1 /* Various utility functions.
2 Copyright (C) 1996-2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software Foundation, Inc.,
18 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
36 #ifdef HAVE_SYS_TIME_H
37 # include <sys/time.h>
43 # include <sys/mman.h>
48 #ifdef HAVE_SYS_UTIME_H
49 # include <sys/utime.h>
57 /* For TIOCGWINSZ and friends: */
58 #ifdef HAVE_SYS_IOCTL_H
59 # include <sys/ioctl.h>
65 /* Needed for Unix version of run_with_timeout. */
69 #ifndef HAVE_SIGSETJMP
70 /* If sigsetjmp is a macro, configure won't pick it up. */
72 # define HAVE_SIGSETJMP
76 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
77 # define USE_SIGNAL_TIMEOUT
84 /* Utility function: like xstrdup(), but also lowercases S. */
87 xstrdup_lower (const char *s)
89 char *copy = xstrdup (s);
96 /* Copy the string formed by two pointers (one on the beginning, other
97 on the char after the last char) to a new, malloc-ed location.
100 strdupdelim (const char *beg, const char *end)
102 char *res = xmalloc (end - beg + 1);
103 memcpy (res, beg, end - beg);
104 res[end - beg] = '\0';
108 /* Parse a string containing comma-separated elements, and return a
109 vector of char pointers with the elements. Spaces following the
110 commas are ignored. */
112 sepstring (const char *s)
126 res = xrealloc (res, (i + 2) * sizeof (char *));
127 res[i] = strdupdelim (p, s);
130 /* Skip the blanks following the ','. */
138 res = xrealloc (res, (i + 2) * sizeof (char *));
139 res[i] = strdupdelim (p, s);
144 /* Like sprintf, but prints into a string of sufficient size freshly
145 allocated with malloc, which is returned. If unable to print due
146 to invalid format, returns NULL. Inability to allocate needed
147 memory results in abort, as with xmalloc. This is in spirit
148 similar to the GNU/BSD extension asprintf, but somewhat easier to
151 Internally the function either calls vasprintf or loops around
152 vsnprintf until the correct size is found. Since Wget also ships a
153 fallback implementation of vsnprintf, this should be portable. */
156 aprintf (const char *fmt, ...)
158 #ifdef HAVE_VASPRINTF
163 va_start (args, fmt);
164 ret = vasprintf (&str, fmt, args);
166 if (ret < 0 && errno == ENOMEM)
167 abort (); /* for consistency with xmalloc/xrealloc */
171 #else /* not HAVE_VASPRINTF */
173 /* vasprintf is unavailable. snprintf into a small buffer and
174 resize it as necessary. */
176 char *str = xmalloc (size);
178 /* #### This code will infloop and eventually abort in xrealloc if
179 passed a FMT that causes snprintf to consistently return -1. */
186 va_start (args, fmt);
187 n = vsnprintf (str, size, fmt, args);
190 /* If the printing worked, return the string. */
191 if (n > -1 && n < size)
194 /* Else try again with a larger buffer. */
195 if (n > -1) /* C99 */
196 size = n + 1; /* precisely what is needed */
198 size <<= 1; /* twice the old size */
199 str = xrealloc (str, size);
201 #endif /* not HAVE_VASPRINTF */
204 /* Concatenate the NULL-terminated list of string arguments into
205 freshly allocated space. */
208 concat_strings (const char *str0, ...)
211 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
214 const char *next_str;
215 int total_length = 0;
218 /* Calculate the length of and allocate the resulting string. */
221 va_start (args, str0);
222 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
224 int len = strlen (next_str);
225 if (argcount < countof (saved_lengths))
226 saved_lengths[argcount++] = len;
230 p = ret = xmalloc (total_length + 1);
232 /* Copy the strings into the allocated space. */
235 va_start (args, str0);
236 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
239 if (argcount < countof (saved_lengths))
240 len = saved_lengths[argcount++];
242 len = strlen (next_str);
243 memcpy (p, next_str, len);
252 /* Return pointer to a static char[] buffer in which zero-terminated
253 string-representation of TM (in form hh:mm:ss) is printed.
255 If TM is NULL, the current time will be used. */
258 time_str (time_t *tm)
260 static char output[15];
262 time_t secs = tm ? *tm : time (NULL);
266 /* In case of error, return the empty string. Maybe we should
267 just abort if this happens? */
271 ptm = localtime (&secs);
272 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
276 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
279 datetime_str (time_t *tm)
281 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
283 time_t secs = tm ? *tm : time (NULL);
287 /* In case of error, return the empty string. Maybe we should
288 just abort if this happens? */
292 ptm = localtime (&secs);
293 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
294 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
295 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
299 /* The Windows versions of the following two functions are defined in
304 fork_to_background (void)
307 /* Whether we arrange our own version of opt.lfilename here. */
308 bool logfile_changed = false;
312 /* We must create the file immediately to avoid either a race
313 condition (which arises from using unique_name and failing to
314 use fopen_excl) or lying to the user about the log file name
315 (which arises from using unique_name, printing the name, and
316 using fopen_excl later on.) */
317 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
320 logfile_changed = true;
333 /* parent, no error */
334 printf (_("Continuing in background, pid %d.\n"), (int) pid);
336 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
337 exit (0); /* #### should we use _exit()? */
340 /* child: give up the privileges and keep running. */
342 freopen ("/dev/null", "r", stdin);
343 freopen ("/dev/null", "w", stdout);
344 freopen ("/dev/null", "w", stderr);
346 #endif /* not WINDOWS */
348 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
349 specified with TM. The atime ("access time") is set to the current
353 touch (const char *file, time_t tm)
355 #ifdef HAVE_STRUCT_UTIMBUF
356 struct utimbuf times;
364 times.actime = time (NULL);
365 if (utime (file, ×) == -1)
366 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
369 /* Checks if FILE is a symbolic link, and removes it if it is. Does
370 nothing under MS-Windows. */
372 remove_link (const char *file)
377 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
379 DEBUGP (("Unlinking %s (symlink).\n", file));
382 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
383 file, strerror (errno));
388 /* Does FILENAME exist? This is quite a lousy implementation, since
389 it supplies no error codes -- only a yes-or-no answer. Thus it
390 will return that a file does not exist if, e.g., the directory is
391 unreadable. I don't mind it too much currently, though. The
392 proper way should, of course, be to have a third, error state,
393 other than true/false, but that would introduce uncalled-for
394 additional complexity to the callers. */
396 file_exists_p (const char *filename)
399 return access (filename, F_OK) >= 0;
402 return stat (filename, &buf) >= 0;
406 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
407 Returns 0 on error. */
409 file_non_directory_p (const char *path)
412 /* Use lstat() rather than stat() so that symbolic links pointing to
413 directories can be identified correctly. */
414 if (lstat (path, &buf) != 0)
416 return S_ISDIR (buf.st_mode) ? false : true;
419 /* Return the size of file named by FILENAME, or -1 if it cannot be
420 opened or seeked into. */
422 file_size (const char *filename)
424 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
426 /* We use fseek rather than stat to determine the file size because
427 that way we can also verify that the file is readable without
428 explicitly checking for permissions. Inspired by the POST patch
430 FILE *fp = fopen (filename, "rb");
433 fseeko (fp, 0, SEEK_END);
439 if (stat (filename, &st) < 0)
445 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
446 doesn't exist is found. Return a freshly allocated copy of the
450 unique_name_1 (const char *prefix)
453 int plen = strlen (prefix);
454 char *template = (char *)alloca (plen + 1 + 24);
455 char *template_tail = template + plen;
457 memcpy (template, prefix, plen);
458 *template_tail++ = '.';
461 number_to_string (template_tail, count++);
462 while (file_exists_p (template));
464 return xstrdup (template);
467 /* Return a unique file name, based on FILE.
469 More precisely, if FILE doesn't exist, it is returned unmodified.
470 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
471 file name that doesn't exist is returned.
473 The resulting file is not created, only verified that it didn't
474 exist at the point in time when the function was called.
475 Therefore, where security matters, don't rely that the file created
476 by this function exists until you open it with O_EXCL or
479 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
480 string. Otherwise, it may return FILE if the file doesn't exist
481 (and therefore doesn't need changing). */
484 unique_name (const char *file, bool allow_passthrough)
486 /* If the FILE itself doesn't exist, return it without
488 if (!file_exists_p (file))
489 return allow_passthrough ? (char *)file : xstrdup (file);
491 /* Otherwise, find a numeric suffix that results in unused file name
493 return unique_name_1 (file);
496 /* Create a file based on NAME, except without overwriting an existing
497 file with that name. Providing O_EXCL is correctly implemented,
498 this function does not have the race condition associated with
499 opening the file returned by unique_name. */
502 unique_create (const char *name, bool binary, char **opened_name)
504 /* unique file name, based on NAME */
505 char *uname = unique_name (name, false);
507 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
510 uname = unique_name (name, false);
512 if (opened_name && fp != NULL)
515 *opened_name = uname;
527 /* Open the file for writing, with the addition that the file is
528 opened "exclusively". This means that, if the file already exists,
529 this function will *fail* and errno will be set to EEXIST. If
530 BINARY is set, the file will be opened in binary mode, equivalent
533 If opening the file fails for any reason, including the file having
534 previously existed, this function returns NULL and sets errno
538 fopen_excl (const char *fname, bool binary)
542 int flags = O_WRONLY | O_CREAT | O_EXCL;
547 fd = open (fname, flags, 0666);
550 return fdopen (fd, binary ? "wb" : "w");
551 #else /* not O_EXCL */
552 /* Manually check whether the file exists. This is prone to race
553 conditions, but systems without O_EXCL haven't deserved
555 if (file_exists_p (fname))
560 return fopen (fname, binary ? "wb" : "w");
561 #endif /* not O_EXCL */
564 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
565 are missing, create them first. In case any mkdir() call fails,
566 return its error status. Returns 0 on successful completion.
568 The behaviour of this function should be identical to the behaviour
569 of `mkdir -p' on systems where mkdir supports the `-p' option. */
571 make_directory (const char *directory)
573 int i, ret, quit = 0;
576 /* Make a copy of dir, to be able to write to it. Otherwise, the
577 function is unsafe if called with a read-only char *argument. */
578 STRDUP_ALLOCA (dir, directory);
580 /* If the first character of dir is '/', skip it (and thus enable
581 creation of absolute-pathname directories. */
582 for (i = (*dir == '/'); 1; ++i)
584 for (; dir[i] && dir[i] != '/'; i++)
589 /* Check whether the directory already exists. Allow creation of
590 of intermediate directories to fail, as the initial path components
591 are not necessarily directories! */
592 if (!file_exists_p (dir))
593 ret = mkdir (dir, 0777);
604 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
605 should be a file name.
607 file_merge("/foo/bar", "baz") => "/foo/baz"
608 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
609 file_merge("foo", "bar") => "bar"
611 In other words, it's a simpler and gentler version of uri_merge. */
614 file_merge (const char *base, const char *file)
617 const char *cut = (const char *)strrchr (base, '/');
620 return xstrdup (file);
622 result = xmalloc (cut - base + 1 + strlen (file) + 1);
623 memcpy (result, base, cut - base);
624 result[cut - base] = '/';
625 strcpy (result + (cut - base) + 1, file);
630 /* Like fnmatch, but performs a case-insensitive match. */
633 fnmatch_nocase (const char *pattern, const char *string, int flags)
636 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
637 also present on *BSD platforms, and possibly elsewhere. */
638 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
640 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
641 char *patcopy = (char *) alloca (strlen (pattern) + 1);
642 char *strcopy = (char *) alloca (strlen (string) + 1);
644 for (p = patcopy; *pattern; pattern++, p++)
645 *p = TOLOWER (*pattern);
647 for (p = strcopy; *string; string++, p++)
648 *p = TOLOWER (*string);
650 return fnmatch (patcopy, strcopy, flags);
654 static bool in_acclist (const char *const *, const char *, bool);
656 /* Determine whether a file is acceptable to be followed, according to
657 lists of patterns to accept/reject. */
659 acceptable (const char *s)
663 while (l && s[l] != '/')
670 return (in_acclist ((const char *const *)opt.accepts, s, true)
671 && !in_acclist ((const char *const *)opt.rejects, s, true));
673 return in_acclist ((const char *const *)opt.accepts, s, true);
675 else if (opt.rejects)
676 return !in_acclist ((const char *const *)opt.rejects, s, true);
680 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
681 `/something', frontcmp() will return true only if S2 begins with
684 frontcmp (const char *s1, const char *s2)
686 if (!opt.ignore_case)
687 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2)
690 for (; *s1 && *s2 && (TOLOWER (*s1) == TOLOWER (*s2)); ++s1, ++s2)
695 /* Iterate through STRLIST, and return the first element that matches
696 S, through wildcards or front comparison (as appropriate). */
698 proclist (char **strlist, const char *s)
701 int (*matcher) (const char *, const char *, int)
702 = opt.ignore_case ? fnmatch_nocase : fnmatch;
704 for (x = strlist; *x; x++)
706 /* Remove leading '/' */
707 char *p = *x + (**x == '/');
708 if (has_wildcards_p (p))
710 if (matcher (p, s, FNM_PATHNAME) == 0)
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 (!proclist (opt.includes, directory))
741 if (proclist (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 (TOLOWER (string[i]) != 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);
1169 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1171 char ***arrayptr = (char ***) arg;
1172 *(*arrayptr)++ = (char *) key;
1176 /* Convert the specified string set to array. ARRAY should be large
1177 enough to hold hash_table_count(ht) char pointers. */
1179 void string_set_to_array (struct hash_table *ht, char **array)
1181 hash_table_map (ht, string_set_to_array_mapper, &array);
1185 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1192 string_set_free (struct hash_table *ht)
1194 hash_table_map (ht, string_set_free_mapper, NULL);
1195 hash_table_destroy (ht);
1199 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1206 /* Another utility function: call free() on all keys and values of HT. */
1209 free_keys_and_values (struct hash_table *ht)
1211 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1214 /* Get digit grouping data for thousand separors by calling
1215 localeconv(). The data includes separator string and grouping info
1216 and is cached after the first call to the function.
1218 In locales that don't set a thousand separator (such as the "C"
1219 locale), this forces it to be ",". We are now only showing
1220 thousand separators in one place, so this shouldn't be a problem in
1224 get_grouping_data (const char **sep, const char **grouping)
1226 static const char *cached_sep;
1227 static const char *cached_grouping;
1228 static bool initialized;
1231 /* Get the grouping info from the locale. */
1232 struct lconv *lconv = localeconv ();
1233 cached_sep = lconv->thousands_sep;
1234 cached_grouping = lconv->grouping;
1237 /* Many locales (such as "C" or "hr_HR") don't specify
1238 grouping, which we still want to use it for legibility.
1239 In those locales set the sep char to ',', unless that
1240 character is used for decimal point, in which case set it
1242 if (*lconv->decimal_point != ',')
1246 cached_grouping = "\x03";
1251 *grouping = cached_grouping;
1254 /* Return a printed representation of N with thousand separators.
1255 This should respect locale settings, with the exception of the "C"
1256 locale which mandates no separator, but we use one anyway.
1258 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1259 the separators because it's too non-portable, and it's hard to test
1260 for this feature at configure time. Besides, it wouldn't display
1261 separators in the "C" locale, still used by many Unix users. */
1264 with_thousand_seps (wgint n)
1266 static char outbuf[48];
1267 char *p = outbuf + sizeof outbuf;
1269 /* Info received from locale */
1270 const char *grouping, *sep;
1273 /* State information */
1274 int i = 0, groupsize;
1275 const char *atgroup;
1277 bool negative = n < 0;
1279 /* Initialize grouping data. */
1280 get_grouping_data (&sep, &grouping);
1281 seplen = strlen (sep);
1283 groupsize = *atgroup++;
1285 /* This would overflow on WGINT_MIN, but printing negative numbers
1286 is not an important goal of this fuinction. */
1290 /* Write the number into the buffer, backwards, inserting the
1291 separators as necessary. */
1295 *--p = n % 10 + '0';
1299 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1300 if (++i == groupsize)
1305 memcpy (p -= seplen, sep, seplen);
1308 groupsize = *atgroup++;
1317 /* N, a byte quantity, is converted to a human-readable abberviated
1318 form a la sizes printed by `ls -lh'. The result is written to a
1319 static buffer, a pointer to which is returned.
1321 Unlike `with_thousand_seps', this approximates to the nearest unit.
1322 Quoting GNU libit: "Most people visually process strings of 3-4
1323 digits effectively, but longer strings of digits are more prone to
1324 misinterpretation. Hence, converting to an abbreviated form
1325 usually improves readability."
1327 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1328 original computer-related meaning of "powers of 1024". We don't
1329 use the "*bibyte" names invented in 1998, and seldom used in
1330 practice. Wikipedia's entry on "binary prefix" discusses this in
1334 human_readable (HR_NUMTYPE n)
1336 /* These suffixes are compatible with those of GNU `ls -lh'. */
1337 static char powers[] =
1339 'K', /* kilobyte, 2^10 bytes */
1340 'M', /* megabyte, 2^20 bytes */
1341 'G', /* gigabyte, 2^30 bytes */
1342 'T', /* terabyte, 2^40 bytes */
1343 'P', /* petabyte, 2^50 bytes */
1344 'E', /* exabyte, 2^60 bytes */
1349 /* If the quantity is smaller than 1K, just print it. */
1352 snprintf (buf, sizeof (buf), "%d", (int) n);
1356 /* Loop over powers, dividing N with 1024 in each iteration. This
1357 works unchanged for all sizes of wgint, while still avoiding
1358 non-portable `long double' arithmetic. */
1359 for (i = 0; i < countof (powers); i++)
1361 /* At each iteration N is greater than the *subsequent* power.
1362 That way N/1024.0 produces a decimal number in the units of
1364 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1366 double val = n / 1024.0;
1367 /* Print values smaller than 10 with one decimal digits, and
1368 others without any decimals. */
1369 snprintf (buf, sizeof (buf), "%.*f%c",
1370 val < 10 ? 1 : 0, val, powers[i]);
1375 return NULL; /* unreached */
1378 /* Count the digits in the provided number. Used to allocate space
1379 when printing numbers. */
1382 numdigit (wgint number)
1386 ++cnt; /* accomodate '-' */
1387 while ((number /= 10) != 0)
1392 #define PR(mask) *p++ = n / (mask) + '0'
1394 /* DIGITS_<D> is used to print a D-digit number and should be called
1395 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1396 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1397 Recursively this continues until DIGITS_1 is invoked. */
1399 #define DIGITS_1(mask) PR (mask)
1400 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1401 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1402 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1403 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1404 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1405 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1406 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1407 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1408 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1410 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1412 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1413 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1414 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1415 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1416 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1417 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1418 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1419 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1420 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1422 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1423 cases and to portably support strange sizes of wgint. Ideally this
1424 would just use "%j" and intmax_t, but many systems don't support
1425 it, so it's used only if nothing else works. */
1426 #if SIZEOF_LONG >= SIZEOF_WGINT
1427 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1428 #elif SIZEOF_LONG_LONG >= SIZEOF_WGINT
1429 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1430 #elif defined(WINDOWS)
1431 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64d", (__int64) (n))
1433 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
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 GNU libc, have been known to be extremely
1446 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 big enough to accept as many bytes as you expect
1453 the number to take up. On machines with 64-bit longs 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 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1465 /* We are running in a strange or misconfigured environment. Let
1466 sprintf cope with it. */
1467 SPRINTF_WGINT (buffer, n);
1468 p += strlen (buffer);
1469 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1475 /* -n would overflow. Have sprintf deal with this. */
1476 SPRINTF_WGINT (buffer, n);
1477 p += strlen (buffer);
1485 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1486 way printing any N is fully open-coded without a loop or jump.
1487 (Also see description of DIGITS_*.) */
1489 if (n < 10) DIGITS_1 (1);
1490 else if (n < 100) DIGITS_2 (10);
1491 else if (n < 1000) DIGITS_3 (100);
1492 else if (n < 10000) DIGITS_4 (1000);
1493 else if (n < 100000) DIGITS_5 (10000);
1494 else if (n < 1000000) DIGITS_6 (100000);
1495 else if (n < 10000000) DIGITS_7 (1000000);
1496 else if (n < 100000000) DIGITS_8 (10000000);
1497 else if (n < 1000000000) DIGITS_9 (100000000);
1498 #if SIZEOF_WGINT == 4
1499 /* wgint is 32 bits wide: no number has more than 10 digits. */
1500 else DIGITS_10 (1000000000);
1502 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1503 Constants are constructed by compile-time multiplication to avoid
1504 dealing with different notations for 64-bit constants
1505 (nL/nLL/nI64, depending on the compiler and architecture). */
1506 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1507 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1508 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1509 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1510 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1511 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1512 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1513 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1514 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1515 else DIGITS_19 (1000000000*(W)1000000000);
1519 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1526 #undef SPRINTF_WGINT
1549 /* Print NUMBER to a statically allocated string and return a pointer
1550 to the printed representation.
1552 This function is intended to be used in conjunction with printf.
1553 It is hard to portably print wgint values:
1554 a) you cannot use printf("%ld", number) because wgint can be long
1555 long on 32-bit machines with LFS.
1556 b) you cannot use printf("%lld", number) because NUMBER could be
1557 long on 32-bit machines without LFS, or on 64-bit machines,
1558 which do not require LFS. Also, Windows doesn't support %lld.
1559 c) you cannot use printf("%j", (int_max_t) number) because not all
1560 versions of printf support "%j", the most notable being the one
1562 d) you cannot #define WGINT_FMT to the appropriate format and use
1563 printf(WGINT_FMT, number) because that would break translations
1564 for user-visible messages, such as printf("Downloaded: %d
1567 What you should use instead is printf("%s", number_to_static_string
1570 CAVEAT: since the function returns pointers to static data, you
1571 must be careful to copy its result before calling it again.
1572 However, to make it more useful with printf, the function maintains
1573 an internal ring of static buffers to return. That way things like
1574 printf("%s %s", number_to_static_string (num1),
1575 number_to_static_string (num2)) work as expected. Three buffers
1576 are currently used, which means that "%s %s %s" will work, but "%s
1577 %s %s %s" won't. If you need to print more than three wgints,
1578 bump the RING_SIZE (or rethink your message.) */
1581 number_to_static_string (wgint number)
1583 static char ring[RING_SIZE][24];
1585 char *buf = ring[ringpos];
1586 number_to_string (buf, number);
1587 ringpos = (ringpos + 1) % RING_SIZE;
1591 /* Determine the width of the terminal we're running on. If that's
1592 not possible, return 0. */
1595 determine_screen_width (void)
1597 /* If there's a way to get the terminal size using POSIX
1598 tcgetattr(), somebody please tell me. */
1603 if (opt.lfilename != NULL)
1606 fd = fileno (stderr);
1607 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1608 return 0; /* most likely ENOTTY */
1611 #elif defined(WINDOWS)
1612 CONSOLE_SCREEN_BUFFER_INFO csbi;
1613 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1615 return csbi.dwSize.X;
1616 #else /* neither TIOCGWINSZ nor WINDOWS */
1618 #endif /* neither TIOCGWINSZ nor WINDOWS */
1621 /* Whether the rnd system (either rand or [dl]rand48) has been
1623 static int rnd_seeded;
1625 /* Return a random number between 0 and MAX-1, inclusive.
1627 If the system does not support lrand48 and MAX is greater than the
1628 value of RAND_MAX+1 on the system, the returned value will be in
1629 the range [0, RAND_MAX]. This may be fixed in a future release.
1630 The random number generator is seeded automatically the first time
1633 This uses lrand48 where available, rand elsewhere. DO NOT use it
1634 for cryptography. It is only meant to be used in situations where
1635 quality of the random numbers returned doesn't really matter. */
1638 random_number (int max)
1643 srand48 ((long) time (NULL) ^ (long) getpid ());
1646 return lrand48 () % max;
1647 #else /* not HAVE_DRAND48 */
1653 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1658 /* Like rand() % max, but uses the high-order bits for better
1659 randomness on architectures where rand() is implemented using a
1660 simple congruential generator. */
1662 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1663 return (int) bounded;
1665 #endif /* not HAVE_DRAND48 */
1668 /* Return a random uniformly distributed floating point number in the
1669 [0, 1) range. Uses drand48 where available, and a really lame
1670 kludge elsewhere. */
1678 srand48 ((long) time (NULL) ^ (long) getpid ());
1682 #else /* not HAVE_DRAND48 */
1683 return ( random_number (10000) / 10000.0
1684 + random_number (10000) / (10000.0 * 10000.0)
1685 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1686 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1687 #endif /* not HAVE_DRAND48 */
1690 /* Implementation of run_with_timeout, a generic timeout-forcing
1691 routine for systems with Unix-like signal handling. */
1693 #ifdef USE_SIGNAL_TIMEOUT
1694 # ifdef HAVE_SIGSETJMP
1695 # define SETJMP(env) sigsetjmp (env, 1)
1697 static sigjmp_buf run_with_timeout_env;
1700 abort_run_with_timeout (int sig)
1702 assert (sig == SIGALRM);
1703 siglongjmp (run_with_timeout_env, -1);
1705 # else /* not HAVE_SIGSETJMP */
1706 # define SETJMP(env) setjmp (env)
1708 static jmp_buf run_with_timeout_env;
1711 abort_run_with_timeout (int sig)
1713 assert (sig == SIGALRM);
1714 /* We don't have siglongjmp to preserve the set of blocked signals;
1715 if we longjumped out of the handler at this point, SIGALRM would
1716 remain blocked. We must unblock it manually. */
1717 int mask = siggetmask ();
1718 mask &= ~sigmask (SIGALRM);
1721 /* Now it's safe to longjump. */
1722 longjmp (run_with_timeout_env, -1);
1724 # endif /* not HAVE_SIGSETJMP */
1726 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1727 setitimer where available, alarm otherwise.
1729 TIMEOUT should be non-zero. If the timeout value is so small that
1730 it would be rounded to zero, it is rounded to the least legal value
1731 instead (1us for setitimer, 1s for alarm). That ensures that
1732 SIGALRM will be delivered in all cases. */
1735 alarm_set (double timeout)
1738 /* Use the modern itimer interface. */
1739 struct itimerval itv;
1741 itv.it_value.tv_sec = (long) timeout;
1742 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1743 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1744 /* Ensure that we wait for at least the minimum interval.
1745 Specifying zero would mean "wait forever". */
1746 itv.it_value.tv_usec = 1;
1747 setitimer (ITIMER_REAL, &itv, NULL);
1748 #else /* not ITIMER_REAL */
1749 /* Use the old alarm() interface. */
1750 int secs = (int) timeout;
1752 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1753 because alarm(0) means "never deliver the alarm", i.e. "wait
1754 forever", which is not what someone who specifies a 0.5s
1755 timeout would expect. */
1758 #endif /* not ITIMER_REAL */
1761 /* Cancel the alarm set with alarm_set. */
1767 struct itimerval disable;
1769 setitimer (ITIMER_REAL, &disable, NULL);
1770 #else /* not ITIMER_REAL */
1772 #endif /* not ITIMER_REAL */
1775 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1776 seconds. Returns true if the function was interrupted with a
1777 timeout, false otherwise.
1779 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1780 using setitimer() or alarm(). The timeout is enforced by
1781 longjumping out of the SIGALRM handler. This has several
1782 advantages compared to the traditional approach of relying on
1783 signals causing system calls to exit with EINTR:
1785 * The callback function is *forcibly* interrupted after the
1786 timeout expires, (almost) regardless of what it was doing and
1787 whether it was in a syscall. For example, a calculation that
1788 takes a long time is interrupted as reliably as an IO
1791 * It works with both SYSV and BSD signals because it doesn't
1792 depend on the default setting of SA_RESTART.
1794 * It doesn't require special handler setup beyond a simple call
1795 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1798 The only downside is that, if FUN allocates internal resources that
1799 are normally freed prior to exit from the functions, they will be
1800 lost in case of timeout. */
1803 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1813 signal (SIGALRM, abort_run_with_timeout);
1814 if (SETJMP (run_with_timeout_env) != 0)
1816 /* Longjumped out of FUN with a timeout. */
1817 signal (SIGALRM, SIG_DFL);
1820 alarm_set (timeout);
1823 /* Preserve errno in case alarm() or signal() modifies it. */
1824 saved_errno = errno;
1826 signal (SIGALRM, SIG_DFL);
1827 errno = saved_errno;
1832 #else /* not USE_SIGNAL_TIMEOUT */
1835 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1836 define it under Windows, because Windows has its own version of
1837 run_with_timeout that uses threads. */
1840 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1845 #endif /* not WINDOWS */
1846 #endif /* not USE_SIGNAL_TIMEOUT */
1850 /* Sleep the specified amount of seconds. On machines without
1851 nanosleep(), this may sleep shorter if interrupted by signals. */
1854 xsleep (double seconds)
1856 #ifdef HAVE_NANOSLEEP
1857 /* nanosleep is the preferred interface because it offers high
1858 accuracy and, more importantly, because it allows us to reliably
1859 restart receiving a signal such as SIGWINCH. (There was an
1860 actual Debian bug report about --limit-rate malfunctioning while
1861 the terminal was being resized.) */
1862 struct timespec sleep, remaining;
1863 sleep.tv_sec = (long) seconds;
1864 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
1865 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1866 /* If nanosleep has been interrupted by a signal, adjust the
1867 sleeping period and return to sleep. */
1869 #elif defined(HAVE_USLEEP)
1870 /* If usleep is available, use it in preference to select. */
1873 /* On some systems, usleep cannot handle values larger than
1874 1,000,000. If the period is larger than that, use sleep
1875 first, then add usleep for subsecond accuracy. */
1877 seconds -= (long) seconds;
1879 usleep (seconds * 1000000);
1880 #else /* fall back select */
1881 /* Note that, although Windows supports select, it can't be used to
1882 implement sleeping because Winsock's select doesn't implement
1883 timeout when it is passed NULL pointers for all fd sets. (But it
1884 does under Cygwin, which implements Unix-compatible select.) */
1885 struct timeval sleep;
1886 sleep.tv_sec = (long) seconds;
1887 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
1888 select (0, NULL, NULL, NULL, &sleep);
1889 /* If select returns -1 and errno is EINTR, it means we were
1890 interrupted by a signal. But without knowing how long we've
1891 actually slept, we can't return to sleep. Using gettimeofday to
1892 track sleeps is slow and unreliable due to clock skew. */
1896 #endif /* not WINDOWS */
1898 /* Encode the string STR of length LENGTH to base64 format and place it
1899 to B64STORE. The output will be \0-terminated, and must point to a
1900 writable buffer of at least 1+BASE64_LENGTH(length) bytes. It
1901 returns the length of the resulting base64 data, not counting the
1904 This implementation will not emit newlines after 76 characters of
1908 base64_encode (const char *str, int length, char *b64store)
1910 /* Conversion table. */
1911 static char tbl[64] = {
1912 'A','B','C','D','E','F','G','H',
1913 'I','J','K','L','M','N','O','P',
1914 'Q','R','S','T','U','V','W','X',
1915 'Y','Z','a','b','c','d','e','f',
1916 'g','h','i','j','k','l','m','n',
1917 'o','p','q','r','s','t','u','v',
1918 'w','x','y','z','0','1','2','3',
1919 '4','5','6','7','8','9','+','/'
1922 const unsigned char *s = (const unsigned char *) str;
1925 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1926 for (i = 0; i < length; i += 3)
1928 *p++ = tbl[s[0] >> 2];
1929 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1930 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1931 *p++ = tbl[s[2] & 0x3f];
1935 /* Pad the result if necessary... */
1936 if (i == length + 1)
1938 else if (i == length + 2)
1939 *(p - 1) = *(p - 2) = '=';
1941 /* ...and zero-terminate it. */
1944 return p - b64store;
1947 /* Store in C the next non-whitespace character from the string, or \0
1948 when end of string is reached. */
1949 #define NEXT_CHAR(c, p) do { \
1950 c = (unsigned char) *p++; \
1951 } while (ISSPACE (c))
1953 #define IS_ASCII(c) (((c) & 0x80) == 0)
1955 /* Decode data from BASE64 (pointer to \0-terminated text) into memory
1956 pointed to by TO. TO should be large enough to accomodate the
1957 decoded data, which is guaranteed to be less than strlen(base64).
1959 Since TO is assumed to contain binary data, it is not
1960 NUL-terminated. The function returns the length of the data
1961 written to TO. -1 is returned in case of error caused by malformed
1965 base64_decode (const char *base64, char *to)
1967 /* Table of base64 values for first 128 characters. Note that this
1968 assumes ASCII (but so does Wget in other places). */
1969 static signed char base64_char_to_value[128] =
1971 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
1972 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
1973 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
1974 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
1975 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
1976 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
1977 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
1978 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
1979 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
1980 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
1981 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
1982 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
1983 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
1985 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
1986 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
1988 const char *p = base64;
1994 unsigned long value;
1996 /* Process first byte of a quadruplet. */
2000 if (c == '=' || !IS_BASE64 (c))
2001 return -1; /* illegal char while decoding base64 */
2002 value = BASE64_CHAR_TO_VALUE (c) << 18;
2004 /* Process second byte of a quadruplet. */
2007 return -1; /* premature EOF while decoding base64 */
2008 if (c == '=' || !IS_BASE64 (c))
2009 return -1; /* illegal char while decoding base64 */
2010 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2013 /* Process third byte of a quadruplet. */
2016 return -1; /* premature EOF while decoding base64 */
2018 return -1; /* illegal char while decoding base64 */
2024 return -1; /* premature EOF while decoding base64 */
2026 return -1; /* padding `=' expected but not found */
2030 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2031 *q++ = 0xff & value >> 8;
2033 /* Process fourth byte of a quadruplet. */
2036 return -1; /* premature EOF while decoding base64 */
2040 return -1; /* illegal char while decoding base64 */
2042 value |= BASE64_CHAR_TO_VALUE (c);
2043 *q++ = 0xff & value;
2046 #undef BASE64_CHAR_TO_VALUE
2054 /* Simple merge sort for use by stable_sort. Implementation courtesy
2055 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2058 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2059 int (*cmpfun) (const void *, const void *))
2061 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2065 size_t mid = (to + from) / 2;
2066 mergesort_internal (base, temp, size, from, mid, cmpfun);
2067 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2070 for (k = from; (i <= mid) && (j <= to); k++)
2071 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2072 memcpy (ELT (temp, k), ELT (base, i++), size);
2074 memcpy (ELT (temp, k), ELT (base, j++), size);
2076 memcpy (ELT (temp, k++), ELT (base, i++), size);
2078 memcpy (ELT (temp, k++), ELT (base, j++), size);
2079 for (k = from; k <= to; k++)
2080 memcpy (ELT (base, k), ELT (temp, k), size);
2085 /* Stable sort with interface exactly like standard library's qsort.
2086 Uses mergesort internally, allocating temporary storage with
2090 stable_sort (void *base, size_t nmemb, size_t size,
2091 int (*cmpfun) (const void *, const void *))
2095 void *temp = alloca (nmemb * size * sizeof (void *));
2096 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2100 /* Print a decimal number. If it is equal to or larger than ten, the
2101 number is rounded. Otherwise it is printed with one significant
2102 digit without trailing zeros and with no more than three fractional
2103 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2104 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2106 This is useful for displaying durations because it provides
2107 order-of-magnitude information without unnecessary clutter --
2108 long-running downloads are shown without the fractional part, and
2109 short ones still retain one significant digit. */
2112 print_decimal (double number)
2114 static char buf[32];
2115 double n = number >= 0 ? number : -number;
2118 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2119 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2120 snprintf (buf, sizeof buf, "%.0f", number);
2122 snprintf (buf, sizeof buf, "%.1f", number);
2123 else if (n >= 0.001)
2124 snprintf (buf, sizeof buf, "%.1g", number);
2125 else if (n >= 0.0005)
2126 /* round [0.0005, 0.001) to 0.001 */
2127 snprintf (buf, sizeof buf, "%.3f", number);
2129 /* print numbers close to 0 as 0, not 0.000 */