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 with xmalloc/xrealloc */
185 #else /* not HAVE_VASPRINTF */
187 /* vasprintf is unavailable. snprintf into a small buffer and
188 resize it as necessary. */
190 char *str = xmalloc (size);
192 /* #### This code will infloop and eventually abort in xrealloc if
193 passed a FMT that causes snprintf to consistently return -1. */
200 va_start (args, fmt);
201 n = vsnprintf (str, size, fmt, args);
204 /* If the printing worked, return the string. */
205 if (n > -1 && n < size)
208 /* Else try again with a larger buffer. */
209 if (n > -1) /* C99 */
210 size = n + 1; /* precisely what is needed */
213 if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
214 { /* maybe we have some wrong format string? */
215 free (str); /* In this case we must free already allocated memory, */
216 return NULL; /* and return NULL pointer... */
218 /* else, we continue to grow our buffer. */
219 size <<= 1; /* twice the old size */
221 str = xrealloc (str, size);
223 #endif /* not HAVE_VASPRINTF */
226 /* Concatenate the NULL-terminated list of string arguments into
227 freshly allocated space. */
230 concat_strings (const char *str0, ...)
233 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
236 const char *next_str;
237 int total_length = 0;
240 /* Calculate the length of and allocate the resulting string. */
243 va_start (args, str0);
244 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
246 int len = strlen (next_str);
247 if (argcount < countof (saved_lengths))
248 saved_lengths[argcount++] = len;
252 p = ret = xmalloc (total_length + 1);
254 /* Copy the strings into the allocated space. */
257 va_start (args, str0);
258 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
261 if (argcount < countof (saved_lengths))
262 len = saved_lengths[argcount++];
264 len = strlen (next_str);
265 memcpy (p, next_str, len);
274 /* Format the provided time according to the specified format. The
275 format is a string with format elements supported by strftime. */
278 fmttime (time_t t, const char *fmt)
280 static char output[32];
281 struct tm *tm = localtime(&t);
284 if (!strftime(output, sizeof(output), fmt, tm))
289 /* Return pointer to a static char[] buffer in which zero-terminated
290 string-representation of TM (in form hh:mm:ss) is printed.
292 If TM is NULL, the current time will be used. */
297 return fmttime(t, "%H:%M:%S");
300 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
303 datetime_str (time_t t)
305 return fmttime(t, "%Y-%m-%d %H:%M:%S");
308 /* The Windows versions of the following two functions are defined in
309 mswindows.c. On MSDOS this function should never be called. */
311 #if !defined(WINDOWS) && !defined(MSDOS)
313 fork_to_background (void)
316 /* Whether we arrange our own version of opt.lfilename here. */
317 bool logfile_changed = false;
321 /* We must create the file immediately to avoid either a race
322 condition (which arises from using unique_name and failing to
323 use fopen_excl) or lying to the user about the log file name
324 (which arises from using unique_name, printing the name, and
325 using fopen_excl later on.) */
326 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
329 logfile_changed = true;
342 /* parent, no error */
343 printf (_("Continuing in background, pid %d.\n"), (int) pid);
345 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
346 exit (0); /* #### should we use _exit()? */
349 /* child: give up the privileges and keep running. */
351 freopen ("/dev/null", "r", stdin);
352 freopen ("/dev/null", "w", stdout);
353 freopen ("/dev/null", "w", stderr);
355 #endif /* !WINDOWS && !MSDOS */
357 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
358 specified with TM. The atime ("access time") is set to the current
362 touch (const char *file, time_t tm)
364 #ifdef HAVE_STRUCT_UTIMBUF
365 struct utimbuf times;
373 times.actime = time (NULL);
374 if (utime (file, ×) == -1)
375 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
378 /* Checks if FILE is a symbolic link, and removes it if it is. Does
379 nothing under MS-Windows. */
381 remove_link (const char *file)
386 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
388 DEBUGP (("Unlinking %s (symlink).\n", file));
391 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
392 file, strerror (errno));
397 /* Does FILENAME exist? This is quite a lousy implementation, since
398 it supplies no error codes -- only a yes-or-no answer. Thus it
399 will return that a file does not exist if, e.g., the directory is
400 unreadable. I don't mind it too much currently, though. The
401 proper way should, of course, be to have a third, error state,
402 other than true/false, but that would introduce uncalled-for
403 additional complexity to the callers. */
405 file_exists_p (const char *filename)
408 return access (filename, F_OK) >= 0;
411 return stat (filename, &buf) >= 0;
415 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
416 Returns 0 on error. */
418 file_non_directory_p (const char *path)
421 /* Use lstat() rather than stat() so that symbolic links pointing to
422 directories can be identified correctly. */
423 if (lstat (path, &buf) != 0)
425 return S_ISDIR (buf.st_mode) ? false : true;
428 /* Return the size of file named by FILENAME, or -1 if it cannot be
429 opened or seeked into. */
431 file_size (const char *filename)
433 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
435 /* We use fseek rather than stat to determine the file size because
436 that way we can also verify that the file is readable without
437 explicitly checking for permissions. Inspired by the POST patch
439 FILE *fp = fopen (filename, "rb");
442 fseeko (fp, 0, SEEK_END);
448 if (stat (filename, &st) < 0)
454 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
455 doesn't exist is found. Return a freshly allocated copy of the
459 unique_name_1 (const char *prefix)
462 int plen = strlen (prefix);
463 char *template = (char *)alloca (plen + 1 + 24);
464 char *template_tail = template + plen;
466 memcpy (template, prefix, plen);
467 *template_tail++ = '.';
470 number_to_string (template_tail, count++);
471 while (file_exists_p (template));
473 return xstrdup (template);
476 /* Return a unique file name, based on FILE.
478 More precisely, if FILE doesn't exist, it is returned unmodified.
479 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
480 file name that doesn't exist is returned.
482 The resulting file is not created, only verified that it didn't
483 exist at the point in time when the function was called.
484 Therefore, where security matters, don't rely that the file created
485 by this function exists until you open it with O_EXCL or
488 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
489 string. Otherwise, it may return FILE if the file doesn't exist
490 (and therefore doesn't need changing). */
493 unique_name (const char *file, bool allow_passthrough)
495 /* If the FILE itself doesn't exist, return it without
497 if (!file_exists_p (file))
498 return allow_passthrough ? (char *)file : xstrdup (file);
500 /* Otherwise, find a numeric suffix that results in unused file name
502 return unique_name_1 (file);
505 /* Create a file based on NAME, except without overwriting an existing
506 file with that name. Providing O_EXCL is correctly implemented,
507 this function does not have the race condition associated with
508 opening the file returned by unique_name. */
511 unique_create (const char *name, bool binary, char **opened_name)
513 /* unique file name, based on NAME */
514 char *uname = unique_name (name, false);
516 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
519 uname = unique_name (name, false);
521 if (opened_name && fp != NULL)
524 *opened_name = uname;
536 /* Open the file for writing, with the addition that the file is
537 opened "exclusively". This means that, if the file already exists,
538 this function will *fail* and errno will be set to EEXIST. If
539 BINARY is set, the file will be opened in binary mode, equivalent
542 If opening the file fails for any reason, including the file having
543 previously existed, this function returns NULL and sets errno
547 fopen_excl (const char *fname, bool binary)
551 int flags = O_WRONLY | O_CREAT | O_EXCL;
556 fd = open (fname, flags, 0666);
559 return fdopen (fd, binary ? "wb" : "w");
560 #else /* not O_EXCL */
561 /* Manually check whether the file exists. This is prone to race
562 conditions, but systems without O_EXCL haven't deserved
564 if (file_exists_p (fname))
569 return fopen (fname, binary ? "wb" : "w");
570 #endif /* not O_EXCL */
573 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
574 are missing, create them first. In case any mkdir() call fails,
575 return its error status. Returns 0 on successful completion.
577 The behaviour of this function should be identical to the behaviour
578 of `mkdir -p' on systems where mkdir supports the `-p' option. */
580 make_directory (const char *directory)
582 int i, ret, quit = 0;
585 /* Make a copy of dir, to be able to write to it. Otherwise, the
586 function is unsafe if called with a read-only char *argument. */
587 STRDUP_ALLOCA (dir, directory);
589 /* If the first character of dir is '/', skip it (and thus enable
590 creation of absolute-pathname directories. */
591 for (i = (*dir == '/'); 1; ++i)
593 for (; dir[i] && dir[i] != '/'; i++)
598 /* Check whether the directory already exists. Allow creation of
599 of intermediate directories to fail, as the initial path components
600 are not necessarily directories! */
601 if (!file_exists_p (dir))
602 ret = mkdir (dir, 0777);
613 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
614 should be a file name.
616 file_merge("/foo/bar", "baz") => "/foo/baz"
617 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
618 file_merge("foo", "bar") => "bar"
620 In other words, it's a simpler and gentler version of uri_merge. */
623 file_merge (const char *base, const char *file)
626 const char *cut = (const char *)strrchr (base, '/');
629 return xstrdup (file);
631 result = xmalloc (cut - base + 1 + strlen (file) + 1);
632 memcpy (result, base, cut - base);
633 result[cut - base] = '/';
634 strcpy (result + (cut - base) + 1, file);
639 /* Like fnmatch, but performs a case-insensitive match. */
642 fnmatch_nocase (const char *pattern, const char *string, int flags)
645 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
646 also present on *BSD platforms, and possibly elsewhere. */
647 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
649 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
650 char *patcopy = (char *) alloca (strlen (pattern) + 1);
651 char *strcopy = (char *) alloca (strlen (string) + 1);
653 for (p = patcopy; *pattern; pattern++, p++)
654 *p = c_tolower (*pattern);
656 for (p = strcopy; *string; string++, p++)
657 *p = c_tolower (*string);
659 return fnmatch (patcopy, strcopy, flags);
663 static bool in_acclist (const char *const *, const char *, bool);
665 /* Determine whether a file is acceptable to be followed, according to
666 lists of patterns to accept/reject. */
668 acceptable (const char *s)
672 while (l && s[l] != '/')
679 return (in_acclist ((const char *const *)opt.accepts, s, true)
680 && !in_acclist ((const char *const *)opt.rejects, s, true));
682 return in_acclist ((const char *const *)opt.accepts, s, true);
684 else if (opt.rejects)
685 return !in_acclist ((const char *const *)opt.rejects, s, true);
689 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
690 will return true if and only if D2 begins with `/something/' or is exactly
693 subdir_p (const char *d1, const char *d2)
697 if (!opt.ignore_case)
698 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
701 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
704 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
707 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
708 first element that matches DIR, through wildcards or front comparison (as
711 dir_matches_p (char **dirlist, const char *dir)
714 int (*matcher) (const char *, const char *, int)
715 = opt.ignore_case ? fnmatch_nocase : fnmatch;
717 for (x = dirlist; *x; x++)
719 /* Remove leading '/' */
720 char *p = *x + (**x == '/');
721 if (has_wildcards_p (p))
723 if (matcher (p, dir, FNM_PATHNAME) == 0)
728 if (subdir_p (p, dir))
733 return *x ? true : false;
736 /* Returns whether DIRECTORY is acceptable for download, wrt the
737 include/exclude lists.
739 The leading `/' is ignored in paths; relative and absolute paths
740 may be freely intermixed. */
743 accdir (const char *directory)
745 /* Remove starting '/'. */
746 if (*directory == '/')
750 if (!dir_matches_p (opt.includes, directory))
755 if (dir_matches_p (opt.excludes, directory))
761 /* Return true if STRING ends with TAIL. For instance:
763 match_tail ("abc", "bc", false) -> 1
764 match_tail ("abc", "ab", false) -> 0
765 match_tail ("abc", "abc", false) -> 1
767 If FOLD_CASE is true, the comparison will be case-insensitive. */
770 match_tail (const char *string, const char *tail, bool fold_case)
774 /* We want this to be fast, so we code two loops, one with
775 case-folding, one without. */
779 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
780 if (string[i] != tail[j])
785 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
786 if (c_tolower (string[i]) != c_tolower (tail[j]))
790 /* If the tail was exhausted, the match was succesful. */
797 /* Checks whether string S matches each element of ACCEPTS. A list
798 element are matched either with fnmatch() or match_tail(),
799 according to whether the element contains wildcards or not.
801 If the BACKWARD is false, don't do backward comparison -- just compare
804 in_acclist (const char *const *accepts, const char *s, bool backward)
806 for (; *accepts; accepts++)
808 if (has_wildcards_p (*accepts))
810 int res = opt.ignore_case
811 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
812 /* fnmatch returns 0 if the pattern *does* match the string. */
820 if (match_tail (s, *accepts, opt.ignore_case))
825 int cmp = opt.ignore_case
826 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
835 /* Return the location of STR's suffix (file extension). Examples:
836 suffix ("foo.bar") -> "bar"
837 suffix ("foo.bar.baz") -> "baz"
838 suffix ("/foo/bar") -> NULL
839 suffix ("/foo.bar/baz") -> NULL */
841 suffix (const char *str)
845 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
849 return (char *)str + i;
854 /* Return true if S contains globbing wildcards (`*', `?', `[' or
858 has_wildcards_p (const char *s)
861 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
866 /* Return true if FNAME ends with a typical HTML suffix. The
867 following (case-insensitive) suffixes are presumed to be HTML
872 ?html (`?' matches one character)
874 #### CAVEAT. This is not necessarily a good indication that FNAME
875 refers to a file that contains HTML! */
877 has_html_suffix_p (const char *fname)
881 if ((suf = suffix (fname)) == NULL)
883 if (!strcasecmp (suf, "html"))
885 if (!strcasecmp (suf, "htm"))
887 if (suf[0] && !strcasecmp (suf + 1, "html"))
892 /* Read a line from FP and return the pointer to freshly allocated
893 storage. The storage space is obtained through malloc() and should
894 be freed with free() when it is no longer needed.
896 The length of the line is not limited, except by available memory.
897 The newline character at the end of line is retained. The line is
898 terminated with a zero character.
900 After end-of-file is encountered without anything being read, NULL
901 is returned. NULL is also returned on error. To distinguish
902 between these two cases, use the stdio function ferror(). */
905 read_whole_line (FILE *fp)
909 char *line = xmalloc (bufsize);
911 while (fgets (line + length, bufsize - length, fp))
913 length += strlen (line + length);
915 /* Possible for example when reading from a binary file where
916 a line begins with \0. */
919 if (line[length - 1] == '\n')
922 /* fgets() guarantees to read the whole line, or to use up the
923 space we've given it. We can double the buffer
926 line = xrealloc (line, bufsize);
928 if (length == 0 || ferror (fp))
933 if (length + 1 < bufsize)
934 /* Relieve the memory from our exponential greediness. We say
935 `length + 1' because the terminating \0 is not included in
936 LENGTH. We don't need to zero-terminate the string ourselves,
937 though, because fgets() does that. */
938 line = xrealloc (line, length + 1);
942 /* Read FILE into memory. A pointer to `struct file_memory' are
943 returned; use struct element `content' to access file contents, and
944 the element `length' to know the file length. `content' is *not*
945 zero-terminated, and you should *not* read or write beyond the [0,
946 length) range of characters.
948 After you are done with the file contents, call read_file_free to
951 Depending on the operating system and the type of file that is
952 being read, read_file() either mmap's the file into memory, or
953 reads the file into the core using read().
955 If file is named "-", fileno(stdin) is used for reading instead.
956 If you want to read from a real file named "-", use "./-" instead. */
959 read_file (const char *file)
962 struct file_memory *fm;
964 bool inhibit_close = false;
966 /* Some magic in the finest tradition of Perl and its kin: if FILE
967 is "-", just use stdin. */
971 inhibit_close = true;
972 /* Note that we don't inhibit mmap() in this case. If stdin is
973 redirected from a regular file, mmap() will still work. */
976 fd = open (file, O_RDONLY);
979 fm = xnew (struct file_memory);
984 if (fstat (fd, &buf) < 0)
986 fm->length = buf.st_size;
987 /* NOTE: As far as I know, the callers of this function never
988 modify the file text. Relying on this would enable us to
989 specify PROT_READ and MAP_SHARED for a marginal gain in
990 efficiency, but at some cost to generality. */
991 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
993 if (fm->content == (char *)MAP_FAILED)
1003 /* The most common reason why mmap() fails is that FD does not point
1004 to a plain file. However, it's also possible that mmap() doesn't
1005 work for a particular type of file. Therefore, whenever mmap()
1006 fails, we just fall back to the regular method. */
1007 #endif /* HAVE_MMAP */
1010 size = 512; /* number of bytes fm->contents can
1011 hold at any given time. */
1012 fm->content = xmalloc (size);
1016 if (fm->length > size / 2)
1018 /* #### I'm not sure whether the whole exponential-growth
1019 thing makes sense with kernel read. On Linux at least,
1020 read() refuses to read more than 4K from a file at a
1021 single chunk anyway. But other Unixes might optimize it
1022 better, and it doesn't *hurt* anything, so I'm leaving
1025 /* Normally, we grow SIZE exponentially to make the number
1026 of calls to read() and realloc() logarithmic in relation
1027 to file size. However, read() can read an amount of data
1028 smaller than requested, and it would be unreasonable to
1029 double SIZE every time *something* was read. Therefore,
1030 we double SIZE only when the length exceeds half of the
1031 entire allocated size. */
1033 fm->content = xrealloc (fm->content, size);
1035 nread = read (fd, fm->content + fm->length, size - fm->length);
1037 /* Successful read. */
1038 fm->length += nread;
1048 if (size > fm->length && fm->length != 0)
1049 /* Due to exponential growth of fm->content, the allocated region
1050 might be much larger than what is actually needed. */
1051 fm->content = xrealloc (fm->content, fm->length);
1058 xfree (fm->content);
1063 /* Release the resources held by FM. Specifically, this calls
1064 munmap() or xfree() on fm->content, depending whether mmap or
1065 malloc/read were used to read in the file. It also frees the
1066 memory needed to hold the FM structure itself. */
1069 read_file_free (struct file_memory *fm)
1074 munmap (fm->content, fm->length);
1079 xfree (fm->content);
1084 /* Free the pointers in a NULL-terminated vector of pointers, then
1085 free the pointer itself. */
1087 free_vec (char **vec)
1098 /* Append vector V2 to vector V1. The function frees V2 and
1099 reallocates V1 (thus you may not use the contents of neither
1100 pointer after the call). If V1 is NULL, V2 is returned. */
1102 merge_vecs (char **v1, char **v2)
1112 /* To avoid j == 0 */
1117 for (i = 0; v1[i]; i++)
1120 for (j = 0; v2[j]; j++)
1122 /* Reallocate v1. */
1123 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1124 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1129 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1130 is allocated as needed. Return the new value of the vector. */
1133 vec_append (char **vec, const char *str)
1135 int cnt; /* count of vector elements, including
1136 the one we're about to append */
1139 for (cnt = 0; vec[cnt]; cnt++)
1145 /* Reallocate the array to fit the new element and the NULL. */
1146 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1147 /* Append a copy of STR to the vector. */
1148 vec[cnt - 1] = xstrdup (str);
1153 /* Sometimes it's useful to create "sets" of strings, i.e. special
1154 hash tables where you want to store strings as keys and merely
1155 query for their existence. Here is a set of utility routines that
1156 makes that transparent. */
1159 string_set_add (struct hash_table *ht, const char *s)
1161 /* First check whether the set element already exists. If it does,
1162 do nothing so that we don't have to free() the old element and
1163 then strdup() a new one. */
1164 if (hash_table_contains (ht, s))
1167 /* We use "1" as value. It provides us a useful and clear arbitrary
1168 value, and it consumes no memory -- the pointers to the same
1169 string "1" will be shared by all the key-value pairs in all `set'
1171 hash_table_put (ht, xstrdup (s), "1");
1174 /* Synonym for hash_table_contains... */
1177 string_set_contains (struct hash_table *ht, const char *s)
1179 return hash_table_contains (ht, s);
1182 /* Convert the specified string set to array. ARRAY should be large
1183 enough to hold hash_table_count(ht) char pointers. */
1185 void string_set_to_array (struct hash_table *ht, char **array)
1187 hash_table_iterator iter;
1188 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1189 *array++ = iter.key;
1192 /* Free the string set. This frees both the storage allocated for
1193 keys and the actual hash table. (hash_table_destroy would only
1194 destroy the hash table.) */
1197 string_set_free (struct hash_table *ht)
1199 hash_table_iterator iter;
1200 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1202 hash_table_destroy (ht);
1205 /* Utility function: simply call xfree() on all keys and values of HT. */
1208 free_keys_and_values (struct hash_table *ht)
1210 hash_table_iterator iter;
1211 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1218 /* Get digit grouping data for thousand separors by calling
1219 localeconv(). The data includes separator string and grouping info
1220 and is cached after the first call to the function.
1222 In locales that don't set a thousand separator (such as the "C"
1223 locale), this forces it to be ",". We are now only showing
1224 thousand separators in one place, so this shouldn't be a problem in
1228 get_grouping_data (const char **sep, const char **grouping)
1230 static const char *cached_sep;
1231 static const char *cached_grouping;
1232 static bool initialized;
1235 /* Get the grouping info from the locale. */
1236 struct lconv *lconv = localeconv ();
1237 cached_sep = lconv->thousands_sep;
1238 cached_grouping = lconv->grouping;
1239 #if ! USE_NLS_PROGRESS_BAR
1240 /* We can't count column widths, so ensure that the separator
1241 * is single-byte only (let check below determine what byte). */
1242 if (strlen(cached_sep) > 1)
1247 /* Many locales (such as "C" or "hr_HR") don't specify
1248 grouping, which we still want to use it for legibility.
1249 In those locales set the sep char to ',', unless that
1250 character is used for decimal point, in which case set it
1252 if (*lconv->decimal_point != ',')
1256 cached_grouping = "\x03";
1261 *grouping = cached_grouping;
1264 /* Return a printed representation of N with thousand separators.
1265 This should respect locale settings, with the exception of the "C"
1266 locale which mandates no separator, but we use one anyway.
1268 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1269 the separators because it's too non-portable, and it's hard to test
1270 for this feature at configure time. Besides, it wouldn't display
1271 separators in the "C" locale, still used by many Unix users. */
1274 with_thousand_seps (wgint n)
1276 static char outbuf[48];
1277 char *p = outbuf + sizeof outbuf;
1279 /* Info received from locale */
1280 const char *grouping, *sep;
1283 /* State information */
1284 int i = 0, groupsize;
1285 const char *atgroup;
1287 bool negative = n < 0;
1289 /* Initialize grouping data. */
1290 get_grouping_data (&sep, &grouping);
1291 seplen = strlen (sep);
1293 groupsize = *atgroup++;
1295 /* This would overflow on WGINT_MIN, but printing negative numbers
1296 is not an important goal of this fuinction. */
1300 /* Write the number into the buffer, backwards, inserting the
1301 separators as necessary. */
1305 *--p = n % 10 + '0';
1309 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1310 if (++i == groupsize)
1315 memcpy (p -= seplen, sep, seplen);
1318 groupsize = *atgroup++;
1327 /* N, a byte quantity, is converted to a human-readable abberviated
1328 form a la sizes printed by `ls -lh'. The result is written to a
1329 static buffer, a pointer to which is returned.
1331 Unlike `with_thousand_seps', this approximates to the nearest unit.
1332 Quoting GNU libit: "Most people visually process strings of 3-4
1333 digits effectively, but longer strings of digits are more prone to
1334 misinterpretation. Hence, converting to an abbreviated form
1335 usually improves readability."
1337 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1338 original computer-related meaning of "powers of 1024". We don't
1339 use the "*bibyte" names invented in 1998, and seldom used in
1340 practice. Wikipedia's entry on "binary prefix" discusses this in
1344 human_readable (HR_NUMTYPE n)
1346 /* These suffixes are compatible with those of GNU `ls -lh'. */
1347 static char powers[] =
1349 'K', /* kilobyte, 2^10 bytes */
1350 'M', /* megabyte, 2^20 bytes */
1351 'G', /* gigabyte, 2^30 bytes */
1352 'T', /* terabyte, 2^40 bytes */
1353 'P', /* petabyte, 2^50 bytes */
1354 'E', /* exabyte, 2^60 bytes */
1359 /* If the quantity is smaller than 1K, just print it. */
1362 snprintf (buf, sizeof (buf), "%d", (int) n);
1366 /* Loop over powers, dividing N with 1024 in each iteration. This
1367 works unchanged for all sizes of wgint, while still avoiding
1368 non-portable `long double' arithmetic. */
1369 for (i = 0; i < countof (powers); i++)
1371 /* At each iteration N is greater than the *subsequent* power.
1372 That way N/1024.0 produces a decimal number in the units of
1374 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1376 double val = n / 1024.0;
1377 /* Print values smaller than 10 with one decimal digits, and
1378 others without any decimals. */
1379 snprintf (buf, sizeof (buf), "%.*f%c",
1380 val < 10 ? 1 : 0, val, powers[i]);
1385 return NULL; /* unreached */
1388 /* Count the digits in the provided number. Used to allocate space
1389 when printing numbers. */
1392 numdigit (wgint number)
1396 ++cnt; /* accomodate '-' */
1397 while ((number /= 10) != 0)
1402 #define PR(mask) *p++ = n / (mask) + '0'
1404 /* DIGITS_<D> is used to print a D-digit number and should be called
1405 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1406 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1407 Recursively this continues until DIGITS_1 is invoked. */
1409 #define DIGITS_1(mask) PR (mask)
1410 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1411 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1412 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1413 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1414 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1415 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1416 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1417 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1418 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1420 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1422 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1423 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1424 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1425 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1426 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1427 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1428 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1429 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1430 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1432 /* Shorthand for casting to wgint. */
1435 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1436 `sprintf(buffer, "%lld", (long long) number)', only typically much
1437 faster and portable to machines without long long.
1439 The speedup may make a difference in programs that frequently
1440 convert numbers to strings. Some implementations of sprintf,
1441 particularly the one in some versions of GNU libc, have been known
1442 to be quite slow when converting integers to strings.
1444 Return the pointer to the location where the terminating zero was
1445 printed. (Equivalent to calling buffer+strlen(buffer) after the
1448 BUFFER should be large enough to accept as many bytes as you expect
1449 the number to take up. On machines with 64-bit wgints the maximum
1450 needed size is 24 bytes. That includes the digits needed for the
1451 largest 64-bit number, the `-' sign in case it's negative, and the
1452 terminating '\0'. */
1455 number_to_string (char *buffer, wgint number)
1460 int last_digit_char = 0;
1462 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1463 /* We are running in a very strange environment. Leave the correct
1464 printing to sprintf. */
1465 p += sprintf (buf, "%j", (intmax_t) (n));
1466 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1472 /* n = -n would overflow because -n would evaluate to a
1473 wgint value larger than WGINT_MAX. Need to make n
1474 smaller and handle the last digit separately. */
1475 int last_digit = n % 10;
1476 /* The sign of n%10 is implementation-defined. */
1478 last_digit_char = '0' - last_digit;
1480 last_digit_char = '0' + last_digit;
1481 /* After n is made smaller, -n will not overflow. */
1489 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1490 way printing any N is fully open-coded without a loop or jump.
1491 (Also see description of DIGITS_*.) */
1493 if (n < 10) DIGITS_1 (1);
1494 else if (n < 100) DIGITS_2 (10);
1495 else if (n < 1000) DIGITS_3 (100);
1496 else if (n < 10000) DIGITS_4 (1000);
1497 else if (n < 100000) DIGITS_5 (10000);
1498 else if (n < 1000000) DIGITS_6 (100000);
1499 else if (n < 10000000) DIGITS_7 (1000000);
1500 else if (n < 100000000) DIGITS_8 (10000000);
1501 else if (n < 1000000000) DIGITS_9 (100000000);
1502 #if SIZEOF_WGINT == 4
1503 /* wgint is 32 bits wide: no number has more than 10 digits. */
1504 else DIGITS_10 (1000000000);
1506 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1507 Constants are constructed by compile-time multiplication to avoid
1508 dealing with different notations for 64-bit constants
1509 (nL/nLL/nI64, depending on the compiler and architecture). */
1510 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1511 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1512 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1513 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1514 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1515 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1516 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1517 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1518 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1519 else DIGITS_19 (1000000000*(W)1000000000);
1522 if (last_digit_char)
1523 *p++ = last_digit_char;
1526 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1533 #undef SPRINTF_WGINT
1556 /* Print NUMBER to a statically allocated string and return a pointer
1557 to the printed representation.
1559 This function is intended to be used in conjunction with printf.
1560 It is hard to portably print wgint values:
1561 a) you cannot use printf("%ld", number) because wgint can be long
1562 long on 32-bit machines with LFS.
1563 b) you cannot use printf("%lld", number) because NUMBER could be
1564 long on 32-bit machines without LFS, or on 64-bit machines,
1565 which do not require LFS. Also, Windows doesn't support %lld.
1566 c) you cannot use printf("%j", (int_max_t) number) because not all
1567 versions of printf support "%j", the most notable being the one
1569 d) you cannot #define WGINT_FMT to the appropriate format and use
1570 printf(WGINT_FMT, number) because that would break translations
1571 for user-visible messages, such as printf("Downloaded: %d
1574 What you should use instead is printf("%s", number_to_static_string
1577 CAVEAT: since the function returns pointers to static data, you
1578 must be careful to copy its result before calling it again.
1579 However, to make it more useful with printf, the function maintains
1580 an internal ring of static buffers to return. That way things like
1581 printf("%s %s", number_to_static_string (num1),
1582 number_to_static_string (num2)) work as expected. Three buffers
1583 are currently used, which means that "%s %s %s" will work, but "%s
1584 %s %s %s" won't. If you need to print more than three wgints,
1585 bump the RING_SIZE (or rethink your message.) */
1588 number_to_static_string (wgint number)
1590 static char ring[RING_SIZE][24];
1592 char *buf = ring[ringpos];
1593 number_to_string (buf, number);
1594 ringpos = (ringpos + 1) % RING_SIZE;
1598 /* Determine the width of the terminal we're running on. If that's
1599 not possible, return 0. */
1602 determine_screen_width (void)
1604 /* If there's a way to get the terminal size using POSIX
1605 tcgetattr(), somebody please tell me. */
1610 if (opt.lfilename != NULL)
1613 fd = fileno (stderr);
1614 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1615 return 0; /* most likely ENOTTY */
1618 #elif defined(WINDOWS)
1619 CONSOLE_SCREEN_BUFFER_INFO csbi;
1620 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1622 return csbi.dwSize.X;
1623 #else /* neither TIOCGWINSZ nor WINDOWS */
1625 #endif /* neither TIOCGWINSZ nor WINDOWS */
1628 /* Whether the rnd system (either rand or [dl]rand48) has been
1630 static int rnd_seeded;
1632 /* Return a random number between 0 and MAX-1, inclusive.
1634 If the system does not support lrand48 and MAX is greater than the
1635 value of RAND_MAX+1 on the system, the returned value will be in
1636 the range [0, RAND_MAX]. This may be fixed in a future release.
1637 The random number generator is seeded automatically the first time
1640 This uses lrand48 where available, rand elsewhere. DO NOT use it
1641 for cryptography. It is only meant to be used in situations where
1642 quality of the random numbers returned doesn't really matter. */
1645 random_number (int max)
1650 srand48 ((long) time (NULL) ^ (long) getpid ());
1653 return lrand48 () % max;
1654 #else /* not HAVE_DRAND48 */
1660 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1665 /* Like rand() % max, but uses the high-order bits for better
1666 randomness on architectures where rand() is implemented using a
1667 simple congruential generator. */
1669 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1670 return (int) bounded;
1672 #endif /* not HAVE_DRAND48 */
1675 /* Return a random uniformly distributed floating point number in the
1676 [0, 1) range. Uses drand48 where available, and a really lame
1677 kludge elsewhere. */
1685 srand48 ((long) time (NULL) ^ (long) getpid ());
1689 #else /* not HAVE_DRAND48 */
1690 return ( random_number (10000) / 10000.0
1691 + random_number (10000) / (10000.0 * 10000.0)
1692 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1693 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1694 #endif /* not HAVE_DRAND48 */
1697 /* Implementation of run_with_timeout, a generic timeout-forcing
1698 routine for systems with Unix-like signal handling. */
1700 #ifdef USE_SIGNAL_TIMEOUT
1701 # ifdef HAVE_SIGSETJMP
1702 # define SETJMP(env) sigsetjmp (env, 1)
1704 static sigjmp_buf run_with_timeout_env;
1707 abort_run_with_timeout (int sig)
1709 assert (sig == SIGALRM);
1710 siglongjmp (run_with_timeout_env, -1);
1712 # else /* not HAVE_SIGSETJMP */
1713 # define SETJMP(env) setjmp (env)
1715 static jmp_buf run_with_timeout_env;
1718 abort_run_with_timeout (int sig)
1720 assert (sig == SIGALRM);
1721 /* We don't have siglongjmp to preserve the set of blocked signals;
1722 if we longjumped out of the handler at this point, SIGALRM would
1723 remain blocked. We must unblock it manually. */
1724 int mask = siggetmask ();
1725 mask &= ~sigmask (SIGALRM);
1728 /* Now it's safe to longjump. */
1729 longjmp (run_with_timeout_env, -1);
1731 # endif /* not HAVE_SIGSETJMP */
1733 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1734 setitimer where available, alarm otherwise.
1736 TIMEOUT should be non-zero. If the timeout value is so small that
1737 it would be rounded to zero, it is rounded to the least legal value
1738 instead (1us for setitimer, 1s for alarm). That ensures that
1739 SIGALRM will be delivered in all cases. */
1742 alarm_set (double timeout)
1745 /* Use the modern itimer interface. */
1746 struct itimerval itv;
1748 itv.it_value.tv_sec = (long) timeout;
1749 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1750 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1751 /* Ensure that we wait for at least the minimum interval.
1752 Specifying zero would mean "wait forever". */
1753 itv.it_value.tv_usec = 1;
1754 setitimer (ITIMER_REAL, &itv, NULL);
1755 #else /* not ITIMER_REAL */
1756 /* Use the old alarm() interface. */
1757 int secs = (int) timeout;
1759 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1760 because alarm(0) means "never deliver the alarm", i.e. "wait
1761 forever", which is not what someone who specifies a 0.5s
1762 timeout would expect. */
1765 #endif /* not ITIMER_REAL */
1768 /* Cancel the alarm set with alarm_set. */
1774 struct itimerval disable;
1776 setitimer (ITIMER_REAL, &disable, NULL);
1777 #else /* not ITIMER_REAL */
1779 #endif /* not ITIMER_REAL */
1782 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1783 seconds. Returns true if the function was interrupted with a
1784 timeout, false otherwise.
1786 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1787 using setitimer() or alarm(). The timeout is enforced by
1788 longjumping out of the SIGALRM handler. This has several
1789 advantages compared to the traditional approach of relying on
1790 signals causing system calls to exit with EINTR:
1792 * The callback function is *forcibly* interrupted after the
1793 timeout expires, (almost) regardless of what it was doing and
1794 whether it was in a syscall. For example, a calculation that
1795 takes a long time is interrupted as reliably as an IO
1798 * It works with both SYSV and BSD signals because it doesn't
1799 depend on the default setting of SA_RESTART.
1801 * It doesn't require special handler setup beyond a simple call
1802 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1805 The only downside is that, if FUN allocates internal resources that
1806 are normally freed prior to exit from the functions, they will be
1807 lost in case of timeout. */
1810 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1820 signal (SIGALRM, abort_run_with_timeout);
1821 if (SETJMP (run_with_timeout_env) != 0)
1823 /* Longjumped out of FUN with a timeout. */
1824 signal (SIGALRM, SIG_DFL);
1827 alarm_set (timeout);
1830 /* Preserve errno in case alarm() or signal() modifies it. */
1831 saved_errno = errno;
1833 signal (SIGALRM, SIG_DFL);
1834 errno = saved_errno;
1839 #else /* not USE_SIGNAL_TIMEOUT */
1842 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1843 define it under Windows, because Windows has its own version of
1844 run_with_timeout that uses threads. */
1847 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1852 #endif /* not WINDOWS */
1853 #endif /* not USE_SIGNAL_TIMEOUT */
1857 /* Sleep the specified amount of seconds. On machines without
1858 nanosleep(), this may sleep shorter if interrupted by signals. */
1861 xsleep (double seconds)
1863 #ifdef HAVE_NANOSLEEP
1864 /* nanosleep is the preferred interface because it offers high
1865 accuracy and, more importantly, because it allows us to reliably
1866 restart receiving a signal such as SIGWINCH. (There was an
1867 actual Debian bug report about --limit-rate malfunctioning while
1868 the terminal was being resized.) */
1869 struct timespec sleep, remaining;
1870 sleep.tv_sec = (long) seconds;
1871 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1872 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1873 /* If nanosleep has been interrupted by a signal, adjust the
1874 sleeping period and return to sleep. */
1876 #elif defined(HAVE_USLEEP)
1877 /* If usleep is available, use it in preference to select. */
1880 /* On some systems, usleep cannot handle values larger than
1881 1,000,000. If the period is larger than that, use sleep
1882 first, then add usleep for subsecond accuracy. */
1884 seconds -= (long) seconds;
1886 usleep (seconds * 1000000);
1887 #else /* fall back select */
1888 /* Note that, although Windows supports select, it can't be used to
1889 implement sleeping because Winsock's select doesn't implement
1890 timeout when it is passed NULL pointers for all fd sets. (But it
1891 does under Cygwin, which implements Unix-compatible select.) */
1892 struct timeval sleep;
1893 sleep.tv_sec = (long) seconds;
1894 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1895 select (0, NULL, NULL, NULL, &sleep);
1896 /* If select returns -1 and errno is EINTR, it means we were
1897 interrupted by a signal. But without knowing how long we've
1898 actually slept, we can't return to sleep. Using gettimeofday to
1899 track sleeps is slow and unreliable due to clock skew. */
1903 #endif /* not WINDOWS */
1905 /* Encode the octets in DATA of length LENGTH to base64 format,
1906 storing the result to DEST. The output will be zero-terminated,
1907 and must point to a writable buffer of at least
1908 1+BASE64_LENGTH(length) bytes. The function returns the length of
1909 the resulting base64 data, not counting the terminating zero.
1911 This implementation does not emit newlines after 76 characters of
1915 base64_encode (const void *data, int length, char *dest)
1917 /* Conversion table. */
1918 static const char tbl[64] = {
1919 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
1920 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
1921 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
1922 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
1924 /* Access bytes in DATA as unsigned char, otherwise the shifts below
1925 don't work for data with MSB set. */
1926 const unsigned char *s = data;
1927 /* Theoretical ANSI violation when length < 3. */
1928 const unsigned char *end = (const unsigned char *) data + length - 2;
1931 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1932 for (; s < end; s += 3)
1934 *p++ = tbl[s[0] >> 2];
1935 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1936 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1937 *p++ = tbl[s[2] & 0x3f];
1940 /* Pad the result if necessary... */
1944 *p++ = tbl[s[0] >> 2];
1945 *p++ = tbl[(s[0] & 3) << 4];
1950 *p++ = tbl[s[0] >> 2];
1951 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1952 *p++ = tbl[((s[1] & 0xf) << 2)];
1956 /* ...and zero-terminate it. */
1962 /* Store in C the next non-whitespace character from the string, or \0
1963 when end of string is reached. */
1964 #define NEXT_CHAR(c, p) do { \
1965 c = (unsigned char) *p++; \
1966 } while (c_isspace (c))
1968 #define IS_ASCII(c) (((c) & 0x80) == 0)
1970 /* Decode data from BASE64 (a null-terminated string) into memory
1971 pointed to by DEST. DEST is assumed to be large enough to
1972 accomodate the decoded data, which is guaranteed to be no more than
1975 Since DEST is assumed to contain binary data, it is not
1976 NUL-terminated. The function returns the length of the data
1977 written to TO. -1 is returned in case of error caused by malformed
1980 This function originates from Free Recode. */
1983 base64_decode (const char *base64, void *dest)
1985 /* Table of base64 values for first 128 characters. Note that this
1986 assumes ASCII (but so does Wget in other places). */
1987 static const signed char base64_char_to_value[128] =
1989 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1990 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1991 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1992 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1993 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1994 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1995 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1996 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1997 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1998 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1999 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2000 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2001 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2003 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2004 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2006 const char *p = base64;
2012 unsigned long value;
2014 /* Process first byte of a quadruplet. */
2018 if (c == '=' || !IS_BASE64 (c))
2019 return -1; /* illegal char while decoding base64 */
2020 value = BASE64_CHAR_TO_VALUE (c) << 18;
2022 /* Process second byte of a quadruplet. */
2025 return -1; /* premature EOF while decoding base64 */
2026 if (c == '=' || !IS_BASE64 (c))
2027 return -1; /* illegal char while decoding base64 */
2028 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2031 /* Process third byte of a quadruplet. */
2034 return -1; /* premature EOF while decoding base64 */
2036 return -1; /* illegal char while decoding base64 */
2042 return -1; /* premature EOF while decoding base64 */
2044 return -1; /* padding `=' expected but not found */
2048 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2049 *q++ = 0xff & value >> 8;
2051 /* Process fourth byte of a quadruplet. */
2054 return -1; /* premature EOF while decoding base64 */
2058 return -1; /* illegal char while decoding base64 */
2060 value |= BASE64_CHAR_TO_VALUE (c);
2061 *q++ = 0xff & value;
2064 #undef BASE64_CHAR_TO_VALUE
2066 return q - (char *) dest;
2072 /* Simple merge sort for use by stable_sort. Implementation courtesy
2073 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2076 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2077 int (*cmpfun) (const void *, const void *))
2079 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2083 size_t mid = (to + from) / 2;
2084 mergesort_internal (base, temp, size, from, mid, cmpfun);
2085 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2088 for (k = from; (i <= mid) && (j <= to); k++)
2089 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2090 memcpy (ELT (temp, k), ELT (base, i++), size);
2092 memcpy (ELT (temp, k), ELT (base, j++), size);
2094 memcpy (ELT (temp, k++), ELT (base, i++), size);
2096 memcpy (ELT (temp, k++), ELT (base, j++), size);
2097 for (k = from; k <= to; k++)
2098 memcpy (ELT (base, k), ELT (temp, k), size);
2103 /* Stable sort with interface exactly like standard library's qsort.
2104 Uses mergesort internally, allocating temporary storage with
2108 stable_sort (void *base, size_t nmemb, size_t size,
2109 int (*cmpfun) (const void *, const void *))
2113 void *temp = alloca (nmemb * size * sizeof (void *));
2114 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2118 /* Print a decimal number. If it is equal to or larger than ten, the
2119 number is rounded. Otherwise it is printed with one significant
2120 digit without trailing zeros and with no more than three fractional
2121 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2122 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2124 This is useful for displaying durations because it provides
2125 order-of-magnitude information without unnecessary clutter --
2126 long-running downloads are shown without the fractional part, and
2127 short ones still retain one significant digit. */
2130 print_decimal (double number)
2132 static char buf[32];
2133 double n = number >= 0 ? number : -number;
2136 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2137 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2138 snprintf (buf, sizeof buf, "%.0f", number);
2140 snprintf (buf, sizeof buf, "%.1f", number);
2141 else if (n >= 0.001)
2142 snprintf (buf, sizeof buf, "%.1g", number);
2143 else if (n >= 0.0005)
2144 /* round [0.0005, 0.001) to 0.001 */
2145 snprintf (buf, sizeof buf, "%.3f", number);
2147 /* print numbers close to 0 as 0, not 0.000 */
2164 { "/somedir", "/somedir", true },
2165 { "/somedir", "/somedir/d2", true },
2166 { "/somedir/d1", "/somedir", false },
2169 for (i = 0; i < countof(test_array); ++i)
2171 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2173 mu_assert ("test_subdir_p: wrong result",
2174 res == test_array[i].result);
2181 test_dir_matches_p()
2189 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2190 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2191 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2192 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2193 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2194 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2195 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2196 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2197 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2198 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2199 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2200 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2201 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2204 for (i = 0; i < countof(test_array); ++i)
2206 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2208 mu_assert ("test_dir_matches_p: wrong result",
2209 res == test_array[i].result);
2215 #endif /* TESTING */