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 logprintf (LOG_ALWAYS,
217 _("%s: aprintf: text buffer is too big (%ld bytes), \
218 free already allocated memory and abort.\n"),
219 exec_name, size); /* printout a log message */
220 abort (); /* and abort... */
222 /* else, we continue to grow our buffer. */
223 size <<= 1; /* twice the old size */
225 str = xrealloc (str, size);
227 #endif /* not HAVE_VASPRINTF */
230 /* Concatenate the NULL-terminated list of string arguments into
231 freshly allocated space. */
234 concat_strings (const char *str0, ...)
237 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
240 const char *next_str;
241 int total_length = 0;
244 /* Calculate the length of and allocate the resulting string. */
247 va_start (args, str0);
248 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
250 int len = strlen (next_str);
251 if (argcount < countof (saved_lengths))
252 saved_lengths[argcount++] = len;
256 p = ret = xmalloc (total_length + 1);
258 /* Copy the strings into the allocated space. */
261 va_start (args, str0);
262 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
265 if (argcount < countof (saved_lengths))
266 len = saved_lengths[argcount++];
268 len = strlen (next_str);
269 memcpy (p, next_str, len);
278 /* Format the provided time according to the specified format. The
279 format is a string with format elements supported by strftime. */
282 fmttime (time_t t, const char *fmt)
284 static char output[32];
285 struct tm *tm = localtime(&t);
288 if (!strftime(output, sizeof(output), fmt, tm))
293 /* Return pointer to a static char[] buffer in which zero-terminated
294 string-representation of TM (in form hh:mm:ss) is printed.
296 If TM is NULL, the current time will be used. */
301 return fmttime(t, "%H:%M:%S");
304 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
307 datetime_str (time_t t)
309 return fmttime(t, "%Y-%m-%d %H:%M:%S");
312 /* The Windows versions of the following two functions are defined in
313 mswindows.c. On MSDOS this function should never be called. */
315 #if !defined(WINDOWS) && !defined(MSDOS)
317 fork_to_background (void)
320 /* Whether we arrange our own version of opt.lfilename here. */
321 bool logfile_changed = false;
325 /* We must create the file immediately to avoid either a race
326 condition (which arises from using unique_name and failing to
327 use fopen_excl) or lying to the user about the log file name
328 (which arises from using unique_name, printing the name, and
329 using fopen_excl later on.) */
330 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
333 logfile_changed = true;
346 /* parent, no error */
347 printf (_("Continuing in background, pid %d.\n"), (int) pid);
349 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
350 exit (0); /* #### should we use _exit()? */
353 /* child: give up the privileges and keep running. */
355 freopen ("/dev/null", "r", stdin);
356 freopen ("/dev/null", "w", stdout);
357 freopen ("/dev/null", "w", stderr);
359 #endif /* !WINDOWS && !MSDOS */
361 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
362 specified with TM. The atime ("access time") is set to the current
366 touch (const char *file, time_t tm)
368 #ifdef HAVE_STRUCT_UTIMBUF
369 struct utimbuf times;
377 times.actime = time (NULL);
378 if (utime (file, ×) == -1)
379 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
382 /* Checks if FILE is a symbolic link, and removes it if it is. Does
383 nothing under MS-Windows. */
385 remove_link (const char *file)
390 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
392 DEBUGP (("Unlinking %s (symlink).\n", file));
395 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
396 file, strerror (errno));
401 /* Does FILENAME exist? This is quite a lousy implementation, since
402 it supplies no error codes -- only a yes-or-no answer. Thus it
403 will return that a file does not exist if, e.g., the directory is
404 unreadable. I don't mind it too much currently, though. The
405 proper way should, of course, be to have a third, error state,
406 other than true/false, but that would introduce uncalled-for
407 additional complexity to the callers. */
409 file_exists_p (const char *filename)
412 return access (filename, F_OK) >= 0;
415 return stat (filename, &buf) >= 0;
419 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
420 Returns 0 on error. */
422 file_non_directory_p (const char *path)
425 /* Use lstat() rather than stat() so that symbolic links pointing to
426 directories can be identified correctly. */
427 if (lstat (path, &buf) != 0)
429 return S_ISDIR (buf.st_mode) ? false : true;
432 /* Return the size of file named by FILENAME, or -1 if it cannot be
433 opened or seeked into. */
435 file_size (const char *filename)
437 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
439 /* We use fseek rather than stat to determine the file size because
440 that way we can also verify that the file is readable without
441 explicitly checking for permissions. Inspired by the POST patch
443 FILE *fp = fopen (filename, "rb");
446 fseeko (fp, 0, SEEK_END);
452 if (stat (filename, &st) < 0)
458 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
459 doesn't exist is found. Return a freshly allocated copy of the
463 unique_name_1 (const char *prefix)
466 int plen = strlen (prefix);
467 char *template = (char *)alloca (plen + 1 + 24);
468 char *template_tail = template + plen;
470 memcpy (template, prefix, plen);
471 *template_tail++ = '.';
474 number_to_string (template_tail, count++);
475 while (file_exists_p (template));
477 return xstrdup (template);
480 /* Return a unique file name, based on FILE.
482 More precisely, if FILE doesn't exist, it is returned unmodified.
483 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
484 file name that doesn't exist is returned.
486 The resulting file is not created, only verified that it didn't
487 exist at the point in time when the function was called.
488 Therefore, where security matters, don't rely that the file created
489 by this function exists until you open it with O_EXCL or
492 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
493 string. Otherwise, it may return FILE if the file doesn't exist
494 (and therefore doesn't need changing). */
497 unique_name (const char *file, bool allow_passthrough)
499 /* If the FILE itself doesn't exist, return it without
501 if (!file_exists_p (file))
502 return allow_passthrough ? (char *)file : xstrdup (file);
504 /* Otherwise, find a numeric suffix that results in unused file name
506 return unique_name_1 (file);
509 /* Create a file based on NAME, except without overwriting an existing
510 file with that name. Providing O_EXCL is correctly implemented,
511 this function does not have the race condition associated with
512 opening the file returned by unique_name. */
515 unique_create (const char *name, bool binary, char **opened_name)
517 /* unique file name, based on NAME */
518 char *uname = unique_name (name, false);
520 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
523 uname = unique_name (name, false);
525 if (opened_name && fp != NULL)
528 *opened_name = uname;
540 /* Open the file for writing, with the addition that the file is
541 opened "exclusively". This means that, if the file already exists,
542 this function will *fail* and errno will be set to EEXIST. If
543 BINARY is set, the file will be opened in binary mode, equivalent
546 If opening the file fails for any reason, including the file having
547 previously existed, this function returns NULL and sets errno
551 fopen_excl (const char *fname, bool binary)
555 int flags = O_WRONLY | O_CREAT | O_EXCL;
560 fd = open (fname, flags, 0666);
563 return fdopen (fd, binary ? "wb" : "w");
564 #else /* not O_EXCL */
565 /* Manually check whether the file exists. This is prone to race
566 conditions, but systems without O_EXCL haven't deserved
568 if (file_exists_p (fname))
573 return fopen (fname, binary ? "wb" : "w");
574 #endif /* not O_EXCL */
577 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
578 are missing, create them first. In case any mkdir() call fails,
579 return its error status. Returns 0 on successful completion.
581 The behaviour of this function should be identical to the behaviour
582 of `mkdir -p' on systems where mkdir supports the `-p' option. */
584 make_directory (const char *directory)
586 int i, ret, quit = 0;
589 /* Make a copy of dir, to be able to write to it. Otherwise, the
590 function is unsafe if called with a read-only char *argument. */
591 STRDUP_ALLOCA (dir, directory);
593 /* If the first character of dir is '/', skip it (and thus enable
594 creation of absolute-pathname directories. */
595 for (i = (*dir == '/'); 1; ++i)
597 for (; dir[i] && dir[i] != '/'; i++)
602 /* Check whether the directory already exists. Allow creation of
603 of intermediate directories to fail, as the initial path components
604 are not necessarily directories! */
605 if (!file_exists_p (dir))
606 ret = mkdir (dir, 0777);
617 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
618 should be a file name.
620 file_merge("/foo/bar", "baz") => "/foo/baz"
621 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
622 file_merge("foo", "bar") => "bar"
624 In other words, it's a simpler and gentler version of uri_merge. */
627 file_merge (const char *base, const char *file)
630 const char *cut = (const char *)strrchr (base, '/');
633 return xstrdup (file);
635 result = xmalloc (cut - base + 1 + strlen (file) + 1);
636 memcpy (result, base, cut - base);
637 result[cut - base] = '/';
638 strcpy (result + (cut - base) + 1, file);
643 /* Like fnmatch, but performs a case-insensitive match. */
646 fnmatch_nocase (const char *pattern, const char *string, int flags)
649 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
650 also present on *BSD platforms, and possibly elsewhere. */
651 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
653 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
654 char *patcopy = (char *) alloca (strlen (pattern) + 1);
655 char *strcopy = (char *) alloca (strlen (string) + 1);
657 for (p = patcopy; *pattern; pattern++, p++)
658 *p = c_tolower (*pattern);
660 for (p = strcopy; *string; string++, p++)
661 *p = c_tolower (*string);
663 return fnmatch (patcopy, strcopy, flags);
667 static bool in_acclist (const char *const *, const char *, bool);
669 /* Determine whether a file is acceptable to be followed, according to
670 lists of patterns to accept/reject. */
672 acceptable (const char *s)
676 while (l && s[l] != '/')
683 return (in_acclist ((const char *const *)opt.accepts, s, true)
684 && !in_acclist ((const char *const *)opt.rejects, s, true));
686 return in_acclist ((const char *const *)opt.accepts, s, true);
688 else if (opt.rejects)
689 return !in_acclist ((const char *const *)opt.rejects, s, true);
693 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
694 will return true if and only if D2 begins with `/something/' or is exactly
697 subdir_p (const char *d1, const char *d2)
701 if (!opt.ignore_case)
702 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
705 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
708 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
711 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
712 first element that matches DIR, through wildcards or front comparison (as
715 dir_matches_p (char **dirlist, const char *dir)
718 int (*matcher) (const char *, const char *, int)
719 = opt.ignore_case ? fnmatch_nocase : fnmatch;
721 for (x = dirlist; *x; x++)
723 /* Remove leading '/' */
724 char *p = *x + (**x == '/');
725 if (has_wildcards_p (p))
727 if (matcher (p, dir, FNM_PATHNAME) == 0)
732 if (subdir_p (p, dir))
737 return *x ? true : false;
740 /* Returns whether DIRECTORY is acceptable for download, wrt the
741 include/exclude lists.
743 The leading `/' is ignored in paths; relative and absolute paths
744 may be freely intermixed. */
747 accdir (const char *directory)
749 /* Remove starting '/'. */
750 if (*directory == '/')
754 if (!dir_matches_p (opt.includes, directory))
759 if (dir_matches_p (opt.excludes, directory))
765 /* Return true if STRING ends with TAIL. For instance:
767 match_tail ("abc", "bc", false) -> 1
768 match_tail ("abc", "ab", false) -> 0
769 match_tail ("abc", "abc", false) -> 1
771 If FOLD_CASE is true, the comparison will be case-insensitive. */
774 match_tail (const char *string, const char *tail, bool fold_case)
778 /* We want this to be fast, so we code two loops, one with
779 case-folding, one without. */
783 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
784 if (string[i] != tail[j])
789 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
790 if (c_tolower (string[i]) != c_tolower (tail[j]))
794 /* If the tail was exhausted, the match was succesful. */
801 /* Checks whether string S matches each element of ACCEPTS. A list
802 element are matched either with fnmatch() or match_tail(),
803 according to whether the element contains wildcards or not.
805 If the BACKWARD is false, don't do backward comparison -- just compare
808 in_acclist (const char *const *accepts, const char *s, bool backward)
810 for (; *accepts; accepts++)
812 if (has_wildcards_p (*accepts))
814 int res = opt.ignore_case
815 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
816 /* fnmatch returns 0 if the pattern *does* match the string. */
824 if (match_tail (s, *accepts, opt.ignore_case))
829 int cmp = opt.ignore_case
830 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
839 /* Return the location of STR's suffix (file extension). Examples:
840 suffix ("foo.bar") -> "bar"
841 suffix ("foo.bar.baz") -> "baz"
842 suffix ("/foo/bar") -> NULL
843 suffix ("/foo.bar/baz") -> NULL */
845 suffix (const char *str)
849 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
853 return (char *)str + i;
858 /* Return true if S contains globbing wildcards (`*', `?', `[' or
862 has_wildcards_p (const char *s)
865 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
870 /* Return true if FNAME ends with a typical HTML suffix. The
871 following (case-insensitive) suffixes are presumed to be HTML
876 ?html (`?' matches one character)
878 #### CAVEAT. This is not necessarily a good indication that FNAME
879 refers to a file that contains HTML! */
881 has_html_suffix_p (const char *fname)
885 if ((suf = suffix (fname)) == NULL)
887 if (!strcasecmp (suf, "html"))
889 if (!strcasecmp (suf, "htm"))
891 if (suf[0] && !strcasecmp (suf + 1, "html"))
896 /* Read a line from FP and return the pointer to freshly allocated
897 storage. The storage space is obtained through malloc() and should
898 be freed with free() when it is no longer needed.
900 The length of the line is not limited, except by available memory.
901 The newline character at the end of line is retained. The line is
902 terminated with a zero character.
904 After end-of-file is encountered without anything being read, NULL
905 is returned. NULL is also returned on error. To distinguish
906 between these two cases, use the stdio function ferror(). */
909 read_whole_line (FILE *fp)
913 char *line = xmalloc (bufsize);
915 while (fgets (line + length, bufsize - length, fp))
917 length += strlen (line + length);
919 /* Possible for example when reading from a binary file where
920 a line begins with \0. */
923 if (line[length - 1] == '\n')
926 /* fgets() guarantees to read the whole line, or to use up the
927 space we've given it. We can double the buffer
930 line = xrealloc (line, bufsize);
932 if (length == 0 || ferror (fp))
937 if (length + 1 < bufsize)
938 /* Relieve the memory from our exponential greediness. We say
939 `length + 1' because the terminating \0 is not included in
940 LENGTH. We don't need to zero-terminate the string ourselves,
941 though, because fgets() does that. */
942 line = xrealloc (line, length + 1);
946 /* Read FILE into memory. A pointer to `struct file_memory' are
947 returned; use struct element `content' to access file contents, and
948 the element `length' to know the file length. `content' is *not*
949 zero-terminated, and you should *not* read or write beyond the [0,
950 length) range of characters.
952 After you are done with the file contents, call read_file_free to
955 Depending on the operating system and the type of file that is
956 being read, read_file() either mmap's the file into memory, or
957 reads the file into the core using read().
959 If file is named "-", fileno(stdin) is used for reading instead.
960 If you want to read from a real file named "-", use "./-" instead. */
963 read_file (const char *file)
966 struct file_memory *fm;
968 bool inhibit_close = false;
970 /* Some magic in the finest tradition of Perl and its kin: if FILE
971 is "-", just use stdin. */
975 inhibit_close = true;
976 /* Note that we don't inhibit mmap() in this case. If stdin is
977 redirected from a regular file, mmap() will still work. */
980 fd = open (file, O_RDONLY);
983 fm = xnew (struct file_memory);
988 if (fstat (fd, &buf) < 0)
990 fm->length = buf.st_size;
991 /* NOTE: As far as I know, the callers of this function never
992 modify the file text. Relying on this would enable us to
993 specify PROT_READ and MAP_SHARED for a marginal gain in
994 efficiency, but at some cost to generality. */
995 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
997 if (fm->content == (char *)MAP_FAILED)
1007 /* The most common reason why mmap() fails is that FD does not point
1008 to a plain file. However, it's also possible that mmap() doesn't
1009 work for a particular type of file. Therefore, whenever mmap()
1010 fails, we just fall back to the regular method. */
1011 #endif /* HAVE_MMAP */
1014 size = 512; /* number of bytes fm->contents can
1015 hold at any given time. */
1016 fm->content = xmalloc (size);
1020 if (fm->length > size / 2)
1022 /* #### I'm not sure whether the whole exponential-growth
1023 thing makes sense with kernel read. On Linux at least,
1024 read() refuses to read more than 4K from a file at a
1025 single chunk anyway. But other Unixes might optimize it
1026 better, and it doesn't *hurt* anything, so I'm leaving
1029 /* Normally, we grow SIZE exponentially to make the number
1030 of calls to read() and realloc() logarithmic in relation
1031 to file size. However, read() can read an amount of data
1032 smaller than requested, and it would be unreasonable to
1033 double SIZE every time *something* was read. Therefore,
1034 we double SIZE only when the length exceeds half of the
1035 entire allocated size. */
1037 fm->content = xrealloc (fm->content, size);
1039 nread = read (fd, fm->content + fm->length, size - fm->length);
1041 /* Successful read. */
1042 fm->length += nread;
1052 if (size > fm->length && fm->length != 0)
1053 /* Due to exponential growth of fm->content, the allocated region
1054 might be much larger than what is actually needed. */
1055 fm->content = xrealloc (fm->content, fm->length);
1062 xfree (fm->content);
1067 /* Release the resources held by FM. Specifically, this calls
1068 munmap() or xfree() on fm->content, depending whether mmap or
1069 malloc/read were used to read in the file. It also frees the
1070 memory needed to hold the FM structure itself. */
1073 read_file_free (struct file_memory *fm)
1078 munmap (fm->content, fm->length);
1083 xfree (fm->content);
1088 /* Free the pointers in a NULL-terminated vector of pointers, then
1089 free the pointer itself. */
1091 free_vec (char **vec)
1102 /* Append vector V2 to vector V1. The function frees V2 and
1103 reallocates V1 (thus you may not use the contents of neither
1104 pointer after the call). If V1 is NULL, V2 is returned. */
1106 merge_vecs (char **v1, char **v2)
1116 /* To avoid j == 0 */
1121 for (i = 0; v1[i]; i++)
1124 for (j = 0; v2[j]; j++)
1126 /* Reallocate v1. */
1127 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1128 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1133 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1134 is allocated as needed. Return the new value of the vector. */
1137 vec_append (char **vec, const char *str)
1139 int cnt; /* count of vector elements, including
1140 the one we're about to append */
1143 for (cnt = 0; vec[cnt]; cnt++)
1149 /* Reallocate the array to fit the new element and the NULL. */
1150 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1151 /* Append a copy of STR to the vector. */
1152 vec[cnt - 1] = xstrdup (str);
1157 /* Sometimes it's useful to create "sets" of strings, i.e. special
1158 hash tables where you want to store strings as keys and merely
1159 query for their existence. Here is a set of utility routines that
1160 makes that transparent. */
1163 string_set_add (struct hash_table *ht, const char *s)
1165 /* First check whether the set element already exists. If it does,
1166 do nothing so that we don't have to free() the old element and
1167 then strdup() a new one. */
1168 if (hash_table_contains (ht, s))
1171 /* We use "1" as value. It provides us a useful and clear arbitrary
1172 value, and it consumes no memory -- the pointers to the same
1173 string "1" will be shared by all the key-value pairs in all `set'
1175 hash_table_put (ht, xstrdup (s), "1");
1178 /* Synonym for hash_table_contains... */
1181 string_set_contains (struct hash_table *ht, const char *s)
1183 return hash_table_contains (ht, s);
1186 /* Convert the specified string set to array. ARRAY should be large
1187 enough to hold hash_table_count(ht) char pointers. */
1189 void string_set_to_array (struct hash_table *ht, char **array)
1191 hash_table_iterator iter;
1192 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1193 *array++ = iter.key;
1196 /* Free the string set. This frees both the storage allocated for
1197 keys and the actual hash table. (hash_table_destroy would only
1198 destroy the hash table.) */
1201 string_set_free (struct hash_table *ht)
1203 hash_table_iterator iter;
1204 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1206 hash_table_destroy (ht);
1209 /* Utility function: simply call xfree() on all keys and values of HT. */
1212 free_keys_and_values (struct hash_table *ht)
1214 hash_table_iterator iter;
1215 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1222 /* Get digit grouping data for thousand separors by calling
1223 localeconv(). The data includes separator string and grouping info
1224 and is cached after the first call to the function.
1226 In locales that don't set a thousand separator (such as the "C"
1227 locale), this forces it to be ",". We are now only showing
1228 thousand separators in one place, so this shouldn't be a problem in
1232 get_grouping_data (const char **sep, const char **grouping)
1234 static const char *cached_sep;
1235 static const char *cached_grouping;
1236 static bool initialized;
1239 /* Get the grouping info from the locale. */
1240 struct lconv *lconv = localeconv ();
1241 cached_sep = lconv->thousands_sep;
1242 cached_grouping = lconv->grouping;
1243 #if ! USE_NLS_PROGRESS_BAR
1244 /* We can't count column widths, so ensure that the separator
1245 * is single-byte only (let check below determine what byte). */
1246 if (strlen(cached_sep) > 1)
1251 /* Many locales (such as "C" or "hr_HR") don't specify
1252 grouping, which we still want to use it for legibility.
1253 In those locales set the sep char to ',', unless that
1254 character is used for decimal point, in which case set it
1256 if (*lconv->decimal_point != ',')
1260 cached_grouping = "\x03";
1265 *grouping = cached_grouping;
1268 /* Return a printed representation of N with thousand separators.
1269 This should respect locale settings, with the exception of the "C"
1270 locale which mandates no separator, but we use one anyway.
1272 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1273 the separators because it's too non-portable, and it's hard to test
1274 for this feature at configure time. Besides, it wouldn't display
1275 separators in the "C" locale, still used by many Unix users. */
1278 with_thousand_seps (wgint n)
1280 static char outbuf[48];
1281 char *p = outbuf + sizeof outbuf;
1283 /* Info received from locale */
1284 const char *grouping, *sep;
1287 /* State information */
1288 int i = 0, groupsize;
1289 const char *atgroup;
1291 bool negative = n < 0;
1293 /* Initialize grouping data. */
1294 get_grouping_data (&sep, &grouping);
1295 seplen = strlen (sep);
1297 groupsize = *atgroup++;
1299 /* This would overflow on WGINT_MIN, but printing negative numbers
1300 is not an important goal of this fuinction. */
1304 /* Write the number into the buffer, backwards, inserting the
1305 separators as necessary. */
1309 *--p = n % 10 + '0';
1313 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1314 if (++i == groupsize)
1319 memcpy (p -= seplen, sep, seplen);
1322 groupsize = *atgroup++;
1331 /* N, a byte quantity, is converted to a human-readable abberviated
1332 form a la sizes printed by `ls -lh'. The result is written to a
1333 static buffer, a pointer to which is returned.
1335 Unlike `with_thousand_seps', this approximates to the nearest unit.
1336 Quoting GNU libit: "Most people visually process strings of 3-4
1337 digits effectively, but longer strings of digits are more prone to
1338 misinterpretation. Hence, converting to an abbreviated form
1339 usually improves readability."
1341 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1342 original computer-related meaning of "powers of 1024". We don't
1343 use the "*bibyte" names invented in 1998, and seldom used in
1344 practice. Wikipedia's entry on "binary prefix" discusses this in
1348 human_readable (HR_NUMTYPE n)
1350 /* These suffixes are compatible with those of GNU `ls -lh'. */
1351 static char powers[] =
1353 'K', /* kilobyte, 2^10 bytes */
1354 'M', /* megabyte, 2^20 bytes */
1355 'G', /* gigabyte, 2^30 bytes */
1356 'T', /* terabyte, 2^40 bytes */
1357 'P', /* petabyte, 2^50 bytes */
1358 'E', /* exabyte, 2^60 bytes */
1363 /* If the quantity is smaller than 1K, just print it. */
1366 snprintf (buf, sizeof (buf), "%d", (int) n);
1370 /* Loop over powers, dividing N with 1024 in each iteration. This
1371 works unchanged for all sizes of wgint, while still avoiding
1372 non-portable `long double' arithmetic. */
1373 for (i = 0; i < countof (powers); i++)
1375 /* At each iteration N is greater than the *subsequent* power.
1376 That way N/1024.0 produces a decimal number in the units of
1378 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1380 double val = n / 1024.0;
1381 /* Print values smaller than 10 with one decimal digits, and
1382 others without any decimals. */
1383 snprintf (buf, sizeof (buf), "%.*f%c",
1384 val < 10 ? 1 : 0, val, powers[i]);
1389 return NULL; /* unreached */
1392 /* Count the digits in the provided number. Used to allocate space
1393 when printing numbers. */
1396 numdigit (wgint number)
1400 ++cnt; /* accomodate '-' */
1401 while ((number /= 10) != 0)
1406 #define PR(mask) *p++ = n / (mask) + '0'
1408 /* DIGITS_<D> is used to print a D-digit number and should be called
1409 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1410 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1411 Recursively this continues until DIGITS_1 is invoked. */
1413 #define DIGITS_1(mask) PR (mask)
1414 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1415 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1416 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1417 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1418 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1419 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1420 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1421 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1422 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1424 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1426 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1427 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1428 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1429 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1430 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1431 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1432 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1433 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1434 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1436 /* Shorthand for casting to wgint. */
1439 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1440 `sprintf(buffer, "%lld", (long long) number)', only typically much
1441 faster and portable to machines without long long.
1443 The speedup may make a difference in programs that frequently
1444 convert numbers to strings. Some implementations of sprintf,
1445 particularly the one in some versions of GNU libc, have been known
1446 to be quite slow when converting integers to strings.
1448 Return the pointer to the location where the terminating zero was
1449 printed. (Equivalent to calling buffer+strlen(buffer) after the
1452 BUFFER should be large enough to accept as many bytes as you expect
1453 the number to take up. On machines with 64-bit wgints the maximum
1454 needed size is 24 bytes. That includes the digits needed for the
1455 largest 64-bit number, the `-' sign in case it's negative, and the
1456 terminating '\0'. */
1459 number_to_string (char *buffer, wgint number)
1464 int last_digit_char = 0;
1466 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1467 /* We are running in a very strange environment. Leave the correct
1468 printing to sprintf. */
1469 p += sprintf (buf, "%j", (intmax_t) (n));
1470 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1476 /* n = -n would overflow because -n would evaluate to a
1477 wgint value larger than WGINT_MAX. Need to make n
1478 smaller and handle the last digit separately. */
1479 int last_digit = n % 10;
1480 /* The sign of n%10 is implementation-defined. */
1482 last_digit_char = '0' - last_digit;
1484 last_digit_char = '0' + last_digit;
1485 /* After n is made smaller, -n will not overflow. */
1493 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1494 way printing any N is fully open-coded without a loop or jump.
1495 (Also see description of DIGITS_*.) */
1497 if (n < 10) DIGITS_1 (1);
1498 else if (n < 100) DIGITS_2 (10);
1499 else if (n < 1000) DIGITS_3 (100);
1500 else if (n < 10000) DIGITS_4 (1000);
1501 else if (n < 100000) DIGITS_5 (10000);
1502 else if (n < 1000000) DIGITS_6 (100000);
1503 else if (n < 10000000) DIGITS_7 (1000000);
1504 else if (n < 100000000) DIGITS_8 (10000000);
1505 else if (n < 1000000000) DIGITS_9 (100000000);
1506 #if SIZEOF_WGINT == 4
1507 /* wgint is 32 bits wide: no number has more than 10 digits. */
1508 else DIGITS_10 (1000000000);
1510 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1511 Constants are constructed by compile-time multiplication to avoid
1512 dealing with different notations for 64-bit constants
1513 (nL/nLL/nI64, depending on the compiler and architecture). */
1514 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1515 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1516 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1517 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1518 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1519 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1520 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1521 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1522 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1523 else DIGITS_19 (1000000000*(W)1000000000);
1526 if (last_digit_char)
1527 *p++ = last_digit_char;
1530 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1537 #undef SPRINTF_WGINT
1560 /* Print NUMBER to a statically allocated string and return a pointer
1561 to the printed representation.
1563 This function is intended to be used in conjunction with printf.
1564 It is hard to portably print wgint values:
1565 a) you cannot use printf("%ld", number) because wgint can be long
1566 long on 32-bit machines with LFS.
1567 b) you cannot use printf("%lld", number) because NUMBER could be
1568 long on 32-bit machines without LFS, or on 64-bit machines,
1569 which do not require LFS. Also, Windows doesn't support %lld.
1570 c) you cannot use printf("%j", (int_max_t) number) because not all
1571 versions of printf support "%j", the most notable being the one
1573 d) you cannot #define WGINT_FMT to the appropriate format and use
1574 printf(WGINT_FMT, number) because that would break translations
1575 for user-visible messages, such as printf("Downloaded: %d
1578 What you should use instead is printf("%s", number_to_static_string
1581 CAVEAT: since the function returns pointers to static data, you
1582 must be careful to copy its result before calling it again.
1583 However, to make it more useful with printf, the function maintains
1584 an internal ring of static buffers to return. That way things like
1585 printf("%s %s", number_to_static_string (num1),
1586 number_to_static_string (num2)) work as expected. Three buffers
1587 are currently used, which means that "%s %s %s" will work, but "%s
1588 %s %s %s" won't. If you need to print more than three wgints,
1589 bump the RING_SIZE (or rethink your message.) */
1592 number_to_static_string (wgint number)
1594 static char ring[RING_SIZE][24];
1596 char *buf = ring[ringpos];
1597 number_to_string (buf, number);
1598 ringpos = (ringpos + 1) % RING_SIZE;
1602 /* Determine the width of the terminal we're running on. If that's
1603 not possible, return 0. */
1606 determine_screen_width (void)
1608 /* If there's a way to get the terminal size using POSIX
1609 tcgetattr(), somebody please tell me. */
1614 if (opt.lfilename != NULL)
1617 fd = fileno (stderr);
1618 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1619 return 0; /* most likely ENOTTY */
1622 #elif defined(WINDOWS)
1623 CONSOLE_SCREEN_BUFFER_INFO csbi;
1624 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1626 return csbi.dwSize.X;
1627 #else /* neither TIOCGWINSZ nor WINDOWS */
1629 #endif /* neither TIOCGWINSZ nor WINDOWS */
1632 /* Whether the rnd system (either rand or [dl]rand48) has been
1634 static int rnd_seeded;
1636 /* Return a random number between 0 and MAX-1, inclusive.
1638 If the system does not support lrand48 and MAX is greater than the
1639 value of RAND_MAX+1 on the system, the returned value will be in
1640 the range [0, RAND_MAX]. This may be fixed in a future release.
1641 The random number generator is seeded automatically the first time
1644 This uses lrand48 where available, rand elsewhere. DO NOT use it
1645 for cryptography. It is only meant to be used in situations where
1646 quality of the random numbers returned doesn't really matter. */
1649 random_number (int max)
1654 srand48 ((long) time (NULL) ^ (long) getpid ());
1657 return lrand48 () % max;
1658 #else /* not HAVE_DRAND48 */
1664 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1669 /* Like rand() % max, but uses the high-order bits for better
1670 randomness on architectures where rand() is implemented using a
1671 simple congruential generator. */
1673 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1674 return (int) bounded;
1676 #endif /* not HAVE_DRAND48 */
1679 /* Return a random uniformly distributed floating point number in the
1680 [0, 1) range. Uses drand48 where available, and a really lame
1681 kludge elsewhere. */
1689 srand48 ((long) time (NULL) ^ (long) getpid ());
1693 #else /* not HAVE_DRAND48 */
1694 return ( random_number (10000) / 10000.0
1695 + random_number (10000) / (10000.0 * 10000.0)
1696 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1697 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1698 #endif /* not HAVE_DRAND48 */
1701 /* Implementation of run_with_timeout, a generic timeout-forcing
1702 routine for systems with Unix-like signal handling. */
1704 #ifdef USE_SIGNAL_TIMEOUT
1705 # ifdef HAVE_SIGSETJMP
1706 # define SETJMP(env) sigsetjmp (env, 1)
1708 static sigjmp_buf run_with_timeout_env;
1711 abort_run_with_timeout (int sig)
1713 assert (sig == SIGALRM);
1714 siglongjmp (run_with_timeout_env, -1);
1716 # else /* not HAVE_SIGSETJMP */
1717 # define SETJMP(env) setjmp (env)
1719 static jmp_buf run_with_timeout_env;
1722 abort_run_with_timeout (int sig)
1724 assert (sig == SIGALRM);
1725 /* We don't have siglongjmp to preserve the set of blocked signals;
1726 if we longjumped out of the handler at this point, SIGALRM would
1727 remain blocked. We must unblock it manually. */
1728 int mask = siggetmask ();
1729 mask &= ~sigmask (SIGALRM);
1732 /* Now it's safe to longjump. */
1733 longjmp (run_with_timeout_env, -1);
1735 # endif /* not HAVE_SIGSETJMP */
1737 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1738 setitimer where available, alarm otherwise.
1740 TIMEOUT should be non-zero. If the timeout value is so small that
1741 it would be rounded to zero, it is rounded to the least legal value
1742 instead (1us for setitimer, 1s for alarm). That ensures that
1743 SIGALRM will be delivered in all cases. */
1746 alarm_set (double timeout)
1749 /* Use the modern itimer interface. */
1750 struct itimerval itv;
1752 itv.it_value.tv_sec = (long) timeout;
1753 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1754 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1755 /* Ensure that we wait for at least the minimum interval.
1756 Specifying zero would mean "wait forever". */
1757 itv.it_value.tv_usec = 1;
1758 setitimer (ITIMER_REAL, &itv, NULL);
1759 #else /* not ITIMER_REAL */
1760 /* Use the old alarm() interface. */
1761 int secs = (int) timeout;
1763 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1764 because alarm(0) means "never deliver the alarm", i.e. "wait
1765 forever", which is not what someone who specifies a 0.5s
1766 timeout would expect. */
1769 #endif /* not ITIMER_REAL */
1772 /* Cancel the alarm set with alarm_set. */
1778 struct itimerval disable;
1780 setitimer (ITIMER_REAL, &disable, NULL);
1781 #else /* not ITIMER_REAL */
1783 #endif /* not ITIMER_REAL */
1786 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1787 seconds. Returns true if the function was interrupted with a
1788 timeout, false otherwise.
1790 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1791 using setitimer() or alarm(). The timeout is enforced by
1792 longjumping out of the SIGALRM handler. This has several
1793 advantages compared to the traditional approach of relying on
1794 signals causing system calls to exit with EINTR:
1796 * The callback function is *forcibly* interrupted after the
1797 timeout expires, (almost) regardless of what it was doing and
1798 whether it was in a syscall. For example, a calculation that
1799 takes a long time is interrupted as reliably as an IO
1802 * It works with both SYSV and BSD signals because it doesn't
1803 depend on the default setting of SA_RESTART.
1805 * It doesn't require special handler setup beyond a simple call
1806 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1809 The only downside is that, if FUN allocates internal resources that
1810 are normally freed prior to exit from the functions, they will be
1811 lost in case of timeout. */
1814 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1824 signal (SIGALRM, abort_run_with_timeout);
1825 if (SETJMP (run_with_timeout_env) != 0)
1827 /* Longjumped out of FUN with a timeout. */
1828 signal (SIGALRM, SIG_DFL);
1831 alarm_set (timeout);
1834 /* Preserve errno in case alarm() or signal() modifies it. */
1835 saved_errno = errno;
1837 signal (SIGALRM, SIG_DFL);
1838 errno = saved_errno;
1843 #else /* not USE_SIGNAL_TIMEOUT */
1846 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1847 define it under Windows, because Windows has its own version of
1848 run_with_timeout that uses threads. */
1851 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1856 #endif /* not WINDOWS */
1857 #endif /* not USE_SIGNAL_TIMEOUT */
1861 /* Sleep the specified amount of seconds. On machines without
1862 nanosleep(), this may sleep shorter if interrupted by signals. */
1865 xsleep (double seconds)
1867 #ifdef HAVE_NANOSLEEP
1868 /* nanosleep is the preferred interface because it offers high
1869 accuracy and, more importantly, because it allows us to reliably
1870 restart receiving a signal such as SIGWINCH. (There was an
1871 actual Debian bug report about --limit-rate malfunctioning while
1872 the terminal was being resized.) */
1873 struct timespec sleep, remaining;
1874 sleep.tv_sec = (long) seconds;
1875 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1876 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1877 /* If nanosleep has been interrupted by a signal, adjust the
1878 sleeping period and return to sleep. */
1880 #elif defined(HAVE_USLEEP)
1881 /* If usleep is available, use it in preference to select. */
1884 /* On some systems, usleep cannot handle values larger than
1885 1,000,000. If the period is larger than that, use sleep
1886 first, then add usleep for subsecond accuracy. */
1888 seconds -= (long) seconds;
1890 usleep (seconds * 1000000);
1891 #else /* fall back select */
1892 /* Note that, although Windows supports select, it can't be used to
1893 implement sleeping because Winsock's select doesn't implement
1894 timeout when it is passed NULL pointers for all fd sets. (But it
1895 does under Cygwin, which implements Unix-compatible select.) */
1896 struct timeval sleep;
1897 sleep.tv_sec = (long) seconds;
1898 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1899 select (0, NULL, NULL, NULL, &sleep);
1900 /* If select returns -1 and errno is EINTR, it means we were
1901 interrupted by a signal. But without knowing how long we've
1902 actually slept, we can't return to sleep. Using gettimeofday to
1903 track sleeps is slow and unreliable due to clock skew. */
1907 #endif /* not WINDOWS */
1909 /* Encode the octets in DATA of length LENGTH to base64 format,
1910 storing the result to DEST. The output will be zero-terminated,
1911 and must point to a writable buffer of at least
1912 1+BASE64_LENGTH(length) bytes. The function returns the length of
1913 the resulting base64 data, not counting the terminating zero.
1915 This implementation does not emit newlines after 76 characters of
1919 base64_encode (const void *data, int length, char *dest)
1921 /* Conversion table. */
1922 static const char tbl[64] = {
1923 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
1924 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
1925 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
1926 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
1928 /* Access bytes in DATA as unsigned char, otherwise the shifts below
1929 don't work for data with MSB set. */
1930 const unsigned char *s = data;
1931 /* Theoretical ANSI violation when length < 3. */
1932 const unsigned char *end = (const unsigned char *) data + length - 2;
1935 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1936 for (; s < end; s += 3)
1938 *p++ = tbl[s[0] >> 2];
1939 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1940 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1941 *p++ = tbl[s[2] & 0x3f];
1944 /* Pad the result if necessary... */
1948 *p++ = tbl[s[0] >> 2];
1949 *p++ = tbl[(s[0] & 3) << 4];
1954 *p++ = tbl[s[0] >> 2];
1955 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1956 *p++ = tbl[((s[1] & 0xf) << 2)];
1960 /* ...and zero-terminate it. */
1966 /* Store in C the next non-whitespace character from the string, or \0
1967 when end of string is reached. */
1968 #define NEXT_CHAR(c, p) do { \
1969 c = (unsigned char) *p++; \
1970 } while (c_isspace (c))
1972 #define IS_ASCII(c) (((c) & 0x80) == 0)
1974 /* Decode data from BASE64 (a null-terminated string) into memory
1975 pointed to by DEST. DEST is assumed to be large enough to
1976 accomodate the decoded data, which is guaranteed to be no more than
1979 Since DEST is assumed to contain binary data, it is not
1980 NUL-terminated. The function returns the length of the data
1981 written to TO. -1 is returned in case of error caused by malformed
1984 This function originates from Free Recode. */
1987 base64_decode (const char *base64, void *dest)
1989 /* Table of base64 values for first 128 characters. Note that this
1990 assumes ASCII (but so does Wget in other places). */
1991 static const signed char base64_char_to_value[128] =
1993 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1994 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1995 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1996 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1997 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1998 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1999 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2000 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2001 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2002 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2003 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2004 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2005 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2007 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2008 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2010 const char *p = base64;
2016 unsigned long value;
2018 /* Process first byte of a quadruplet. */
2022 if (c == '=' || !IS_BASE64 (c))
2023 return -1; /* illegal char while decoding base64 */
2024 value = BASE64_CHAR_TO_VALUE (c) << 18;
2026 /* Process second byte of a quadruplet. */
2029 return -1; /* premature EOF while decoding base64 */
2030 if (c == '=' || !IS_BASE64 (c))
2031 return -1; /* illegal char while decoding base64 */
2032 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2035 /* Process third byte of a quadruplet. */
2038 return -1; /* premature EOF while decoding base64 */
2040 return -1; /* illegal char while decoding base64 */
2046 return -1; /* premature EOF while decoding base64 */
2048 return -1; /* padding `=' expected but not found */
2052 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2053 *q++ = 0xff & value >> 8;
2055 /* Process fourth byte of a quadruplet. */
2058 return -1; /* premature EOF while decoding base64 */
2062 return -1; /* illegal char while decoding base64 */
2064 value |= BASE64_CHAR_TO_VALUE (c);
2065 *q++ = 0xff & value;
2068 #undef BASE64_CHAR_TO_VALUE
2070 return q - (char *) dest;
2076 /* Simple merge sort for use by stable_sort. Implementation courtesy
2077 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2080 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2081 int (*cmpfun) (const void *, const void *))
2083 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2087 size_t mid = (to + from) / 2;
2088 mergesort_internal (base, temp, size, from, mid, cmpfun);
2089 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2092 for (k = from; (i <= mid) && (j <= to); k++)
2093 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2094 memcpy (ELT (temp, k), ELT (base, i++), size);
2096 memcpy (ELT (temp, k), ELT (base, j++), size);
2098 memcpy (ELT (temp, k++), ELT (base, i++), size);
2100 memcpy (ELT (temp, k++), ELT (base, j++), size);
2101 for (k = from; k <= to; k++)
2102 memcpy (ELT (base, k), ELT (temp, k), size);
2107 /* Stable sort with interface exactly like standard library's qsort.
2108 Uses mergesort internally, allocating temporary storage with
2112 stable_sort (void *base, size_t nmemb, size_t size,
2113 int (*cmpfun) (const void *, const void *))
2117 void *temp = alloca (nmemb * size * sizeof (void *));
2118 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2122 /* Print a decimal number. If it is equal to or larger than ten, the
2123 number is rounded. Otherwise it is printed with one significant
2124 digit without trailing zeros and with no more than three fractional
2125 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2126 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2128 This is useful for displaying durations because it provides
2129 order-of-magnitude information without unnecessary clutter --
2130 long-running downloads are shown without the fractional part, and
2131 short ones still retain one significant digit. */
2134 print_decimal (double number)
2136 static char buf[32];
2137 double n = number >= 0 ? number : -number;
2140 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2141 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2142 snprintf (buf, sizeof buf, "%.0f", number);
2144 snprintf (buf, sizeof buf, "%.1f", number);
2145 else if (n >= 0.001)
2146 snprintf (buf, sizeof buf, "%.1g", number);
2147 else if (n >= 0.0005)
2148 /* round [0.0005, 0.001) to 0.001 */
2149 snprintf (buf, sizeof buf, "%.3f", number);
2151 /* print numbers close to 0 as 0, not 0.000 */
2168 { "/somedir", "/somedir", true },
2169 { "/somedir", "/somedir/d2", true },
2170 { "/somedir/d1", "/somedir", false },
2173 for (i = 0; i < countof(test_array); ++i)
2175 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2177 mu_assert ("test_subdir_p: wrong result",
2178 res == test_array[i].result);
2185 test_dir_matches_p()
2193 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2194 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2195 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2196 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2197 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2198 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2199 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2200 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2201 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2202 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2203 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2204 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2205 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2208 for (i = 0; i < countof(test_array); ++i)
2210 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2212 mu_assert ("test_dir_matches_p: wrong result",
2213 res == test_array[i].result);
2219 #endif /* TESTING */