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. */
163 aprintf (const char *fmt, ...)
165 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
170 va_start (args, fmt);
171 ret = vasprintf (&str, fmt, args);
173 if (ret < 0 && errno == ENOMEM)
174 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency with xmalloc/xrealloc */
178 #else /* not HAVE_VASPRINTF */
180 /* vasprintf is unavailable. snprintf into a small buffer and
181 resize it as necessary. */
183 char *str = xmalloc (size);
185 /* #### This code will infloop and eventually abort in xrealloc if
186 passed a FMT that causes snprintf to consistently return -1. */
193 va_start (args, fmt);
194 n = vsnprintf (str, size, fmt, args);
197 /* If the printing worked, return the string. */
198 if (n > -1 && n < size)
201 /* Else try again with a larger buffer. */
202 if (n > -1) /* C99 */
203 size = n + 1; /* precisely what is needed */
205 size <<= 1; /* twice the old size */
206 str = xrealloc (str, size);
208 #endif /* not HAVE_VASPRINTF */
211 /* Concatenate the NULL-terminated list of string arguments into
212 freshly allocated space. */
215 concat_strings (const char *str0, ...)
218 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
221 const char *next_str;
222 int total_length = 0;
225 /* Calculate the length of and allocate the resulting string. */
228 va_start (args, str0);
229 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
231 int len = strlen (next_str);
232 if (argcount < countof (saved_lengths))
233 saved_lengths[argcount++] = len;
237 p = ret = xmalloc (total_length + 1);
239 /* Copy the strings into the allocated space. */
242 va_start (args, str0);
243 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
246 if (argcount < countof (saved_lengths))
247 len = saved_lengths[argcount++];
249 len = strlen (next_str);
250 memcpy (p, next_str, len);
259 /* Format the provided time according to the specified format. The
260 format is a string with format elements supported by strftime. */
263 fmttime (time_t t, const char *fmt)
265 static char output[32];
266 struct tm *tm = localtime(&t);
269 if (!strftime(output, sizeof(output), fmt, tm))
274 /* Return pointer to a static char[] buffer in which zero-terminated
275 string-representation of TM (in form hh:mm:ss) is printed.
277 If TM is NULL, the current time will be used. */
282 return fmttime(t, "%H:%M:%S");
285 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
288 datetime_str (time_t t)
290 return fmttime(t, "%Y-%m-%d %H:%M:%S");
293 /* The Windows versions of the following two functions are defined in
294 mswindows.c. On MSDOS this function should never be called. */
296 #if !defined(WINDOWS) && !defined(MSDOS)
298 fork_to_background (void)
301 /* Whether we arrange our own version of opt.lfilename here. */
302 bool logfile_changed = false;
304 if (!opt.lfilename && (!opt.quiet || opt.server_response))
306 /* We must create the file immediately to avoid either a race
307 condition (which arises from using unique_name and failing to
308 use fopen_excl) or lying to the user about the log file name
309 (which arises from using unique_name, printing the name, and
310 using fopen_excl later on.) */
311 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
314 logfile_changed = true;
327 /* parent, no error */
329 printf (_("Continuing in background, pid %d.\n"), (int) pid);
331 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
332 exit (0); /* #### should we use _exit()? */
335 /* child: give up the privileges and keep running. */
337 freopen ("/dev/null", "r", stdin);
338 freopen ("/dev/null", "w", stdout);
339 freopen ("/dev/null", "w", stderr);
341 #endif /* !WINDOWS && !MSDOS */
343 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
344 specified with TM. The atime ("access time") is set to the current
348 touch (const char *file, time_t tm)
350 #ifdef HAVE_STRUCT_UTIMBUF
351 struct utimbuf times;
359 times.actime = time (NULL);
360 if (utime (file, ×) == -1)
361 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
364 /* Checks if FILE is a symbolic link, and removes it if it is. Does
365 nothing under MS-Windows. */
367 remove_link (const char *file)
372 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
374 DEBUGP (("Unlinking %s (symlink).\n", file));
377 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
378 file, strerror (errno));
383 /* Does FILENAME exist? This is quite a lousy implementation, since
384 it supplies no error codes -- only a yes-or-no answer. Thus it
385 will return that a file does not exist if, e.g., the directory is
386 unreadable. I don't mind it too much currently, though. The
387 proper way should, of course, be to have a third, error state,
388 other than true/false, but that would introduce uncalled-for
389 additional complexity to the callers. */
391 file_exists_p (const char *filename)
394 return access (filename, F_OK) >= 0;
397 return stat (filename, &buf) >= 0;
401 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
402 Returns 0 on error. */
404 file_non_directory_p (const char *path)
407 /* Use lstat() rather than stat() so that symbolic links pointing to
408 directories can be identified correctly. */
409 if (lstat (path, &buf) != 0)
411 return S_ISDIR (buf.st_mode) ? false : true;
414 /* Return the size of file named by FILENAME, or -1 if it cannot be
415 opened or seeked into. */
417 file_size (const char *filename)
419 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
421 /* We use fseek rather than stat to determine the file size because
422 that way we can also verify that the file is readable without
423 explicitly checking for permissions. Inspired by the POST patch
425 FILE *fp = fopen (filename, "rb");
428 fseeko (fp, 0, SEEK_END);
434 if (stat (filename, &st) < 0)
440 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
441 doesn't exist is found. Return a freshly allocated copy of the
445 unique_name_1 (const char *prefix)
448 int plen = strlen (prefix);
449 char *template = (char *)alloca (plen + 1 + 24);
450 char *template_tail = template + plen;
452 memcpy (template, prefix, plen);
453 *template_tail++ = '.';
456 number_to_string (template_tail, count++);
457 while (file_exists_p (template));
459 return xstrdup (template);
462 /* Return a unique file name, based on FILE.
464 More precisely, if FILE doesn't exist, it is returned unmodified.
465 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
466 file name that doesn't exist is returned.
468 The resulting file is not created, only verified that it didn't
469 exist at the point in time when the function was called.
470 Therefore, where security matters, don't rely that the file created
471 by this function exists until you open it with O_EXCL or
474 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
475 string. Otherwise, it may return FILE if the file doesn't exist
476 (and therefore doesn't need changing). */
479 unique_name (const char *file, bool allow_passthrough)
481 /* If the FILE itself doesn't exist, return it without
483 if (!file_exists_p (file))
484 return allow_passthrough ? (char *)file : xstrdup (file);
486 /* Otherwise, find a numeric suffix that results in unused file name
488 return unique_name_1 (file);
491 /* Create a file based on NAME, except without overwriting an existing
492 file with that name. Providing O_EXCL is correctly implemented,
493 this function does not have the race condition associated with
494 opening the file returned by unique_name. */
497 unique_create (const char *name, bool binary, char **opened_name)
499 /* unique file name, based on NAME */
500 char *uname = unique_name (name, false);
502 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
505 uname = unique_name (name, false);
507 if (opened_name && fp != NULL)
510 *opened_name = uname;
522 /* Open the file for writing, with the addition that the file is
523 opened "exclusively". This means that, if the file already exists,
524 this function will *fail* and errno will be set to EEXIST. If
525 BINARY is set, the file will be opened in binary mode, equivalent
528 If opening the file fails for any reason, including the file having
529 previously existed, this function returns NULL and sets errno
533 fopen_excl (const char *fname, bool binary)
537 int flags = O_WRONLY | O_CREAT | O_EXCL;
542 fd = open (fname, flags, 0666);
545 return fdopen (fd, binary ? "wb" : "w");
546 #else /* not O_EXCL */
547 /* Manually check whether the file exists. This is prone to race
548 conditions, but systems without O_EXCL haven't deserved
550 if (file_exists_p (fname))
555 return fopen (fname, binary ? "wb" : "w");
556 #endif /* not O_EXCL */
559 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
560 are missing, create them first. In case any mkdir() call fails,
561 return its error status. Returns 0 on successful completion.
563 The behaviour of this function should be identical to the behaviour
564 of `mkdir -p' on systems where mkdir supports the `-p' option. */
566 make_directory (const char *directory)
568 int i, ret, quit = 0;
571 /* Make a copy of dir, to be able to write to it. Otherwise, the
572 function is unsafe if called with a read-only char *argument. */
573 STRDUP_ALLOCA (dir, directory);
575 /* If the first character of dir is '/', skip it (and thus enable
576 creation of absolute-pathname directories. */
577 for (i = (*dir == '/'); 1; ++i)
579 for (; dir[i] && dir[i] != '/'; i++)
584 /* Check whether the directory already exists. Allow creation of
585 of intermediate directories to fail, as the initial path components
586 are not necessarily directories! */
587 if (!file_exists_p (dir))
588 ret = mkdir (dir, 0777);
599 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
600 should be a file name.
602 file_merge("/foo/bar", "baz") => "/foo/baz"
603 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
604 file_merge("foo", "bar") => "bar"
606 In other words, it's a simpler and gentler version of uri_merge. */
609 file_merge (const char *base, const char *file)
612 const char *cut = (const char *)strrchr (base, '/');
615 return xstrdup (file);
617 result = xmalloc (cut - base + 1 + strlen (file) + 1);
618 memcpy (result, base, cut - base);
619 result[cut - base] = '/';
620 strcpy (result + (cut - base) + 1, file);
625 /* Like fnmatch, but performs a case-insensitive match. */
628 fnmatch_nocase (const char *pattern, const char *string, int flags)
631 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
632 also present on *BSD platforms, and possibly elsewhere. */
633 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
635 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
636 char *patcopy = (char *) alloca (strlen (pattern) + 1);
637 char *strcopy = (char *) alloca (strlen (string) + 1);
639 for (p = patcopy; *pattern; pattern++, p++)
640 *p = c_tolower (*pattern);
642 for (p = strcopy; *string; string++, p++)
643 *p = c_tolower (*string);
645 return fnmatch (patcopy, strcopy, flags);
649 static bool in_acclist (const char *const *, const char *, bool);
651 /* Determine whether a file is acceptable to be followed, according to
652 lists of patterns to accept/reject. */
654 acceptable (const char *s)
658 while (l && s[l] != '/')
665 return (in_acclist ((const char *const *)opt.accepts, s, true)
666 && !in_acclist ((const char *const *)opt.rejects, s, true));
668 return in_acclist ((const char *const *)opt.accepts, s, true);
670 else if (opt.rejects)
671 return !in_acclist ((const char *const *)opt.rejects, s, true);
675 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
676 will return true if and only if D2 begins with `/something/' or is exactly
679 subdir_p (const char *d1, const char *d2)
683 if (!opt.ignore_case)
684 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
687 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
690 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
693 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
694 first element that matches DIR, through wildcards or front comparison (as
697 dir_matches_p (char **dirlist, const char *dir)
700 int (*matcher) (const char *, const char *, int)
701 = opt.ignore_case ? fnmatch_nocase : fnmatch;
703 for (x = dirlist; *x; x++)
705 /* Remove leading '/' */
706 char *p = *x + (**x == '/');
707 if (has_wildcards_p (p))
709 if (matcher (p, dir, FNM_PATHNAME) == 0)
714 if (subdir_p (p, dir))
719 return *x ? true : false;
722 /* Returns whether DIRECTORY is acceptable for download, wrt the
723 include/exclude lists.
725 The leading `/' is ignored in paths; relative and absolute paths
726 may be freely intermixed. */
729 accdir (const char *directory)
731 /* Remove starting '/'. */
732 if (*directory == '/')
736 if (!dir_matches_p (opt.includes, directory))
741 if (dir_matches_p (opt.excludes, directory))
747 /* Return true if STRING ends with TAIL. For instance:
749 match_tail ("abc", "bc", false) -> 1
750 match_tail ("abc", "ab", false) -> 0
751 match_tail ("abc", "abc", false) -> 1
753 If FOLD_CASE is true, the comparison will be case-insensitive. */
756 match_tail (const char *string, const char *tail, bool fold_case)
760 /* We want this to be fast, so we code two loops, one with
761 case-folding, one without. */
765 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
766 if (string[i] != tail[j])
771 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
772 if (c_tolower (string[i]) != c_tolower (tail[j]))
776 /* If the tail was exhausted, the match was succesful. */
783 /* Checks whether string S matches each element of ACCEPTS. A list
784 element are matched either with fnmatch() or match_tail(),
785 according to whether the element contains wildcards or not.
787 If the BACKWARD is false, don't do backward comparison -- just compare
790 in_acclist (const char *const *accepts, const char *s, bool backward)
792 for (; *accepts; accepts++)
794 if (has_wildcards_p (*accepts))
796 int res = opt.ignore_case
797 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
798 /* fnmatch returns 0 if the pattern *does* match the string. */
806 if (match_tail (s, *accepts, opt.ignore_case))
811 int cmp = opt.ignore_case
812 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
821 /* Return the location of STR's suffix (file extension). Examples:
822 suffix ("foo.bar") -> "bar"
823 suffix ("foo.bar.baz") -> "baz"
824 suffix ("/foo/bar") -> NULL
825 suffix ("/foo.bar/baz") -> NULL */
827 suffix (const char *str)
831 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
835 return (char *)str + i;
840 /* Return true if S contains globbing wildcards (`*', `?', `[' or
844 has_wildcards_p (const char *s)
847 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
852 /* Return true if FNAME ends with a typical HTML suffix. The
853 following (case-insensitive) suffixes are presumed to be HTML
858 ?html (`?' matches one character)
860 #### CAVEAT. This is not necessarily a good indication that FNAME
861 refers to a file that contains HTML! */
863 has_html_suffix_p (const char *fname)
867 if ((suf = suffix (fname)) == NULL)
869 if (!strcasecmp (suf, "html"))
871 if (!strcasecmp (suf, "htm"))
873 if (suf[0] && !strcasecmp (suf + 1, "html"))
878 /* Read a line from FP and return the pointer to freshly allocated
879 storage. The storage space is obtained through malloc() and should
880 be freed with free() when it is no longer needed.
882 The length of the line is not limited, except by available memory.
883 The newline character at the end of line is retained. The line is
884 terminated with a zero character.
886 After end-of-file is encountered without anything being read, NULL
887 is returned. NULL is also returned on error. To distinguish
888 between these two cases, use the stdio function ferror(). */
891 read_whole_line (FILE *fp)
895 char *line = xmalloc (bufsize);
897 while (fgets (line + length, bufsize - length, fp))
899 length += strlen (line + length);
901 /* Possible for example when reading from a binary file where
902 a line begins with \0. */
905 if (line[length - 1] == '\n')
908 /* fgets() guarantees to read the whole line, or to use up the
909 space we've given it. We can double the buffer
912 line = xrealloc (line, bufsize);
914 if (length == 0 || ferror (fp))
919 if (length + 1 < bufsize)
920 /* Relieve the memory from our exponential greediness. We say
921 `length + 1' because the terminating \0 is not included in
922 LENGTH. We don't need to zero-terminate the string ourselves,
923 though, because fgets() does that. */
924 line = xrealloc (line, length + 1);
928 /* Read FILE into memory. A pointer to `struct file_memory' are
929 returned; use struct element `content' to access file contents, and
930 the element `length' to know the file length. `content' is *not*
931 zero-terminated, and you should *not* read or write beyond the [0,
932 length) range of characters.
934 After you are done with the file contents, call read_file_free to
937 Depending on the operating system and the type of file that is
938 being read, read_file() either mmap's the file into memory, or
939 reads the file into the core using read().
941 If file is named "-", fileno(stdin) is used for reading instead.
942 If you want to read from a real file named "-", use "./-" instead. */
945 read_file (const char *file)
948 struct file_memory *fm;
950 bool inhibit_close = false;
952 /* Some magic in the finest tradition of Perl and its kin: if FILE
953 is "-", just use stdin. */
957 inhibit_close = true;
958 /* Note that we don't inhibit mmap() in this case. If stdin is
959 redirected from a regular file, mmap() will still work. */
962 fd = open (file, O_RDONLY);
965 fm = xnew (struct file_memory);
970 if (fstat (fd, &buf) < 0)
972 fm->length = buf.st_size;
973 /* NOTE: As far as I know, the callers of this function never
974 modify the file text. Relying on this would enable us to
975 specify PROT_READ and MAP_SHARED for a marginal gain in
976 efficiency, but at some cost to generality. */
977 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
979 if (fm->content == (char *)MAP_FAILED)
989 /* The most common reason why mmap() fails is that FD does not point
990 to a plain file. However, it's also possible that mmap() doesn't
991 work for a particular type of file. Therefore, whenever mmap()
992 fails, we just fall back to the regular method. */
993 #endif /* HAVE_MMAP */
996 size = 512; /* number of bytes fm->contents can
997 hold at any given time. */
998 fm->content = xmalloc (size);
1002 if (fm->length > size / 2)
1004 /* #### I'm not sure whether the whole exponential-growth
1005 thing makes sense with kernel read. On Linux at least,
1006 read() refuses to read more than 4K from a file at a
1007 single chunk anyway. But other Unixes might optimize it
1008 better, and it doesn't *hurt* anything, so I'm leaving
1011 /* Normally, we grow SIZE exponentially to make the number
1012 of calls to read() and realloc() logarithmic in relation
1013 to file size. However, read() can read an amount of data
1014 smaller than requested, and it would be unreasonable to
1015 double SIZE every time *something* was read. Therefore,
1016 we double SIZE only when the length exceeds half of the
1017 entire allocated size. */
1019 fm->content = xrealloc (fm->content, size);
1021 nread = read (fd, fm->content + fm->length, size - fm->length);
1023 /* Successful read. */
1024 fm->length += nread;
1034 if (size > fm->length && fm->length != 0)
1035 /* Due to exponential growth of fm->content, the allocated region
1036 might be much larger than what is actually needed. */
1037 fm->content = xrealloc (fm->content, fm->length);
1044 xfree (fm->content);
1049 /* Release the resources held by FM. Specifically, this calls
1050 munmap() or xfree() on fm->content, depending whether mmap or
1051 malloc/read were used to read in the file. It also frees the
1052 memory needed to hold the FM structure itself. */
1055 read_file_free (struct file_memory *fm)
1060 munmap (fm->content, fm->length);
1065 xfree (fm->content);
1070 /* Free the pointers in a NULL-terminated vector of pointers, then
1071 free the pointer itself. */
1073 free_vec (char **vec)
1084 /* Append vector V2 to vector V1. The function frees V2 and
1085 reallocates V1 (thus you may not use the contents of neither
1086 pointer after the call). If V1 is NULL, V2 is returned. */
1088 merge_vecs (char **v1, char **v2)
1098 /* To avoid j == 0 */
1103 for (i = 0; v1[i]; i++)
1106 for (j = 0; v2[j]; j++)
1108 /* Reallocate v1. */
1109 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1110 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1115 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1116 is allocated as needed. Return the new value of the vector. */
1119 vec_append (char **vec, const char *str)
1121 int cnt; /* count of vector elements, including
1122 the one we're about to append */
1125 for (cnt = 0; vec[cnt]; cnt++)
1131 /* Reallocate the array to fit the new element and the NULL. */
1132 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1133 /* Append a copy of STR to the vector. */
1134 vec[cnt - 1] = xstrdup (str);
1139 /* Sometimes it's useful to create "sets" of strings, i.e. special
1140 hash tables where you want to store strings as keys and merely
1141 query for their existence. Here is a set of utility routines that
1142 makes that transparent. */
1145 string_set_add (struct hash_table *ht, const char *s)
1147 /* First check whether the set element already exists. If it does,
1148 do nothing so that we don't have to free() the old element and
1149 then strdup() a new one. */
1150 if (hash_table_contains (ht, s))
1153 /* We use "1" as value. It provides us a useful and clear arbitrary
1154 value, and it consumes no memory -- the pointers to the same
1155 string "1" will be shared by all the key-value pairs in all `set'
1157 hash_table_put (ht, xstrdup (s), "1");
1160 /* Synonym for hash_table_contains... */
1163 string_set_contains (struct hash_table *ht, const char *s)
1165 return hash_table_contains (ht, s);
1168 /* Convert the specified string set to array. ARRAY should be large
1169 enough to hold hash_table_count(ht) char pointers. */
1171 void string_set_to_array (struct hash_table *ht, char **array)
1173 hash_table_iterator iter;
1174 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1175 *array++ = iter.key;
1178 /* Free the string set. This frees both the storage allocated for
1179 keys and the actual hash table. (hash_table_destroy would only
1180 destroy the hash table.) */
1183 string_set_free (struct hash_table *ht)
1185 hash_table_iterator iter;
1186 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1188 hash_table_destroy (ht);
1191 /* Utility function: simply call xfree() on all keys and values of HT. */
1194 free_keys_and_values (struct hash_table *ht)
1196 hash_table_iterator iter;
1197 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1204 /* Get digit grouping data for thousand separors by calling
1205 localeconv(). The data includes separator string and grouping info
1206 and is cached after the first call to the function.
1208 In locales that don't set a thousand separator (such as the "C"
1209 locale), this forces it to be ",". We are now only showing
1210 thousand separators in one place, so this shouldn't be a problem in
1214 get_grouping_data (const char **sep, const char **grouping)
1216 static const char *cached_sep;
1217 static const char *cached_grouping;
1218 static bool initialized;
1221 /* Get the grouping info from the locale. */
1222 struct lconv *lconv = localeconv ();
1223 cached_sep = lconv->thousands_sep;
1224 cached_grouping = lconv->grouping;
1225 #if ! USE_NLS_PROGRESS_BAR
1226 /* We can't count column widths, so ensure that the separator
1227 * is single-byte only (let check below determine what byte). */
1228 if (strlen(cached_sep) > 1)
1233 /* Many locales (such as "C" or "hr_HR") don't specify
1234 grouping, which we still want to use it for legibility.
1235 In those locales set the sep char to ',', unless that
1236 character is used for decimal point, in which case set it
1238 if (*lconv->decimal_point != ',')
1242 cached_grouping = "\x03";
1247 *grouping = cached_grouping;
1250 /* Return a printed representation of N with thousand separators.
1251 This should respect locale settings, with the exception of the "C"
1252 locale which mandates no separator, but we use one anyway.
1254 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1255 the separators because it's too non-portable, and it's hard to test
1256 for this feature at configure time. Besides, it wouldn't display
1257 separators in the "C" locale, still used by many Unix users. */
1260 with_thousand_seps (wgint n)
1262 static char outbuf[48];
1263 char *p = outbuf + sizeof outbuf;
1265 /* Info received from locale */
1266 const char *grouping, *sep;
1269 /* State information */
1270 int i = 0, groupsize;
1271 const char *atgroup;
1273 bool negative = n < 0;
1275 /* Initialize grouping data. */
1276 get_grouping_data (&sep, &grouping);
1277 seplen = strlen (sep);
1279 groupsize = *atgroup++;
1281 /* This would overflow on WGINT_MIN, but printing negative numbers
1282 is not an important goal of this fuinction. */
1286 /* Write the number into the buffer, backwards, inserting the
1287 separators as necessary. */
1291 *--p = n % 10 + '0';
1295 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1296 if (++i == groupsize)
1301 memcpy (p -= seplen, sep, seplen);
1304 groupsize = *atgroup++;
1313 /* N, a byte quantity, is converted to a human-readable abberviated
1314 form a la sizes printed by `ls -lh'. The result is written to a
1315 static buffer, a pointer to which is returned.
1317 Unlike `with_thousand_seps', this approximates to the nearest unit.
1318 Quoting GNU libit: "Most people visually process strings of 3-4
1319 digits effectively, but longer strings of digits are more prone to
1320 misinterpretation. Hence, converting to an abbreviated form
1321 usually improves readability."
1323 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1324 original computer-related meaning of "powers of 1024". We don't
1325 use the "*bibyte" names invented in 1998, and seldom used in
1326 practice. Wikipedia's entry on "binary prefix" discusses this in
1330 human_readable (HR_NUMTYPE n)
1332 /* These suffixes are compatible with those of GNU `ls -lh'. */
1333 static char powers[] =
1335 'K', /* kilobyte, 2^10 bytes */
1336 'M', /* megabyte, 2^20 bytes */
1337 'G', /* gigabyte, 2^30 bytes */
1338 'T', /* terabyte, 2^40 bytes */
1339 'P', /* petabyte, 2^50 bytes */
1340 'E', /* exabyte, 2^60 bytes */
1345 /* If the quantity is smaller than 1K, just print it. */
1348 snprintf (buf, sizeof (buf), "%d", (int) n);
1352 /* Loop over powers, dividing N with 1024 in each iteration. This
1353 works unchanged for all sizes of wgint, while still avoiding
1354 non-portable `long double' arithmetic. */
1355 for (i = 0; i < countof (powers); i++)
1357 /* At each iteration N is greater than the *subsequent* power.
1358 That way N/1024.0 produces a decimal number in the units of
1360 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1362 double val = n / 1024.0;
1363 /* Print values smaller than 10 with one decimal digits, and
1364 others without any decimals. */
1365 snprintf (buf, sizeof (buf), "%.*f%c",
1366 val < 10 ? 1 : 0, val, powers[i]);
1371 return NULL; /* unreached */
1374 /* Count the digits in the provided number. Used to allocate space
1375 when printing numbers. */
1378 numdigit (wgint number)
1382 ++cnt; /* accomodate '-' */
1383 while ((number /= 10) != 0)
1388 #define PR(mask) *p++ = n / (mask) + '0'
1390 /* DIGITS_<D> is used to print a D-digit number and should be called
1391 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1392 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1393 Recursively this continues until DIGITS_1 is invoked. */
1395 #define DIGITS_1(mask) PR (mask)
1396 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1397 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1398 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1399 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1400 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1401 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1402 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1403 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1404 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1406 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1408 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1409 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1410 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1411 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1412 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1413 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1414 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1415 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1416 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1418 /* Shorthand for casting to wgint. */
1421 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1422 `sprintf(buffer, "%lld", (long long) number)', only typically much
1423 faster and portable to machines without long long.
1425 The speedup may make a difference in programs that frequently
1426 convert numbers to strings. Some implementations of sprintf,
1427 particularly the one in some versions of GNU libc, have been known
1428 to be quite slow when converting integers to strings.
1430 Return the pointer to the location where the terminating zero was
1431 printed. (Equivalent to calling buffer+strlen(buffer) after the
1434 BUFFER should be large enough to accept as many bytes as you expect
1435 the number to take up. On machines with 64-bit wgints the maximum
1436 needed size is 24 bytes. That includes the digits needed for the
1437 largest 64-bit number, the `-' sign in case it's negative, and the
1438 terminating '\0'. */
1441 number_to_string (char *buffer, wgint number)
1446 int last_digit_char = 0;
1448 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1449 /* We are running in a very strange environment. Leave the correct
1450 printing to sprintf. */
1451 p += sprintf (buf, "%j", (intmax_t) (n));
1452 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1458 /* n = -n would overflow because -n would evaluate to a
1459 wgint value larger than WGINT_MAX. Need to make n
1460 smaller and handle the last digit separately. */
1461 int last_digit = n % 10;
1462 /* The sign of n%10 is implementation-defined. */
1464 last_digit_char = '0' - last_digit;
1466 last_digit_char = '0' + last_digit;
1467 /* After n is made smaller, -n will not overflow. */
1475 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1476 way printing any N is fully open-coded without a loop or jump.
1477 (Also see description of DIGITS_*.) */
1479 if (n < 10) DIGITS_1 (1);
1480 else if (n < 100) DIGITS_2 (10);
1481 else if (n < 1000) DIGITS_3 (100);
1482 else if (n < 10000) DIGITS_4 (1000);
1483 else if (n < 100000) DIGITS_5 (10000);
1484 else if (n < 1000000) DIGITS_6 (100000);
1485 else if (n < 10000000) DIGITS_7 (1000000);
1486 else if (n < 100000000) DIGITS_8 (10000000);
1487 else if (n < 1000000000) DIGITS_9 (100000000);
1488 #if SIZEOF_WGINT == 4
1489 /* wgint is 32 bits wide: no number has more than 10 digits. */
1490 else DIGITS_10 (1000000000);
1492 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1493 Constants are constructed by compile-time multiplication to avoid
1494 dealing with different notations for 64-bit constants
1495 (nL/nLL/nI64, depending on the compiler and architecture). */
1496 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1497 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1498 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1499 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1500 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1501 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1502 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1503 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1504 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1505 else DIGITS_19 (1000000000*(W)1000000000);
1508 if (last_digit_char)
1509 *p++ = last_digit_char;
1512 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1519 #undef SPRINTF_WGINT
1542 /* Print NUMBER to a statically allocated string and return a pointer
1543 to the printed representation.
1545 This function is intended to be used in conjunction with printf.
1546 It is hard to portably print wgint values:
1547 a) you cannot use printf("%ld", number) because wgint can be long
1548 long on 32-bit machines with LFS.
1549 b) you cannot use printf("%lld", number) because NUMBER could be
1550 long on 32-bit machines without LFS, or on 64-bit machines,
1551 which do not require LFS. Also, Windows doesn't support %lld.
1552 c) you cannot use printf("%j", (int_max_t) number) because not all
1553 versions of printf support "%j", the most notable being the one
1555 d) you cannot #define WGINT_FMT to the appropriate format and use
1556 printf(WGINT_FMT, number) because that would break translations
1557 for user-visible messages, such as printf("Downloaded: %d
1560 What you should use instead is printf("%s", number_to_static_string
1563 CAVEAT: since the function returns pointers to static data, you
1564 must be careful to copy its result before calling it again.
1565 However, to make it more useful with printf, the function maintains
1566 an internal ring of static buffers to return. That way things like
1567 printf("%s %s", number_to_static_string (num1),
1568 number_to_static_string (num2)) work as expected. Three buffers
1569 are currently used, which means that "%s %s %s" will work, but "%s
1570 %s %s %s" won't. If you need to print more than three wgints,
1571 bump the RING_SIZE (or rethink your message.) */
1574 number_to_static_string (wgint number)
1576 static char ring[RING_SIZE][24];
1578 char *buf = ring[ringpos];
1579 number_to_string (buf, number);
1580 ringpos = (ringpos + 1) % RING_SIZE;
1584 /* Determine the width of the terminal we're running on. If that's
1585 not possible, return 0. */
1588 determine_screen_width (void)
1590 /* If there's a way to get the terminal size using POSIX
1591 tcgetattr(), somebody please tell me. */
1596 if (opt.lfilename != NULL)
1599 fd = fileno (stderr);
1600 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1601 return 0; /* most likely ENOTTY */
1604 #elif defined(WINDOWS)
1605 CONSOLE_SCREEN_BUFFER_INFO csbi;
1606 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1608 return csbi.dwSize.X;
1609 #else /* neither TIOCGWINSZ nor WINDOWS */
1611 #endif /* neither TIOCGWINSZ nor WINDOWS */
1614 /* Whether the rnd system (either rand or [dl]rand48) has been
1616 static int rnd_seeded;
1618 /* Return a random number between 0 and MAX-1, inclusive.
1620 If the system does not support lrand48 and MAX is greater than the
1621 value of RAND_MAX+1 on the system, the returned value will be in
1622 the range [0, RAND_MAX]. This may be fixed in a future release.
1623 The random number generator is seeded automatically the first time
1626 This uses lrand48 where available, rand elsewhere. DO NOT use it
1627 for cryptography. It is only meant to be used in situations where
1628 quality of the random numbers returned doesn't really matter. */
1631 random_number (int max)
1636 srand48 ((long) time (NULL) ^ (long) getpid ());
1639 return lrand48 () % max;
1640 #else /* not HAVE_DRAND48 */
1646 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1651 /* Like rand() % max, but uses the high-order bits for better
1652 randomness on architectures where rand() is implemented using a
1653 simple congruential generator. */
1655 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1656 return (int) bounded;
1658 #endif /* not HAVE_DRAND48 */
1661 /* Return a random uniformly distributed floating point number in the
1662 [0, 1) range. Uses drand48 where available, and a really lame
1663 kludge elsewhere. */
1671 srand48 ((long) time (NULL) ^ (long) getpid ());
1675 #else /* not HAVE_DRAND48 */
1676 return ( random_number (10000) / 10000.0
1677 + random_number (10000) / (10000.0 * 10000.0)
1678 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1679 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1680 #endif /* not HAVE_DRAND48 */
1683 /* Implementation of run_with_timeout, a generic timeout-forcing
1684 routine for systems with Unix-like signal handling. */
1686 #ifdef USE_SIGNAL_TIMEOUT
1687 # ifdef HAVE_SIGSETJMP
1688 # define SETJMP(env) sigsetjmp (env, 1)
1690 static sigjmp_buf run_with_timeout_env;
1693 abort_run_with_timeout (int sig)
1695 assert (sig == SIGALRM);
1696 siglongjmp (run_with_timeout_env, -1);
1698 # else /* not HAVE_SIGSETJMP */
1699 # define SETJMP(env) setjmp (env)
1701 static jmp_buf run_with_timeout_env;
1704 abort_run_with_timeout (int sig)
1706 assert (sig == SIGALRM);
1707 /* We don't have siglongjmp to preserve the set of blocked signals;
1708 if we longjumped out of the handler at this point, SIGALRM would
1709 remain blocked. We must unblock it manually. */
1710 int mask = siggetmask ();
1711 mask &= ~sigmask (SIGALRM);
1714 /* Now it's safe to longjump. */
1715 longjmp (run_with_timeout_env, -1);
1717 # endif /* not HAVE_SIGSETJMP */
1719 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1720 setitimer where available, alarm otherwise.
1722 TIMEOUT should be non-zero. If the timeout value is so small that
1723 it would be rounded to zero, it is rounded to the least legal value
1724 instead (1us for setitimer, 1s for alarm). That ensures that
1725 SIGALRM will be delivered in all cases. */
1728 alarm_set (double timeout)
1731 /* Use the modern itimer interface. */
1732 struct itimerval itv;
1734 itv.it_value.tv_sec = (long) timeout;
1735 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1736 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1737 /* Ensure that we wait for at least the minimum interval.
1738 Specifying zero would mean "wait forever". */
1739 itv.it_value.tv_usec = 1;
1740 setitimer (ITIMER_REAL, &itv, NULL);
1741 #else /* not ITIMER_REAL */
1742 /* Use the old alarm() interface. */
1743 int secs = (int) timeout;
1745 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1746 because alarm(0) means "never deliver the alarm", i.e. "wait
1747 forever", which is not what someone who specifies a 0.5s
1748 timeout would expect. */
1751 #endif /* not ITIMER_REAL */
1754 /* Cancel the alarm set with alarm_set. */
1760 struct itimerval disable;
1762 setitimer (ITIMER_REAL, &disable, NULL);
1763 #else /* not ITIMER_REAL */
1765 #endif /* not ITIMER_REAL */
1768 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1769 seconds. Returns true if the function was interrupted with a
1770 timeout, false otherwise.
1772 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1773 using setitimer() or alarm(). The timeout is enforced by
1774 longjumping out of the SIGALRM handler. This has several
1775 advantages compared to the traditional approach of relying on
1776 signals causing system calls to exit with EINTR:
1778 * The callback function is *forcibly* interrupted after the
1779 timeout expires, (almost) regardless of what it was doing and
1780 whether it was in a syscall. For example, a calculation that
1781 takes a long time is interrupted as reliably as an IO
1784 * It works with both SYSV and BSD signals because it doesn't
1785 depend on the default setting of SA_RESTART.
1787 * It doesn't require special handler setup beyond a simple call
1788 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1791 The only downside is that, if FUN allocates internal resources that
1792 are normally freed prior to exit from the functions, they will be
1793 lost in case of timeout. */
1796 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1806 signal (SIGALRM, abort_run_with_timeout);
1807 if (SETJMP (run_with_timeout_env) != 0)
1809 /* Longjumped out of FUN with a timeout. */
1810 signal (SIGALRM, SIG_DFL);
1813 alarm_set (timeout);
1816 /* Preserve errno in case alarm() or signal() modifies it. */
1817 saved_errno = errno;
1819 signal (SIGALRM, SIG_DFL);
1820 errno = saved_errno;
1825 #else /* not USE_SIGNAL_TIMEOUT */
1828 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1829 define it under Windows, because Windows has its own version of
1830 run_with_timeout that uses threads. */
1833 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1838 #endif /* not WINDOWS */
1839 #endif /* not USE_SIGNAL_TIMEOUT */
1843 /* Sleep the specified amount of seconds. On machines without
1844 nanosleep(), this may sleep shorter if interrupted by signals. */
1847 xsleep (double seconds)
1849 #ifdef HAVE_NANOSLEEP
1850 /* nanosleep is the preferred interface because it offers high
1851 accuracy and, more importantly, because it allows us to reliably
1852 restart receiving a signal such as SIGWINCH. (There was an
1853 actual Debian bug report about --limit-rate malfunctioning while
1854 the terminal was being resized.) */
1855 struct timespec sleep, remaining;
1856 sleep.tv_sec = (long) seconds;
1857 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1858 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1859 /* If nanosleep has been interrupted by a signal, adjust the
1860 sleeping period and return to sleep. */
1862 #elif defined(HAVE_USLEEP)
1863 /* If usleep is available, use it in preference to select. */
1866 /* On some systems, usleep cannot handle values larger than
1867 1,000,000. If the period is larger than that, use sleep
1868 first, then add usleep for subsecond accuracy. */
1870 seconds -= (long) seconds;
1872 usleep (seconds * 1000000);
1873 #else /* fall back select */
1874 /* Note that, although Windows supports select, it can't be used to
1875 implement sleeping because Winsock's select doesn't implement
1876 timeout when it is passed NULL pointers for all fd sets. (But it
1877 does under Cygwin, which implements Unix-compatible select.) */
1878 struct timeval sleep;
1879 sleep.tv_sec = (long) seconds;
1880 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1881 select (0, NULL, NULL, NULL, &sleep);
1882 /* If select returns -1 and errno is EINTR, it means we were
1883 interrupted by a signal. But without knowing how long we've
1884 actually slept, we can't return to sleep. Using gettimeofday to
1885 track sleeps is slow and unreliable due to clock skew. */
1889 #endif /* not WINDOWS */
1891 /* Encode the octets in DATA of length LENGTH to base64 format,
1892 storing the result to DEST. The output will be zero-terminated,
1893 and must point to a writable buffer of at least
1894 1+BASE64_LENGTH(length) bytes. The function returns the length of
1895 the resulting base64 data, not counting the terminating zero.
1897 This implementation does not emit newlines after 76 characters of
1901 base64_encode (const void *data, int length, char *dest)
1903 /* Conversion table. */
1904 static const char tbl[64] = {
1905 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
1906 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
1907 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
1908 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
1910 /* Access bytes in DATA as unsigned char, otherwise the shifts below
1911 don't work for data with MSB set. */
1912 const unsigned char *s = data;
1913 /* Theoretical ANSI violation when length < 3. */
1914 const unsigned char *end = (const unsigned char *) data + length - 2;
1917 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1918 for (; s < end; s += 3)
1920 *p++ = tbl[s[0] >> 2];
1921 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1922 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1923 *p++ = tbl[s[2] & 0x3f];
1926 /* Pad the result if necessary... */
1930 *p++ = tbl[s[0] >> 2];
1931 *p++ = tbl[(s[0] & 3) << 4];
1936 *p++ = tbl[s[0] >> 2];
1937 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1938 *p++ = tbl[((s[1] & 0xf) << 2)];
1942 /* ...and zero-terminate it. */
1948 /* Store in C the next non-whitespace character from the string, or \0
1949 when end of string is reached. */
1950 #define NEXT_CHAR(c, p) do { \
1951 c = (unsigned char) *p++; \
1952 } while (c_isspace (c))
1954 #define IS_ASCII(c) (((c) & 0x80) == 0)
1956 /* Decode data from BASE64 (a null-terminated string) into memory
1957 pointed to by DEST. DEST is assumed to be large enough to
1958 accomodate the decoded data, which is guaranteed to be no more than
1961 Since DEST is assumed to contain binary data, it is not
1962 NUL-terminated. The function returns the length of the data
1963 written to TO. -1 is returned in case of error caused by malformed
1966 This function originates from Free Recode. */
1969 base64_decode (const char *base64, void *dest)
1971 /* Table of base64 values for first 128 characters. Note that this
1972 assumes ASCII (but so does Wget in other places). */
1973 static const signed char base64_char_to_value[128] =
1975 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1976 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1977 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1978 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1979 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1980 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1981 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1982 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1983 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1984 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1985 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1986 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1987 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1989 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
1990 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
1992 const char *p = base64;
1998 unsigned long value;
2000 /* Process first byte of a quadruplet. */
2004 if (c == '=' || !IS_BASE64 (c))
2005 return -1; /* illegal char while decoding base64 */
2006 value = BASE64_CHAR_TO_VALUE (c) << 18;
2008 /* Process second byte of a quadruplet. */
2011 return -1; /* premature EOF while decoding base64 */
2012 if (c == '=' || !IS_BASE64 (c))
2013 return -1; /* illegal char while decoding base64 */
2014 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2017 /* Process third byte of a quadruplet. */
2020 return -1; /* premature EOF while decoding base64 */
2022 return -1; /* illegal char while decoding base64 */
2028 return -1; /* premature EOF while decoding base64 */
2030 return -1; /* padding `=' expected but not found */
2034 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2035 *q++ = 0xff & value >> 8;
2037 /* Process fourth byte of a quadruplet. */
2040 return -1; /* premature EOF while decoding base64 */
2044 return -1; /* illegal char while decoding base64 */
2046 value |= BASE64_CHAR_TO_VALUE (c);
2047 *q++ = 0xff & value;
2050 #undef BASE64_CHAR_TO_VALUE
2052 return q - (char *) dest;
2058 /* Simple merge sort for use by stable_sort. Implementation courtesy
2059 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2062 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2063 int (*cmpfun) (const void *, const void *))
2065 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2069 size_t mid = (to + from) / 2;
2070 mergesort_internal (base, temp, size, from, mid, cmpfun);
2071 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2074 for (k = from; (i <= mid) && (j <= to); k++)
2075 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2076 memcpy (ELT (temp, k), ELT (base, i++), size);
2078 memcpy (ELT (temp, k), ELT (base, j++), size);
2080 memcpy (ELT (temp, k++), ELT (base, i++), size);
2082 memcpy (ELT (temp, k++), ELT (base, j++), size);
2083 for (k = from; k <= to; k++)
2084 memcpy (ELT (base, k), ELT (temp, k), size);
2089 /* Stable sort with interface exactly like standard library's qsort.
2090 Uses mergesort internally, allocating temporary storage with
2094 stable_sort (void *base, size_t nmemb, size_t size,
2095 int (*cmpfun) (const void *, const void *))
2099 void *temp = alloca (nmemb * size * sizeof (void *));
2100 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2104 /* Print a decimal number. If it is equal to or larger than ten, the
2105 number is rounded. Otherwise it is printed with one significant
2106 digit without trailing zeros and with no more than three fractional
2107 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2108 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2110 This is useful for displaying durations because it provides
2111 order-of-magnitude information without unnecessary clutter --
2112 long-running downloads are shown without the fractional part, and
2113 short ones still retain one significant digit. */
2116 print_decimal (double number)
2118 static char buf[32];
2119 double n = number >= 0 ? number : -number;
2122 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2123 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2124 snprintf (buf, sizeof buf, "%.0f", number);
2126 snprintf (buf, sizeof buf, "%.1f", number);
2127 else if (n >= 0.001)
2128 snprintf (buf, sizeof buf, "%.1g", number);
2129 else if (n >= 0.0005)
2130 /* round [0.0005, 0.001) to 0.001 */
2131 snprintf (buf, sizeof buf, "%.3f", number);
2133 /* print numbers close to 0 as 0, not 0.000 */
2150 { "/somedir", "/somedir", true },
2151 { "/somedir", "/somedir/d2", true },
2152 { "/somedir/d1", "/somedir", false },
2155 for (i = 0; i < countof(test_array); ++i)
2157 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2159 mu_assert ("test_subdir_p: wrong result",
2160 res == test_array[i].result);
2167 test_dir_matches_p()
2175 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2176 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2177 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2178 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2179 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2180 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2181 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2182 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2183 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2184 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2185 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2186 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2187 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2190 for (i = 0; i < countof(test_array); ++i)
2192 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2194 mu_assert ("test_dir_matches_p: wrong result",
2195 res == test_array[i].result);
2201 #endif /* TESTING */