xmalloc (end - beg + 1);
110 memcpy (res, beg, end - beg);
111 res[end - beg] = '\0';
115 /* Parse a string containing comma-separated elements, and return a
116 vector of char pointers with the elements. Spaces following the
117 commas are ignored. */
119 sepstring (const char *s)
133 res = xrealloc (res, (i + 2) * sizeof (char *));
134 res[i] = strdupdelim (p, s);
137 /* Skip the blanks following the ','. */
138 while (c_isspace (*s))
145 res = xrealloc (res, (i + 2) * sizeof (char *));
146 res[i] = strdupdelim (p, s);
151 /* Like sprintf, but prints into a string of sufficient size freshly
152 allocated with malloc, which is returned. If unable to print due
153 to invalid format, returns NULL. Inability to allocate needed
154 memory results in abort, as with xmalloc. This is in spirit
155 similar to the GNU/BSD extension asprintf, but somewhat easier to
158 Internally the function either calls vasprintf or loops around
159 vsnprintf until the correct size is found. Since Wget also ships a
160 fallback implementation of vsnprintf, this should be portable. */
162 /* Constant is using for limits memory allocation for text buffer.
163 Applicable in situation when: vasprintf is not available in the system
164 and vsnprintf return -1 when long line is truncated (in old versions of
165 glibc and in other system where C99 doesn`t support) */
167 #define FMT_MAX_LENGTH 1048576
170 aprintf (const char *fmt, ...)
172 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
177 va_start (args, fmt);
178 ret = vasprintf (&str, fmt, args);
180 if (ret < 0 && errno == ENOMEM)
181 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency
182 with xmalloc/xrealloc */
186 #else /* not HAVE_VASPRINTF */
188 /* vasprintf is unavailable. snprintf into a small buffer and
189 resize it as necessary. */
191 char *str = xmalloc (size);
193 /* #### This code will infloop and eventually abort in xrealloc if
194 passed a FMT that causes snprintf to consistently return -1. */
201 va_start (args, fmt);
202 n = vsnprintf (str, size, fmt, args);
205 /* If the printing worked, return the string. */
206 if (n > -1 && n < size)
209 /* Else try again with a larger buffer. */
210 if (n > -1) /* C99 */
211 size = n + 1; /* precisely what is needed */
212 else if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
213 { /* maybe we have some wrong
215 logprintf (LOG_ALWAYS,
216 _("%s: aprintf: text buffer is too big (%ld bytes), "
218 exec_name, size); /* printout a log message */
219 abort (); /* and abort... */
223 /* else, we continue to grow our
224 * buffer: Twice the old size. */
227 str = xrealloc (str, size);
229 #endif /* not HAVE_VASPRINTF */
232 /* Concatenate the NULL-terminated list of string arguments into
233 freshly allocated space. */
236 concat_strings (const char *str0, ...)
239 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
242 const char *next_str;
243 int total_length = 0;
246 /* Calculate the length of and allocate the resulting string. */
249 va_start (args, str0);
250 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
252 int len = strlen (next_str);
253 if (argcount < countof (saved_lengths))
254 saved_lengths[argcount++] = len;
258 p = ret = xmalloc (total_length + 1);
260 /* Copy the strings into the allocated space. */
263 va_start (args, str0);
264 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
267 if (argcount < countof (saved_lengths))
268 len = saved_lengths[argcount++];
270 len = strlen (next_str);
271 memcpy (p, next_str, len);
280 /* Format the provided time according to the specified format. The
281 format is a string with format elements supported by strftime. */
284 fmttime (time_t t, const char *fmt)
286 static char output[32];
287 struct tm *tm = localtime(&t);
290 if (!strftime(output, sizeof(output), fmt, tm))
295 /* Return pointer to a static char[] buffer in which zero-terminated
296 string-representation of TM (in form hh:mm:ss) is printed.
298 If TM is NULL, the current time will be used. */
303 return fmttime(t, "%H:%M:%S");
306 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
309 datetime_str (time_t t)
311 return fmttime(t, "%Y-%m-%d %H:%M:%S");
314 /* The Windows versions of the following two functions are defined in
315 mswindows.c. On MSDOS this function should never be called. */
317 #if !defined(WINDOWS) && !defined(MSDOS)
319 fork_to_background (void)
322 /* Whether we arrange our own version of opt.lfilename here. */
323 bool logfile_changed = false;
325 if (!opt.lfilename && (!opt.quiet || opt.server_response))
327 /* We must create the file immediately to avoid either a race
328 condition (which arises from using unique_name and failing to
329 use fopen_excl) or lying to the user about the log file name
330 (which arises from using unique_name, printing the name, and
331 using fopen_excl later on.) */
332 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
335 logfile_changed = true;
348 /* parent, no error */
349 printf (_("Continuing in background, pid %d.\n"), (int) pid);
351 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
352 exit (0); /* #### should we use _exit()? */
355 /* child: give up the privileges and keep running. */
357 freopen ("/dev/null", "r", stdin);
358 freopen ("/dev/null", "w", stdout);
359 freopen ("/dev/null", "w", stderr);
361 #endif /* !WINDOWS && !MSDOS */
363 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
364 specified with TM. The atime ("access time") is set to the current
368 touch (const char *file, time_t tm)
370 #ifdef HAVE_STRUCT_UTIMBUF
371 struct utimbuf times;
379 times.actime = time (NULL);
380 if (utime (file, ×) == -1)
381 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
384 /* Checks if FILE is a symbolic link, and removes it if it is. Does
385 nothing under MS-Windows. */
387 remove_link (const char *file)
392 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
394 DEBUGP (("Unlinking %s (symlink).\n", file));
397 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
398 file, strerror (errno));
403 /* Does FILENAME exist? This is quite a lousy implementation, since
404 it supplies no error codes -- only a yes-or-no answer. Thus it
405 will return that a file does not exist if, e.g., the directory is
406 unreadable. I don't mind it too much currently, though. The
407 proper way should, of course, be to have a third, error state,
408 other than true/false, but that would introduce uncalled-for
409 additional complexity to the callers. */
411 file_exists_p (const char *filename)
414 return access (filename, F_OK) >= 0;
417 return stat (filename, &buf) >= 0;
421 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
422 Returns 0 on error. */
424 file_non_directory_p (const char *path)
427 /* Use lstat() rather than stat() so that symbolic links pointing to
428 directories can be identified correctly. */
429 if (lstat (path, &buf) != 0)
431 return S_ISDIR (buf.st_mode) ? false : true;
434 /* Return the size of file named by FILENAME, or -1 if it cannot be
435 opened or seeked into. */
437 file_size (const char *filename)
439 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
441 /* We use fseek rather than stat to determine the file size because
442 that way we can also verify that the file is readable without
443 explicitly checking for permissions. Inspired by the POST patch
445 FILE *fp = fopen (filename, "rb");
448 fseeko (fp, 0, SEEK_END);
454 if (stat (filename, &st) < 0)
460 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
461 doesn't exist is found. Return a freshly allocated copy of the
465 unique_name_1 (const char *prefix)
468 int plen = strlen (prefix);
469 char *template = (char *)alloca (plen + 1 + 24);
470 char *template_tail = template + plen;
472 memcpy (template, prefix, plen);
473 *template_tail++ = '.';
476 number_to_string (template_tail, count++);
477 while (file_exists_p (template));
479 return xstrdup (template);
482 /* Return a unique file name, based on FILE.
484 More precisely, if FILE doesn't exist, it is returned unmodified.
485 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
486 file name that doesn't exist is returned.
488 The resulting file is not created, only verified that it didn't
489 exist at the point in time when the function was called.
490 Therefore, where security matters, don't rely that the file created
491 by this function exists until you open it with O_EXCL or
494 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
495 string. Otherwise, it may return FILE if the file doesn't exist
496 (and therefore doesn't need changing). */
499 unique_name (const char *file, bool allow_passthrough)
501 /* If the FILE itself doesn't exist, return it without
503 if (!file_exists_p (file))
504 return allow_passthrough ? (char *)file : xstrdup (file);
506 /* Otherwise, find a numeric suffix that results in unused file name
508 return unique_name_1 (file);
511 /* Create a file based on NAME, except without overwriting an existing
512 file with that name. Providing O_EXCL is correctly implemented,
513 this function does not have the race condition associated with
514 opening the file returned by unique_name. */
517 unique_create (const char *name, bool binary, char **opened_name)
519 /* unique file name, based on NAME */
520 char *uname = unique_name (name, false);
522 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
525 uname = unique_name (name, false);
527 if (opened_name && fp != NULL)
530 *opened_name = uname;
542 /* Open the file for writing, with the addition that the file is
543 opened "exclusively". This means that, if the file already exists,
544 this function will *fail* and errno will be set to EEXIST. If
545 BINARY is set, the file will be opened in binary mode, equivalent
548 If opening the file fails for any reason, including the file having
549 previously existed, this function returns NULL and sets errno
553 fopen_excl (const char *fname, bool binary)
557 int flags = O_WRONLY | O_CREAT | O_EXCL;
562 fd = open (fname, flags, 0666);
565 return fdopen (fd, binary ? "wb" : "w");
566 #else /* not O_EXCL */
567 /* Manually check whether the file exists. This is prone to race
568 conditions, but systems without O_EXCL haven't deserved
570 if (file_exists_p (fname))
575 return fopen (fname, binary ? "wb" : "w");
576 #endif /* not O_EXCL */
579 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
580 are missing, create them first. In case any mkdir() call fails,
581 return its error status. Returns 0 on successful completion.
583 The behaviour of this function should be identical to the behaviour
584 of `mkdir -p' on systems where mkdir supports the `-p' option. */
586 make_directory (const char *directory)
588 int i, ret, quit = 0;
591 /* Make a copy of dir, to be able to write to it. Otherwise, the
592 function is unsafe if called with a read-only char *argument. */
593 STRDUP_ALLOCA (dir, directory);
595 /* If the first character of dir is '/', skip it (and thus enable
596 creation of absolute-pathname directories. */
597 for (i = (*dir == '/'); 1; ++i)
599 for (; dir[i] && dir[i] != '/'; i++)
604 /* Check whether the directory already exists. Allow creation of
605 of intermediate directories to fail, as the initial path components
606 are not necessarily directories! */
607 if (!file_exists_p (dir))
608 ret = mkdir (dir, 0777);
619 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
620 should be a file name.
622 file_merge("/foo/bar", "baz") => "/foo/baz"
623 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
624 file_merge("foo", "bar") => "bar"
626 In other words, it's a simpler and gentler version of uri_merge. */
629 file_merge (const char *base, const char *file)
632 const char *cut = (const char *)strrchr (base, '/');
635 return xstrdup (file);
637 result = xmalloc (cut - base + 1 + strlen (file) + 1);
638 memcpy (result, base, cut - base);
639 result[cut - base] = '/';
640 strcpy (result + (cut - base) + 1, file);
645 /* Like fnmatch, but performs a case-insensitive match. */
648 fnmatch_nocase (const char *pattern, const char *string, int flags)
651 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
652 also present on *BSD platforms, and possibly elsewhere. */
653 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
655 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
656 char *patcopy = (char *) alloca (strlen (pattern) + 1);
657 char *strcopy = (char *) alloca (strlen (string) + 1);
659 for (p = patcopy; *pattern; pattern++, p++)
660 *p = c_tolower (*pattern);
662 for (p = strcopy; *string; string++, p++)
663 *p = c_tolower (*string);
665 return fnmatch (patcopy, strcopy, flags);
669 static bool in_acclist (const char *const *, const char *, bool);
671 /* Determine whether a file is acceptable to be followed, according to
672 lists of patterns to accept/reject. */
674 acceptable (const char *s)
678 while (l && s[l] != '/')
685 return (in_acclist ((const char *const *)opt.accepts, s, true)
686 && !in_acclist ((const char *const *)opt.rejects, s, true));
688 return in_acclist ((const char *const *)opt.accepts, s, true);
690 else if (opt.rejects)
691 return !in_acclist ((const char *const *)opt.rejects, s, true);
695 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
696 will return true if and only if D2 begins with `/something/' or is exactly
699 subdir_p (const char *d1, const char *d2)
703 if (!opt.ignore_case)
704 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
707 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
710 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
713 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
714 first element that matches DIR, through wildcards or front comparison (as
717 dir_matches_p (char **dirlist, const char *dir)
720 int (*matcher) (const char *, const char *, int)
721 = opt.ignore_case ? fnmatch_nocase : fnmatch;
723 for (x = dirlist; *x; x++)
725 /* Remove leading '/' */
726 char *p = *x + (**x == '/');
727 if (has_wildcards_p (p))
729 if (matcher (p, dir, FNM_PATHNAME) == 0)
734 if (subdir_p (p, dir))
739 return *x ? true : false;
742 /* Returns whether DIRECTORY is acceptable for download, wrt the
743 include/exclude lists.
745 The leading `/' is ignored in paths; relative and absolute paths
746 may be freely intermixed. */
749 accdir (const char *directory)
751 /* Remove starting '/'. */
752 if (*directory == '/')
756 if (!dir_matches_p (opt.includes, directory))
761 if (dir_matches_p (opt.excludes, directory))
767 /* Return true if STRING ends with TAIL. For instance:
769 match_tail ("abc", "bc", false) -> 1
770 match_tail ("abc", "ab", false) -> 0
771 match_tail ("abc", "abc", false) -> 1
773 If FOLD_CASE is true, the comparison will be case-insensitive. */
776 match_tail (const char *string, const char *tail, bool fold_case)
780 /* We want this to be fast, so we code two loops, one with
781 case-folding, one without. */
785 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
786 if (string[i] != tail[j])
791 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
792 if (c_tolower (string[i]) != c_tolower (tail[j]))
796 /* If the tail was exhausted, the match was succesful. */
803 /* Checks whether string S matches each element of ACCEPTS. A list
804 element are matched either with fnmatch() or match_tail(),
805 according to whether the element contains wildcards or not.
807 If the BACKWARD is false, don't do backward comparison -- just compare
810 in_acclist (const char *const *accepts, const char *s, bool backward)
812 for (; *accepts; accepts++)
814 if (has_wildcards_p (*accepts))
816 int res = opt.ignore_case
817 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
818 /* fnmatch returns 0 if the pattern *does* match the string. */
826 if (match_tail (s, *accepts, opt.ignore_case))
831 int cmp = opt.ignore_case
832 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
841 /* Return the location of STR's suffix (file extension). Examples:
842 suffix ("foo.bar") -> "bar"
843 suffix ("foo.bar.baz") -> "baz"
844 suffix ("/foo/bar") -> NULL
845 suffix ("/foo.bar/baz") -> NULL */
847 suffix (const char *str)
851 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
855 return (char *)str + i;
860 /* Return true if S contains globbing wildcards (`*', `?', `[' or
864 has_wildcards_p (const char *s)
867 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
872 /* Return true if FNAME ends with a typical HTML suffix. The
873 following (case-insensitive) suffixes are presumed to be HTML
878 ?html (`?' matches one character)
880 #### CAVEAT. This is not necessarily a good indication that FNAME
881 refers to a file that contains HTML! */
883 has_html_suffix_p (const char *fname)
887 if ((suf = suffix (fname)) == NULL)
889 if (!strcasecmp (suf, "html"))
891 if (!strcasecmp (suf, "htm"))
893 if (suf[0] && !strcasecmp (suf + 1, "html"))
898 /* Read a line from FP and return the pointer to freshly allocated
899 storage. The storage space is obtained through malloc() and should
900 be freed with free() when it is no longer needed.
902 The length of the line is not limited, except by available memory.
903 The newline character at the end of line is retained. The line is
904 terminated with a zero character.
906 After end-of-file is encountered without anything being read, NULL
907 is returned. NULL is also returned on error. To distinguish
908 between these two cases, use the stdio function ferror(). */
911 read_whole_line (FILE *fp)
915 char *line = xmalloc (bufsize);
917 while (fgets (line + length, bufsize - length, fp))
919 length += strlen (line + length);
921 /* Possible for example when reading from a binary file where
922 a line begins with \0. */
925 if (line[length - 1] == '\n')
928 /* fgets() guarantees to read the whole line, or to use up the
929 space we've given it. We can double the buffer
932 line = xrealloc (line, bufsize);
934 if (length == 0 || ferror (fp))
939 if (length + 1 < bufsize)
940 /* Relieve the memory from our exponential greediness. We say
941 `length + 1' because the terminating \0 is not included in
942 LENGTH. We don't need to zero-terminate the string ourselves,
943 though, because fgets() does that. */
944 line = xrealloc (line, length + 1);
948 /* Read FILE into memory. A pointer to `struct file_memory' are
949 returned; use struct element `content' to access file contents, and
950 the element `length' to know the file length. `content' is *not*
951 zero-terminated, and you should *not* read or write beyond the [0,
952 length) range of characters.
954 After you are done with the file contents, call read_file_free to
957 Depending on the operating system and the type of file that is
958 being read, read_file() either mmap's the file into memory, or
959 reads the file into the core using read().
961 If file is named "-", fileno(stdin) is used for reading instead.
962 If you want to read from a real file named "-", use "./-" instead. */
965 read_file (const char *file)
968 struct file_memory *fm;
970 bool inhibit_close = false;
972 /* Some magic in the finest tradition of Perl and its kin: if FILE
973 is "-", just use stdin. */
977 inhibit_close = true;
978 /* Note that we don't inhibit mmap() in this case. If stdin is
979 redirected from a regular file, mmap() will still work. */
982 fd = open (file, O_RDONLY);
985 fm = xnew (struct file_memory);
990 if (fstat (fd, &buf) < 0)
992 fm->length = buf.st_size;
993 /* NOTE: As far as I know, the callers of this function never
994 modify the file text. Relying on this would enable us to
995 specify PROT_READ and MAP_SHARED for a marginal gain in
996 efficiency, but at some cost to generality. */
997 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
999 if (fm->content == (char *)MAP_FAILED)
1009 /* The most common reason why mmap() fails is that FD does not point
1010 to a plain file. However, it's also possible that mmap() doesn't
1011 work for a particular type of file. Therefore, whenever mmap()
1012 fails, we just fall back to the regular method. */
1013 #endif /* HAVE_MMAP */
1016 size = 512; /* number of bytes fm->contents can
1017 hold at any given time. */
1018 fm->content = xmalloc (size);
1022 if (fm->length > size / 2)
1024 /* #### I'm not sure whether the whole exponential-growth
1025 thing makes sense with kernel read. On Linux at least,
1026 read() refuses to read more than 4K from a file at a
1027 single chunk anyway. But other Unixes might optimize it
1028 better, and it doesn't *hurt* anything, so I'm leaving
1031 /* Normally, we grow SIZE exponentially to make the number
1032 of calls to read() and realloc() logarithmic in relation
1033 to file size. However, read() can read an amount of data
1034 smaller than requested, and it would be unreasonable to
1035 double SIZE every time *something* was read. Therefore,
1036 we double SIZE only when the length exceeds half of the
1037 entire allocated size. */
1039 fm->content = xrealloc (fm->content, size);
1041 nread = read (fd, fm->content + fm->length, size - fm->length);
1043 /* Successful read. */
1044 fm->length += nread;
1054 if (size > fm->length && fm->length != 0)
1055 /* Due to exponential growth of fm->content, the allocated region
1056 might be much larger than what is actually needed. */
1057 fm->content = xrealloc (fm->content, fm->length);
1064 xfree (fm->content);
1069 /* Release the resources held by FM. Specifically, this calls
1070 munmap() or xfree() on fm->content, depending whether mmap or
1071 malloc/read were used to read in the file. It also frees the
1072 memory needed to hold the FM structure itself. */
1075 read_file_free (struct file_memory *fm)
1080 munmap (fm->content, fm->length);
1085 xfree (fm->content);
1090 /* Free the pointers in a NULL-terminated vector of pointers, then
1091 free the pointer itself. */
1093 free_vec (char **vec)
1104 /* Append vector V2 to vector V1. The function frees V2 and
1105 reallocates V1 (thus you may not use the contents of neither
1106 pointer after the call). If V1 is NULL, V2 is returned. */
1108 merge_vecs (char **v1, char **v2)
1118 /* To avoid j == 0 */
1123 for (i = 0; v1[i]; i++)
1126 for (j = 0; v2[j]; j++)
1128 /* Reallocate v1. */
1129 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1130 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1135 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1136 is allocated as needed. Return the new value of the vector. */
1139 vec_append (char **vec, const char *str)
1141 int cnt; /* count of vector elements, including
1142 the one we're about to append */
1145 for (cnt = 0; vec[cnt]; cnt++)
1151 /* Reallocate the array to fit the new element and the NULL. */
1152 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1153 /* Append a copy of STR to the vector. */
1154 vec[cnt - 1] = xstrdup (str);
1159 /* Sometimes it's useful to create "sets" of strings, i.e. special
1160 hash tables where you want to store strings as keys and merely
1161 query for their existence. Here is a set of utility routines that
1162 makes that transparent. */
1165 string_set_add (struct hash_table *ht, const char *s)
1167 /* First check whether the set element already exists. If it does,
1168 do nothing so that we don't have to free() the old element and
1169 then strdup() a new one. */
1170 if (hash_table_contains (ht, s))
1173 /* We use "1" as value. It provides us a useful and clear arbitrary
1174 value, and it consumes no memory -- the pointers to the same
1175 string "1" will be shared by all the key-value pairs in all `set'
1177 hash_table_put (ht, xstrdup (s), "1");
1180 /* Synonym for hash_table_contains... */
1183 string_set_contains (struct hash_table *ht, const char *s)
1185 return hash_table_contains (ht, s);
1188 /* Convert the specified string set to array. ARRAY should be large
1189 enough to hold hash_table_count(ht) char pointers. */
1191 void string_set_to_array (struct hash_table *ht, char **array)
1193 hash_table_iterator iter;
1194 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1195 *array++ = iter.key;
1198 /* Free the string set. This frees both the storage allocated for
1199 keys and the actual hash table. (hash_table_destroy would only
1200 destroy the hash table.) */
1203 string_set_free (struct hash_table *ht)
1205 hash_table_iterator iter;
1206 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1208 hash_table_destroy (ht);
1211 /* Utility function: simply call xfree() on all keys and values of HT. */
1214 free_keys_and_values (struct hash_table *ht)
1216 hash_table_iterator iter;
1217 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1224 /* Get digit grouping data for thousand separors by calling
1225 localeconv(). The data includes separator string and grouping info
1226 and is cached after the first call to the function.
1228 In locales that don't set a thousand separator (such as the "C"
1229 locale), this forces it to be ",". We are now only showing
1230 thousand separators in one place, so this shouldn't be a problem in
1234 get_grouping_data (const char **sep, const char **grouping)
1236 static const char *cached_sep;
1237 static const char *cached_grouping;
1238 static bool initialized;
1241 /* Get the grouping info from the locale. */
1242 struct lconv *lconv = localeconv ();
1243 cached_sep = lconv->thousands_sep;
1244 cached_grouping = lconv->grouping;
1245 #if ! USE_NLS_PROGRESS_BAR
1246 /* We can't count column widths, so ensure that the separator
1247 * is single-byte only (let check below determine what byte). */
1248 if (strlen(cached_sep) > 1)
1253 /* Many locales (such as "C" or "hr_HR") don't specify
1254 grouping, which we still want to use it for legibility.
1255 In those locales set the sep char to ',', unless that
1256 character is used for decimal point, in which case set it
1258 if (*lconv->decimal_point != ',')
1262 cached_grouping = "\x03";
1267 *grouping = cached_grouping;
1270 /* Return a printed representation of N with thousand separators.
1271 This should respect locale settings, with the exception of the "C"
1272 locale which mandates no separator, but we use one anyway.
1274 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1275 the separators because it's too non-portable, and it's hard to test
1276 for this feature at configure time. Besides, it wouldn't display
1277 separators in the "C" locale, still used by many Unix users. */
1280 with_thousand_seps (wgint n)
1282 static char outbuf[48];
1283 char *p = outbuf + sizeof outbuf;
1285 /* Info received from locale */
1286 const char *grouping, *sep;
1289 /* State information */
1290 int i = 0, groupsize;
1291 const char *atgroup;
1293 bool negative = n < 0;
1295 /* Initialize grouping data. */
1296 get_grouping_data (&sep, &grouping);
1297 seplen = strlen (sep);
1299 groupsize = *atgroup++;
1301 /* This would overflow on WGINT_MIN, but printing negative numbers
1302 is not an important goal of this fuinction. */
1306 /* Write the number into the buffer, backwards, inserting the
1307 separators as necessary. */
1311 *--p = n % 10 + '0';
1315 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1316 if (++i == groupsize)
1321 memcpy (p -= seplen, sep, seplen);
1324 groupsize = *atgroup++;
1333 /* N, a byte quantity, is converted to a human-readable abberviated
1334 form a la sizes printed by `ls -lh'. The result is written to a
1335 static buffer, a pointer to which is returned.
1337 Unlike `with_thousand_seps', this approximates to the nearest unit.
1338 Quoting GNU libit: "Most people visually process strings of 3-4
1339 digits effectively, but longer strings of digits are more prone to
1340 misinterpretation. Hence, converting to an abbreviated form
1341 usually improves readability."
1343 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1344 original computer-related meaning of "powers of 1024". We don't
1345 use the "*bibyte" names invented in 1998, and seldom used in
1346 practice. Wikipedia's entry on "binary prefix" discusses this in
1350 human_readable (HR_NUMTYPE n)
1352 /* These suffixes are compatible with those of GNU `ls -lh'. */
1353 static char powers[] =
1355 'K', /* kilobyte, 2^10 bytes */
1356 'M', /* megabyte, 2^20 bytes */
1357 'G', /* gigabyte, 2^30 bytes */
1358 'T', /* terabyte, 2^40 bytes */
1359 'P', /* petabyte, 2^50 bytes */
1360 'E', /* exabyte, 2^60 bytes */
1365 /* If the quantity is smaller than 1K, just print it. */
1368 snprintf (buf, sizeof (buf), "%d", (int) n);
1372 /* Loop over powers, dividing N with 1024 in each iteration. This
1373 works unchanged for all sizes of wgint, while still avoiding
1374 non-portable `long double' arithmetic. */
1375 for (i = 0; i < countof (powers); i++)
1377 /* At each iteration N is greater than the *subsequent* power.
1378 That way N/1024.0 produces a decimal number in the units of
1380 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1382 double val = n / 1024.0;
1383 /* Print values smaller than 10 with one decimal digits, and
1384 others without any decimals. */
1385 snprintf (buf, sizeof (buf), "%.*f%c",
1386 val < 10 ? 1 : 0, val, powers[i]);
1391 return NULL; /* unreached */
1394 /* Count the digits in the provided number. Used to allocate space
1395 when printing numbers. */
1398 numdigit (wgint number)
1402 ++cnt; /* accomodate '-' */
1403 while ((number /= 10) != 0)
1408 #define PR(mask) *p++ = n / (mask) + '0'
1410 /* DIGITS_<D> is used to print a D-digit number and should be called
1411 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1412 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1413 Recursively this continues until DIGITS_1 is invoked. */
1415 #define DIGITS_1(mask) PR (mask)
1416 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1417 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1418 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1419 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1420 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1421 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1422 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1423 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1424 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1426 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1428 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1429 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1430 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1431 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1432 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1433 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1434 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1435 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1436 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1438 /* Shorthand for casting to wgint. */
1441 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1442 `sprintf(buffer, "%lld", (long long) number)', only typically much
1443 faster and portable to machines without long long.
1445 The speedup may make a difference in programs that frequently
1446 convert numbers to strings. Some implementations of sprintf,
1447 particularly the one in some versions of GNU libc, have been known
1448 to be quite slow when converting integers to strings.
1450 Return the pointer to the location where the terminating zero was
1451 printed. (Equivalent to calling buffer+strlen(buffer) after the
1454 BUFFER should be large enough to accept as many bytes as you expect
1455 the number to take up. On machines with 64-bit wgints the maximum
1456 needed size is 24 bytes. That includes the digits needed for the
1457 largest 64-bit number, the `-' sign in case it's negative, and the
1458 terminating '\0'. */
1461 number_to_string (char *buffer, wgint number)
1466 int last_digit_char = 0;
1468 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1469 /* We are running in a very strange environment. Leave the correct
1470 printing to sprintf. */
1471 p += sprintf (buf, "%j", (intmax_t) (n));
1472 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1478 /* n = -n would overflow because -n would evaluate to a
1479 wgint value larger than WGINT_MAX. Need to make n
1480 smaller and handle the last digit separately. */
1481 int last_digit = n % 10;
1482 /* The sign of n%10 is implementation-defined. */
1484 last_digit_char = '0' - last_digit;
1486 last_digit_char = '0' + last_digit;
1487 /* After n is made smaller, -n will not overflow. */
1495 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1496 way printing any N is fully open-coded without a loop or jump.
1497 (Also see description of DIGITS_*.) */
1499 if (n < 10) DIGITS_1 (1);
1500 else if (n < 100) DIGITS_2 (10);
1501 else if (n < 1000) DIGITS_3 (100);
1502 else if (n < 10000) DIGITS_4 (1000);
1503 else if (n < 100000) DIGITS_5 (10000);
1504 else if (n < 1000000) DIGITS_6 (100000);
1505 else if (n < 10000000) DIGITS_7 (1000000);
1506 else if (n < 100000000) DIGITS_8 (10000000);
1507 else if (n < 1000000000) DIGITS_9 (100000000);
1508 #if SIZEOF_WGINT == 4
1509 /* wgint is 32 bits wide: no number has more than 10 digits. */
1510 else DIGITS_10 (1000000000);
1512 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1513 Constants are constructed by compile-time multiplication to avoid
1514 dealing with different notations for 64-bit constants
1515 (nL/nLL/nI64, depending on the compiler and architecture). */
1516 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1517 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1518 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1519 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1520 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1521 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1522 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1523 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1524 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1525 else DIGITS_19 (1000000000*(W)1000000000);
1528 if (last_digit_char)
1529 *p++ = last_digit_char;
1532 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1539 #undef SPRINTF_WGINT
1562 /* Print NUMBER to a statically allocated string and return a pointer
1563 to the printed representation.
1565 This function is intended to be used in conjunction with printf.
1566 It is hard to portably print wgint values:
1567 a) you cannot use printf("%ld", number) because wgint can be long
1568 long on 32-bit machines with LFS.
1569 b) you cannot use printf("%lld", number) because NUMBER could be
1570 long on 32-bit machines without LFS, or on 64-bit machines,
1571 which do not require LFS. Also, Windows doesn't support %lld.
1572 c) you cannot use printf("%j", (int_max_t) number) because not all
1573 versions of printf support "%j", the most notable being the one
1575 d) you cannot #define WGINT_FMT to the appropriate format and use
1576 printf(WGINT_FMT, number) because that would break translations
1577 for user-visible messages, such as printf("Downloaded: %d
1580 What you should use instead is printf("%s", number_to_static_string
1583 CAVEAT: since the function returns pointers to static data, you
1584 must be careful to copy its result before calling it again.
1585 However, to make it more useful with printf, the function maintains
1586 an internal ring of static buffers to return. That way things like
1587 printf("%s %s", number_to_static_string (num1),
1588 number_to_static_string (num2)) work as expected. Three buffers
1589 are currently used, which means that "%s %s %s" will work, but "%s
1590 %s %s %s" won't. If you need to print more than three wgints,
1591 bump the RING_SIZE (or rethink your message.) */
1594 number_to_static_string (wgint number)
1596 static char ring[RING_SIZE][24];
1598 char *buf = ring[ringpos];
1599 number_to_string (buf, number);
1600 ringpos = (ringpos + 1) % RING_SIZE;
1604 /* Determine the width of the terminal we're running on. If that's
1605 not possible, return 0. */
1608 determine_screen_width (void)
1610 /* If there's a way to get the terminal size using POSIX
1611 tcgetattr(), somebody please tell me. */
1616 if (opt.lfilename != NULL)
1619 fd = fileno (stderr);
1620 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1621 return 0; /* most likely ENOTTY */
1624 #elif defined(WINDOWS)
1625 CONSOLE_SCREEN_BUFFER_INFO csbi;
1626 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1628 return csbi.dwSize.X;
1629 #else /* neither TIOCGWINSZ nor WINDOWS */
1631 #endif /* neither TIOCGWINSZ nor WINDOWS */
1634 /* Whether the rnd system (either rand or [dl]rand48) has been
1636 static int rnd_seeded;
1638 /* Return a random number between 0 and MAX-1, inclusive.
1640 If the system does not support lrand48 and MAX is greater than the
1641 value of RAND_MAX+1 on the system, the returned value will be in
1642 the range [0, RAND_MAX]. This may be fixed in a future release.
1643 The random number generator is seeded automatically the first time
1646 This uses lrand48 where available, rand elsewhere. DO NOT use it
1647 for cryptography. It is only meant to be used in situations where
1648 quality of the random numbers returned doesn't really matter. */
1651 random_number (int max)
1656 srand48 ((long) time (NULL) ^ (long) getpid ());
1659 return lrand48 () % max;
1660 #else /* not HAVE_DRAND48 */
1666 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1671 /* Like rand() % max, but uses the high-order bits for better
1672 randomness on architectures where rand() is implemented using a
1673 simple congruential generator. */
1675 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1676 return (int) bounded;
1678 #endif /* not HAVE_DRAND48 */
1681 /* Return a random uniformly distributed floating point number in the
1682 [0, 1) range. Uses drand48 where available, and a really lame
1683 kludge elsewhere. */
1691 srand48 ((long) time (NULL) ^ (long) getpid ());
1695 #else /* not HAVE_DRAND48 */
1696 return ( random_number (10000) / 10000.0
1697 + random_number (10000) / (10000.0 * 10000.0)
1698 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1699 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1700 #endif /* not HAVE_DRAND48 */
1703 /* Implementation of run_with_timeout, a generic timeout-forcing
1704 routine for systems with Unix-like signal handling. */
1706 #ifdef USE_SIGNAL_TIMEOUT
1707 # ifdef HAVE_SIGSETJMP
1708 # define SETJMP(env) sigsetjmp (env, 1)
1710 static sigjmp_buf run_with_timeout_env;
1713 abort_run_with_timeout (int sig)
1715 assert (sig == SIGALRM);
1716 siglongjmp (run_with_timeout_env, -1);
1718 # else /* not HAVE_SIGSETJMP */
1719 # define SETJMP(env) setjmp (env)
1721 static jmp_buf run_with_timeout_env;
1724 abort_run_with_timeout (int sig)
1726 assert (sig == SIGALRM);
1727 /* We don't have siglongjmp to preserve the set of blocked signals;
1728 if we longjumped out of the handler at this point, SIGALRM would
1729 remain blocked. We must unblock it manually. */
1730 int mask = siggetmask ();
1731 mask &= ~sigmask (SIGALRM);
1734 /* Now it's safe to longjump. */
1735 longjmp (run_with_timeout_env, -1);
1737 # endif /* not HAVE_SIGSETJMP */
1739 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1740 setitimer where available, alarm otherwise.
1742 TIMEOUT should be non-zero. If the timeout value is so small that
1743 it would be rounded to zero, it is rounded to the least legal value
1744 instead (1us for setitimer, 1s for alarm). That ensures that
1745 SIGALRM will be delivered in all cases. */
1748 alarm_set (double timeout)
1751 /* Use the modern itimer interface. */
1752 struct itimerval itv;
1754 itv.it_value.tv_sec = (long) timeout;
1755 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1756 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1757 /* Ensure that we wait for at least the minimum interval.
1758 Specifying zero would mean "wait forever". */
1759 itv.it_value.tv_usec = 1;
1760 setitimer (ITIMER_REAL, &itv, NULL);
1761 #else /* not ITIMER_REAL */
1762 /* Use the old alarm() interface. */
1763 int secs = (int) timeout;
1765 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1766 because alarm(0) means "never deliver the alarm", i.e. "wait
1767 forever", which is not what someone who specifies a 0.5s
1768 timeout would expect. */
1771 #endif /* not ITIMER_REAL */
1774 /* Cancel the alarm set with alarm_set. */
1780 struct itimerval disable;
1782 setitimer (ITIMER_REAL, &disable, NULL);
1783 #else /* not ITIMER_REAL */
1785 #endif /* not ITIMER_REAL */
1788 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1789 seconds. Returns true if the function was interrupted with a
1790 timeout, false otherwise.
1792 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1793 using setitimer() or alarm(). The timeout is enforced by
1794 longjumping out of the SIGALRM handler. This has several
1795 advantages compared to the traditional approach of relying on
1796 signals causing system calls to exit with EINTR:
1798 * The callback function is *forcibly* interrupted after the
1799 timeout expires, (almost) regardless of what it was doing and
1800 whether it was in a syscall. For example, a calculation that
1801 takes a long time is interrupted as reliably as an IO
1804 * It works with both SYSV and BSD signals because it doesn't
1805 depend on the default setting of SA_RESTART.
1807 * It doesn't require special handler setup beyond a simple call
1808 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1811 The only downside is that, if FUN allocates internal resources that
1812 are normally freed prior to exit from the functions, they will be
1813 lost in case of timeout. */
1816 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1826 signal (SIGALRM, abort_run_with_timeout);
1827 if (SETJMP (run_with_timeout_env) != 0)
1829 /* Longjumped out of FUN with a timeout. */
1830 signal (SIGALRM, SIG_DFL);
1833 alarm_set (timeout);
1836 /* Preserve errno in case alarm() or signal() modifies it. */
1837 saved_errno = errno;
1839 signal (SIGALRM, SIG_DFL);
1840 errno = saved_errno;
1845 #else /* not USE_SIGNAL_TIMEOUT */
1848 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1849 define it under Windows, because Windows has its own version of
1850 run_with_timeout that uses threads. */
1853 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1858 #endif /* not WINDOWS */
1859 #endif /* not USE_SIGNAL_TIMEOUT */
1863 /* Sleep the specified amount of seconds. On machines without
1864 nanosleep(), this may sleep shorter if interrupted by signals. */
1867 xsleep (double seconds)
1869 #ifdef HAVE_NANOSLEEP
1870 /* nanosleep is the preferred interface because it offers high
1871 accuracy and, more importantly, because it allows us to reliably
1872 restart receiving a signal such as SIGWINCH. (There was an
1873 actual Debian bug report about --limit-rate malfunctioning while
1874 the terminal was being resized.) */
1875 struct timespec sleep, remaining;
1876 sleep.tv_sec = (long) seconds;
1877 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1878 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1879 /* If nanosleep has been interrupted by a signal, adjust the
1880 sleeping period and return to sleep. */
1882 #elif defined(HAVE_USLEEP)
1883 /* If usleep is available, use it in preference to select. */
1886 /* On some systems, usleep cannot handle values larger than
1887 1,000,000. If the period is larger than that, use sleep
1888 first, then add usleep for subsecond accuracy. */
1890 seconds -= (long) seconds;
1892 usleep (seconds * 1000000);
1893 #else /* fall back select */
1894 /* Note that, although Windows supports select, it can't be used to
1895 implement sleeping because Winsock's select doesn't implement
1896 timeout when it is passed NULL pointers for all fd sets. (But it
1897 does under Cygwin, which implements Unix-compatible select.) */
1898 struct timeval sleep;
1899 sleep.tv_sec = (long) seconds;
1900 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1901 select (0, NULL, NULL, NULL, &sleep);
1902 /* If select returns -1 and errno is EINTR, it means we were
1903 interrupted by a signal. But without knowing how long we've
1904 actually slept, we can't return to sleep. Using gettimeofday to
1905 track sleeps is slow and unreliable due to clock skew. */
1909 #endif /* not WINDOWS */
1911 /* Encode the octets in DATA of length LENGTH to base64 format,
1912 storing the result to DEST. The output will be zero-terminated,
1913 and must point to a writable buffer of at least
1914 1+BASE64_LENGTH(length) bytes. The function returns the length of
1915 the resulting base64 data, not counting the terminating zero.
1917 This implementation does not emit newlines after 76 characters of
1921 base64_encode (const void *data, int length, char *dest)
1923 /* Conversion table. */
1924 static const char tbl[64] = {
1925 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
1926 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
1927 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
1928 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
1930 /* Access bytes in DATA as unsigned char, otherwise the shifts below
1931 don't work for data with MSB set. */
1932 const unsigned char *s = data;
1933 /* Theoretical ANSI violation when length < 3. */
1934 const unsigned char *end = (const unsigned char *) data + length - 2;
1937 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1938 for (; s < end; s += 3)
1940 *p++ = tbl[s[0] >> 2];
1941 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1942 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1943 *p++ = tbl[s[2] & 0x3f];
1946 /* Pad the result if necessary... */
1950 *p++ = tbl[s[0] >> 2];
1951 *p++ = tbl[(s[0] & 3) << 4];
1956 *p++ = tbl[s[0] >> 2];
1957 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1958 *p++ = tbl[((s[1] & 0xf) << 2)];
1962 /* ...and zero-terminate it. */
1968 /* Store in C the next non-whitespace character from the string, or \0
1969 when end of string is reached. */
1970 #define NEXT_CHAR(c, p) do { \
1971 c = (unsigned char) *p++; \
1972 } while (c_isspace (c))
1974 #define IS_ASCII(c) (((c) & 0x80) == 0)
1976 /* Decode data from BASE64 (a null-terminated string) into memory
1977 pointed to by DEST. DEST is assumed to be large enough to
1978 accomodate the decoded data, which is guaranteed to be no more than
1981 Since DEST is assumed to contain binary data, it is not
1982 NUL-terminated. The function returns the length of the data
1983 written to TO. -1 is returned in case of error caused by malformed
1986 This function originates from Free Recode. */
1989 base64_decode (const char *base64, void *dest)
1991 /* Table of base64 values for first 128 characters. Note that this
1992 assumes ASCII (but so does Wget in other places). */
1993 static const signed char base64_char_to_value[128] =
1995 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1996 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1997 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1998 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1999 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2000 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2001 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2002 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2003 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2004 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2005 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2006 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2007 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2009 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2010 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2012 const char *p = base64;
2018 unsigned long value;
2020 /* Process first byte of a quadruplet. */
2024 if (c == '=' || !IS_BASE64 (c))
2025 return -1; /* illegal char while decoding base64 */
2026 value = BASE64_CHAR_TO_VALUE (c) << 18;
2028 /* Process second byte of a quadruplet. */
2031 return -1; /* premature EOF while decoding base64 */
2032 if (c == '=' || !IS_BASE64 (c))
2033 return -1; /* illegal char while decoding base64 */
2034 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2037 /* Process third byte of a quadruplet. */
2040 return -1; /* premature EOF while decoding base64 */
2042 return -1; /* illegal char while decoding base64 */
2048 return -1; /* premature EOF while decoding base64 */
2050 return -1; /* padding `=' expected but not found */
2054 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2055 *q++ = 0xff & value >> 8;
2057 /* Process fourth byte of a quadruplet. */
2060 return -1; /* premature EOF while decoding base64 */
2064 return -1; /* illegal char while decoding base64 */
2066 value |= BASE64_CHAR_TO_VALUE (c);
2067 *q++ = 0xff & value;
2070 #undef BASE64_CHAR_TO_VALUE
2072 return q - (char *) dest;
2078 /* Simple merge sort for use by stable_sort. Implementation courtesy
2079 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2082 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2083 int (*cmpfun) (const void *, const void *))
2085 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2089 size_t mid = (to + from) / 2;
2090 mergesort_internal (base, temp, size, from, mid, cmpfun);
2091 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2094 for (k = from; (i <= mid) && (j <= to); k++)
2095 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2096 memcpy (ELT (temp, k), ELT (base, i++), size);
2098 memcpy (ELT (temp, k), ELT (base, j++), size);
2100 memcpy (ELT (temp, k++), ELT (base, i++), size);
2102 memcpy (ELT (temp, k++), ELT (base, j++), size);
2103 for (k = from; k <= to; k++)
2104 memcpy (ELT (base, k), ELT (temp, k), size);
2109 /* Stable sort with interface exactly like standard library's qsort.
2110 Uses mergesort internally, allocating temporary storage with
2114 stable_sort (void *base, size_t nmemb, size_t size,
2115 int (*cmpfun) (const void *, const void *))
2119 void *temp = alloca (nmemb * size * sizeof (void *));
2120 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2124 /* Print a decimal number. If it is equal to or larger than ten, the
2125 number is rounded. Otherwise it is printed with one significant
2126 digit without trailing zeros and with no more than three fractional
2127 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2128 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2130 This is useful for displaying durations because it provides
2131 order-of-magnitude information without unnecessary clutter --
2132 long-running downloads are shown without the fractional part, and
2133 short ones still retain one significant digit. */
2136 print_decimal (double number)
2138 static char buf[32];
2139 double n = number >= 0 ? number : -number;
2142 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2143 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2144 snprintf (buf, sizeof buf, "%.0f", number);
2146 snprintf (buf, sizeof buf, "%.1f", number);
2147 else if (n >= 0.001)
2148 snprintf (buf, sizeof buf, "%.1g", number);
2149 else if (n >= 0.0005)
2150 /* round [0.0005, 0.001) to 0.001 */
2151 snprintf (buf, sizeof buf, "%.3f", number);
2153 /* print numbers close to 0 as 0, not 0.000 */
2170 { "/somedir", "/somedir", true },
2171 { "/somedir", "/somedir/d2", true },
2172 { "/somedir/d1", "/somedir", false },
2175 for (i = 0; i < countof(test_array); ++i)
2177 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2179 mu_assert ("test_subdir_p: wrong result",
2180 res == test_array[i].result);
2187 test_dir_matches_p()
2195 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2196 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2197 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2198 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2199 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2200 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2201 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2202 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2203 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2204 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2205 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2206 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2207 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2210 for (i = 0; i < countof(test_array); ++i)
2212 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2214 mu_assert ("test_dir_matches_p: wrong result",
2215 res == test_array[i].result);
2221 #endif /* TESTING */