1 /* Various utility functions.
2 Copyright (C) 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
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, 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 #else /* not HAVE_STRING_H */
38 #endif /* not HAVE_STRING_H */
39 #include <sys/types.h>
44 # include <sys/mman.h>
55 #ifdef HAVE_SYS_UTIME_H
56 # include <sys/utime.h>
60 # include <libc.h> /* for access() */
64 #ifdef WGET_USE_STDARG
70 /* For TIOCGWINSZ and friends: */
71 #ifdef HAVE_SYS_IOCTL_H
72 # include <sys/ioctl.h>
78 /* Needed for run_with_timeout. */
79 #undef USE_SIGNAL_TIMEOUT
87 #ifndef HAVE_SIGSETJMP
88 /* If sigsetjmp is a macro, configure won't pick it up. */
90 # define HAVE_SIGSETJMP
95 # ifdef HAVE_SIGSETJMP
96 # define USE_SIGNAL_TIMEOUT
99 # define USE_SIGNAL_TIMEOUT
111 /* Utility function: like xstrdup(), but also lowercases S. */
114 xstrdup_lower (const char *s)
116 char *copy = xstrdup (s);
123 /* Copy the string formed by two pointers (one on the beginning, other
124 on the char after the last char) to a new, malloc-ed location.
127 strdupdelim (const char *beg, const char *end)
129 char *res = (char *)xmalloc (end - beg + 1);
130 memcpy (res, beg, end - beg);
131 res[end - beg] = '\0';
135 /* Parse a string containing comma-separated elements, and return a
136 vector of char pointers with the elements. Spaces following the
137 commas are ignored. */
139 sepstring (const char *s)
153 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
154 res[i] = strdupdelim (p, s);
157 /* Skip the blanks following the ','. */
165 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
166 res[i] = strdupdelim (p, s);
171 #ifdef WGET_USE_STDARG
172 # define VA_START(args, arg1) va_start (args, arg1)
174 # define VA_START(args, ignored) va_start (args)
177 /* Like sprintf, but allocates a string of sufficient size with malloc
178 and returns it. GNU libc has a similar function named asprintf,
179 which requires the pointer to the string to be passed. */
182 aprintf (const char *fmt, ...)
184 /* This function is implemented using vsnprintf, which we provide
185 for the systems that don't have it. Therefore, it should be 100%
189 char *str = xmalloc (size);
196 /* See log_vprintf_internal for explanation why it's OK to rely
197 on the return value of vsnprintf. */
199 VA_START (args, fmt);
200 n = vsnprintf (str, size, fmt, args);
203 /* If the printing worked, return the string. */
204 if (n > -1 && n < size)
207 /* Else try again with a larger buffer. */
208 if (n > -1) /* C99 */
209 size = n + 1; /* precisely what is needed */
211 size <<= 1; /* twice the old size */
212 str = xrealloc (str, size);
214 return NULL; /* unreached */
217 /* Concatenate the NULL-terminated list of string arguments into
218 freshly allocated space. */
221 concat_strings (const char *str0, ...)
224 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
227 const char *next_str;
228 int total_length = 0;
231 /* Calculate the length of and allocate the resulting string. */
234 VA_START (args, str0);
235 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
237 int len = strlen (next_str);
238 if (argcount < countof (saved_lengths))
239 saved_lengths[argcount++] = len;
243 p = ret = xmalloc (total_length + 1);
245 /* Copy the strings into the allocated space. */
248 VA_START (args, str0);
249 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
252 if (argcount < countof (saved_lengths))
253 len = saved_lengths[argcount++];
255 len = strlen (next_str);
256 memcpy (p, next_str, len);
265 /* Return pointer to a static char[] buffer in which zero-terminated
266 string-representation of TM (in form hh:mm:ss) is printed.
268 If TM is NULL, the current time will be used. */
271 time_str (time_t *tm)
273 static char output[15];
275 time_t secs = tm ? *tm : time (NULL);
279 /* In case of error, return the empty string. Maybe we should
280 just abort if this happens? */
284 ptm = localtime (&secs);
285 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
289 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
292 datetime_str (time_t *tm)
294 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
296 time_t secs = tm ? *tm : time (NULL);
300 /* In case of error, return the empty string. Maybe we should
301 just abort if this happens? */
305 ptm = localtime (&secs);
306 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
307 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
308 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
312 /* The Windows versions of the following two functions are defined in
317 fork_to_background (void)
320 /* Whether we arrange our own version of opt.lfilename here. */
321 int logfile_changed = 0;
325 /* We must create the file immediately to avoid either a race
326 condition (which arises from using unique_name and failing to
327 use fopen_excl) or lying to the user about the log file name
328 (which arises from using unique_name, printing the name, and
329 using fopen_excl later on.) */
330 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, 0, &opt.lfilename);
346 /* parent, no error */
347 printf (_("Continuing in background, pid %d.\n"), (int)pid);
349 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
350 exit (0); /* #### should we use _exit()? */
353 /* child: give up the privileges and keep running. */
355 freopen ("/dev/null", "r", stdin);
356 freopen ("/dev/null", "w", stdout);
357 freopen ("/dev/null", "w", stderr);
359 #endif /* not WINDOWS */
361 /* "Touch" FILE, i.e. make its atime and mtime equal to the time
362 specified with TM. */
364 touch (const char *file, time_t tm)
366 #ifdef HAVE_STRUCT_UTIMBUF
367 struct utimbuf times;
368 times.actime = times.modtime = tm;
371 times[0] = times[1] = tm;
374 if (utime (file, ×) == -1)
375 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
378 /* Checks if FILE is a symbolic link, and removes it if it is. Does
379 nothing under MS-Windows. */
381 remove_link (const char *file)
386 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
388 DEBUGP (("Unlinking %s (symlink).\n", file));
391 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
392 file, strerror (errno));
397 /* Does FILENAME exist? This is quite a lousy implementation, since
398 it supplies no error codes -- only a yes-or-no answer. Thus it
399 will return that a file does not exist if, e.g., the directory is
400 unreadable. I don't mind it too much currently, though. The
401 proper way should, of course, be to have a third, error state,
402 other than true/false, but that would introduce uncalled-for
403 additional complexity to the callers. */
405 file_exists_p (const char *filename)
408 return access (filename, F_OK) >= 0;
411 return stat (filename, &buf) >= 0;
415 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
416 Returns 0 on error. */
418 file_non_directory_p (const char *path)
421 /* Use lstat() rather than stat() so that symbolic links pointing to
422 directories can be identified correctly. */
423 if (lstat (path, &buf) != 0)
425 return S_ISDIR (buf.st_mode) ? 0 : 1;
428 /* Return the size of file named by FILENAME, or -1 if it cannot be
429 opened or seeked into. */
431 file_size (const char *filename)
433 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
435 /* We use fseek rather than stat to determine the file size because
436 that way we can also verify that the file is readable without
437 explicitly checking for permissions. Inspired by the POST patch
439 FILE *fp = fopen (filename, "rb");
442 fseeko (fp, 0, SEEK_END);
448 if (stat (filename, &st) < 0)
454 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
455 doesn't exist is found. Return a freshly allocated copy of the
459 unique_name_1 (const char *prefix)
462 int plen = strlen (prefix);
463 char *template = (char *)alloca (plen + 1 + 24);
464 char *template_tail = template + plen;
466 memcpy (template, prefix, plen);
467 *template_tail++ = '.';
470 number_to_string (template_tail, count++);
471 while (file_exists_p (template));
473 return xstrdup (template);
476 /* Return a unique file name, based on FILE.
478 More precisely, if FILE doesn't exist, it is returned unmodified.
479 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
480 file name that doesn't exist is returned.
482 The resulting file is not created, only verified that it didn't
483 exist at the point in time when the function was called.
484 Therefore, where security matters, don't rely that the file created
485 by this function exists until you open it with O_EXCL or
488 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
489 string. Otherwise, it may return FILE if the file doesn't exist
490 (and therefore doesn't need changing). */
493 unique_name (const char *file, int allow_passthrough)
495 /* If the FILE itself doesn't exist, return it without
497 if (!file_exists_p (file))
498 return allow_passthrough ? (char *)file : xstrdup (file);
500 /* Otherwise, find a numeric suffix that results in unused file name
502 return unique_name_1 (file);
505 /* Create a file based on NAME, except without overwriting an existing
506 file with that name. Providing O_EXCL is correctly implemented,
507 this function does not have the race condition associated with
508 opening the file returned by unique_name. */
511 unique_create (const char *name, int binary, char **opened_name)
513 /* unique file name, based on NAME */
514 char *uname = unique_name (name, 0);
516 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
519 uname = unique_name (name, 0);
521 if (opened_name && fp != NULL)
524 *opened_name = uname;
536 /* Open the file for writing, with the addition that the file is
537 opened "exclusively". This means that, if the file already exists,
538 this function will *fail* and errno will be set to EEXIST. If
539 BINARY is set, the file will be opened in binary mode, equivalent
542 If opening the file fails for any reason, including the file having
543 previously existed, this function returns NULL and sets errno
547 fopen_excl (const char *fname, int binary)
551 int flags = O_WRONLY | O_CREAT | O_EXCL;
556 fd = open (fname, flags, 0666);
559 return fdopen (fd, binary ? "wb" : "w");
560 #else /* not O_EXCL */
561 /* Manually check whether the file exists. This is prone to race
562 conditions, but systems without O_EXCL haven't deserved
564 if (file_exists_p (fname))
569 return fopen (fname, binary ? "wb" : "w");
570 #endif /* not O_EXCL */
573 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
574 are missing, create them first. In case any mkdir() call fails,
575 return its error status. Returns 0 on successful completion.
577 The behaviour of this function should be identical to the behaviour
578 of `mkdir -p' on systems where mkdir supports the `-p' option. */
580 make_directory (const char *directory)
582 int i, ret, quit = 0;
585 /* Make a copy of dir, to be able to write to it. Otherwise, the
586 function is unsafe if called with a read-only char *argument. */
587 STRDUP_ALLOCA (dir, directory);
589 /* If the first character of dir is '/', skip it (and thus enable
590 creation of absolute-pathname directories. */
591 for (i = (*dir == '/'); 1; ++i)
593 for (; dir[i] && dir[i] != '/'; i++)
598 /* Check whether the directory already exists. Allow creation of
599 of intermediate directories to fail, as the initial path components
600 are not necessarily directories! */
601 if (!file_exists_p (dir))
602 ret = mkdir (dir, 0777);
613 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
614 should be a file name.
616 file_merge("/foo/bar", "baz") => "/foo/baz"
617 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
618 file_merge("foo", "bar") => "bar"
620 In other words, it's a simpler and gentler version of uri_merge_1. */
623 file_merge (const char *base, const char *file)
626 const char *cut = (const char *)strrchr (base, '/');
629 return xstrdup (file);
631 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
632 memcpy (result, base, cut - base);
633 result[cut - base] = '/';
634 strcpy (result + (cut - base) + 1, file);
639 static int in_acclist PARAMS ((const char *const *, const char *, int));
641 /* Determine whether a file is acceptable to be followed, according to
642 lists of patterns to accept/reject. */
644 acceptable (const char *s)
648 while (l && s[l] != '/')
655 return (in_acclist ((const char *const *)opt.accepts, s, 1)
656 && !in_acclist ((const char *const *)opt.rejects, s, 1));
658 return in_acclist ((const char *const *)opt.accepts, s, 1);
660 else if (opt.rejects)
661 return !in_acclist ((const char *const *)opt.rejects, s, 1);
665 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
666 `/something', frontcmp() will return 1 only if S2 begins with
667 `/something'. Otherwise, 0 is returned. */
669 frontcmp (const char *s1, const char *s2)
671 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
675 /* Iterate through STRLIST, and return the first element that matches
676 S, through wildcards or front comparison (as appropriate). */
678 proclist (char **strlist, const char *s, enum accd flags)
682 for (x = strlist; *x; x++)
683 if (has_wildcards_p (*x))
685 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
690 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
697 /* Returns whether DIRECTORY is acceptable for download, wrt the
698 include/exclude lists.
700 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
701 and absolute paths may be freely intermixed. */
703 accdir (const char *directory, enum accd flags)
705 /* Remove starting '/'. */
706 if (flags & ALLABS && *directory == '/')
710 if (!proclist (opt.includes, directory, flags))
715 if (proclist (opt.excludes, directory, flags))
721 /* Return non-zero if STRING ends with TAIL. For instance:
723 match_tail ("abc", "bc", 0) -> 1
724 match_tail ("abc", "ab", 0) -> 0
725 match_tail ("abc", "abc", 0) -> 1
727 If FOLD_CASE_P is non-zero, the comparison will be
731 match_tail (const char *string, const char *tail, int fold_case_p)
735 /* We want this to be fast, so we code two loops, one with
736 case-folding, one without. */
740 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
741 if (string[i] != tail[j])
746 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
747 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
751 /* If the tail was exhausted, the match was succesful. */
758 /* Checks whether string S matches each element of ACCEPTS. A list
759 element are matched either with fnmatch() or match_tail(),
760 according to whether the element contains wildcards or not.
762 If the BACKWARD is 0, don't do backward comparison -- just compare
765 in_acclist (const char *const *accepts, const char *s, int backward)
767 for (; *accepts; accepts++)
769 if (has_wildcards_p (*accepts))
771 /* fnmatch returns 0 if the pattern *does* match the
773 if (fnmatch (*accepts, s, 0) == 0)
780 if (match_tail (s, *accepts, 0))
785 if (!strcmp (s, *accepts))
793 /* Return the location of STR's suffix (file extension). Examples:
794 suffix ("foo.bar") -> "bar"
795 suffix ("foo.bar.baz") -> "baz"
796 suffix ("/foo/bar") -> NULL
797 suffix ("/foo.bar/baz") -> NULL */
799 suffix (const char *str)
803 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
807 return (char *)str + i;
812 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
816 has_wildcards_p (const char *s)
819 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
824 /* Return non-zero if FNAME ends with a typical HTML suffix. The
825 following (case-insensitive) suffixes are presumed to be HTML files:
829 ?html (`?' matches one character)
831 #### CAVEAT. This is not necessarily a good indication that FNAME
832 refers to a file that contains HTML! */
834 has_html_suffix_p (const char *fname)
838 if ((suf = suffix (fname)) == NULL)
840 if (!strcasecmp (suf, "html"))
842 if (!strcasecmp (suf, "htm"))
844 if (suf[0] && !strcasecmp (suf + 1, "html"))
849 /* Read a line from FP and return the pointer to freshly allocated
850 storage. The storage space is obtained through malloc() and should
851 be freed with free() when it is no longer needed.
853 The length of the line is not limited, except by available memory.
854 The newline character at the end of line is retained. The line is
855 terminated with a zero character.
857 After end-of-file is encountered without anything being read, NULL
858 is returned. NULL is also returned on error. To distinguish
859 between these two cases, use the stdio function ferror(). */
862 read_whole_line (FILE *fp)
866 char *line = (char *)xmalloc (bufsize);
868 while (fgets (line + length, bufsize - length, fp))
870 length += strlen (line + length);
872 /* Possible for example when reading from a binary file where
873 a line begins with \0. */
876 if (line[length - 1] == '\n')
879 /* fgets() guarantees to read the whole line, or to use up the
880 space we've given it. We can double the buffer
883 line = xrealloc (line, bufsize);
885 if (length == 0 || ferror (fp))
890 if (length + 1 < bufsize)
891 /* Relieve the memory from our exponential greediness. We say
892 `length + 1' because the terminating \0 is not included in
893 LENGTH. We don't need to zero-terminate the string ourselves,
894 though, because fgets() does that. */
895 line = xrealloc (line, length + 1);
899 /* Read FILE into memory. A pointer to `struct file_memory' are
900 returned; use struct element `content' to access file contents, and
901 the element `length' to know the file length. `content' is *not*
902 zero-terminated, and you should *not* read or write beyond the [0,
903 length) range of characters.
905 After you are done with the file contents, call read_file_free to
908 Depending on the operating system and the type of file that is
909 being read, read_file() either mmap's the file into memory, or
910 reads the file into the core using read().
912 If file is named "-", fileno(stdin) is used for reading instead.
913 If you want to read from a real file named "-", use "./-" instead. */
916 read_file (const char *file)
919 struct file_memory *fm;
921 int inhibit_close = 0;
923 /* Some magic in the finest tradition of Perl and its kin: if FILE
924 is "-", just use stdin. */
929 /* Note that we don't inhibit mmap() in this case. If stdin is
930 redirected from a regular file, mmap() will still work. */
933 fd = open (file, O_RDONLY);
936 fm = xnew (struct file_memory);
941 if (fstat (fd, &buf) < 0)
943 fm->length = buf.st_size;
944 /* NOTE: As far as I know, the callers of this function never
945 modify the file text. Relying on this would enable us to
946 specify PROT_READ and MAP_SHARED for a marginal gain in
947 efficiency, but at some cost to generality. */
948 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
950 if (fm->content == (char *)MAP_FAILED)
960 /* The most common reason why mmap() fails is that FD does not point
961 to a plain file. However, it's also possible that mmap() doesn't
962 work for a particular type of file. Therefore, whenever mmap()
963 fails, we just fall back to the regular method. */
964 #endif /* HAVE_MMAP */
967 size = 512; /* number of bytes fm->contents can
968 hold at any given time. */
969 fm->content = xmalloc (size);
973 if (fm->length > size / 2)
975 /* #### I'm not sure whether the whole exponential-growth
976 thing makes sense with kernel read. On Linux at least,
977 read() refuses to read more than 4K from a file at a
978 single chunk anyway. But other Unixes might optimize it
979 better, and it doesn't *hurt* anything, so I'm leaving
982 /* Normally, we grow SIZE exponentially to make the number
983 of calls to read() and realloc() logarithmic in relation
984 to file size. However, read() can read an amount of data
985 smaller than requested, and it would be unreasonable to
986 double SIZE every time *something* was read. Therefore,
987 we double SIZE only when the length exceeds half of the
988 entire allocated size. */
990 fm->content = xrealloc (fm->content, size);
992 nread = read (fd, fm->content + fm->length, size - fm->length);
994 /* Successful read. */
1005 if (size > fm->length && fm->length != 0)
1006 /* Due to exponential growth of fm->content, the allocated region
1007 might be much larger than what is actually needed. */
1008 fm->content = xrealloc (fm->content, fm->length);
1015 xfree (fm->content);
1020 /* Release the resources held by FM. Specifically, this calls
1021 munmap() or xfree() on fm->content, depending whether mmap or
1022 malloc/read were used to read in the file. It also frees the
1023 memory needed to hold the FM structure itself. */
1026 read_file_free (struct file_memory *fm)
1031 munmap (fm->content, fm->length);
1036 xfree (fm->content);
1041 /* Free the pointers in a NULL-terminated vector of pointers, then
1042 free the pointer itself. */
1044 free_vec (char **vec)
1055 /* Append vector V2 to vector V1. The function frees V2 and
1056 reallocates V1 (thus you may not use the contents of neither
1057 pointer after the call). If V1 is NULL, V2 is returned. */
1059 merge_vecs (char **v1, char **v2)
1069 /* To avoid j == 0 */
1074 for (i = 0; v1[i]; i++);
1076 for (j = 0; v2[j]; j++);
1077 /* Reallocate v1. */
1078 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1079 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1084 /* A set of simple-minded routines to store strings in a linked list.
1085 This used to also be used for searching, but now we have hash
1088 /* It's a shame that these simple things like linked lists and hash
1089 tables (see hash.c) need to be implemented over and over again. It
1090 would be nice to be able to use the routines from glib -- see
1091 www.gtk.org for details. However, that would make Wget depend on
1092 glib, and I want to avoid dependencies to external libraries for
1093 reasons of convenience and portability (I suspect Wget is more
1094 portable than anything ever written for Gnome). */
1096 /* Append an element to the list. If the list has a huge number of
1097 elements, this can get slow because it has to find the list's
1098 ending. If you think you have to call slist_append in a loop,
1099 think about calling slist_prepend() followed by slist_nreverse(). */
1102 slist_append (slist *l, const char *s)
1104 slist *newel = xnew (slist);
1107 newel->string = xstrdup (s);
1112 /* Find the last element. */
1119 /* Prepend S to the list. Unlike slist_append(), this is O(1). */
1122 slist_prepend (slist *l, const char *s)
1124 slist *newel = xnew (slist);
1125 newel->string = xstrdup (s);
1130 /* Destructively reverse L. */
1133 slist_nreverse (slist *l)
1138 slist *next = l->next;
1146 /* Is there a specific entry in the list? */
1148 slist_contains (slist *l, const char *s)
1150 for (; l; l = l->next)
1151 if (!strcmp (l->string, s))
1156 /* Free the whole slist. */
1158 slist_free (slist *l)
1169 /* Sometimes it's useful to create "sets" of strings, i.e. special
1170 hash tables where you want to store strings as keys and merely
1171 query for their existence. Here is a set of utility routines that
1172 makes that transparent. */
1175 string_set_add (struct hash_table *ht, const char *s)
1177 /* First check whether the set element already exists. If it does,
1178 do nothing so that we don't have to free() the old element and
1179 then strdup() a new one. */
1180 if (hash_table_contains (ht, s))
1183 /* We use "1" as value. It provides us a useful and clear arbitrary
1184 value, and it consumes no memory -- the pointers to the same
1185 string "1" will be shared by all the key-value pairs in all `set'
1187 hash_table_put (ht, xstrdup (s), "1");
1190 /* Synonym for hash_table_contains... */
1193 string_set_contains (struct hash_table *ht, const char *s)
1195 return hash_table_contains (ht, s);
1199 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1206 string_set_free (struct hash_table *ht)
1208 hash_table_map (ht, string_set_free_mapper, NULL);
1209 hash_table_destroy (ht);
1213 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1220 /* Another utility function: call free() on all keys and values of HT. */
1223 free_keys_and_values (struct hash_table *ht)
1225 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1229 /* Add thousand separators to a number already in string form. Used
1230 by with_thousand_seps and with_thousand_seps_large. */
1233 add_thousand_seps (const char *repr)
1235 static char outbuf[48];
1240 /* Reset the pointers. */
1244 /* Ignore the sign for the purpose of adding thousand
1251 /* How many digits before the first separator? */
1252 mod = strlen (inptr) % 3;
1254 for (i = 0; i < mod; i++)
1255 *outptr++ = inptr[i];
1256 /* Now insert the rest of them, putting separator before every
1258 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1260 if (i % 3 == 0 && i1 != 0)
1262 *outptr++ = inptr[i1];
1264 /* Zero-terminate the string. */
1269 /* Return a static pointer to the number printed with thousand
1270 separators inserted at the right places. */
1273 with_thousand_seps (wgint l)
1276 /* Print the number into the buffer. */
1277 number_to_string (inbuf, l);
1278 return add_thousand_seps (inbuf);
1281 /* Write a string representation of LARGE_INT NUMBER into the provided
1284 It would be dangerous to use sprintf, because the code wouldn't
1285 work on a machine with gcc-provided long long support, but without
1286 libc support for "%lld". However, such old systems platforms
1287 typically lack snprintf and will end up using our version, which
1288 does support "%lld" whereever long longs are available. */
1291 large_int_to_string (char *buffer, int bufsize, LARGE_INT number)
1293 snprintf (buffer, bufsize, LARGE_INT_FMT, number);
1296 /* The same as with_thousand_seps, but works on LARGE_INT. */
1299 with_thousand_seps_large (LARGE_INT l)
1302 large_int_to_string (inbuf, sizeof (inbuf), l);
1303 return add_thousand_seps (inbuf);
1306 /* N, a byte quantity, is converted to a human-readable abberviated
1307 form a la sizes printed by `ls -lh'. The result is written to a
1308 static buffer, a pointer to which is returned.
1310 Unlike `with_thousand_seps', this approximates to the nearest unit.
1311 Quoting GNU libit: "Most people visually process strings of 3-4
1312 digits effectively, but longer strings of digits are more prone to
1313 misinterpretation. Hence, converting to an abbreviated form
1314 usually improves readability."
1316 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1317 original computer science meaning of "multiples of 1024".
1318 Multiples of 1000 would be useless since Wget already adds thousand
1319 separators for legibility. We don't use the "*bibyte" names
1320 invented in 1998, and seldom used in practice. Wikipedia's entry
1321 on kilobyte discusses this in some detail. */
1324 human_readable (wgint n)
1326 /* These suffixes are compatible with those of GNU `ls -lh'. */
1327 static char powers[] =
1329 'K', /* kilobyte, 2^10 bytes */
1330 'M', /* megabyte, 2^20 bytes */
1331 'G', /* gigabyte, 2^30 bytes */
1332 'T', /* terabyte, 2^40 bytes */
1333 'P', /* petabyte, 2^50 bytes */
1334 'E', /* exabyte, 2^60 bytes */
1339 /* If the quantity is smaller than 1K, just print it. */
1342 snprintf (buf, sizeof (buf), "%d", (int) n);
1346 /* Loop over powers, dividing N with 1024 in each iteration. This
1347 works unchanged for all sizes of wgint, while still avoiding
1348 non-portable `long double' arithmetic. */
1349 for (i = 0; i < countof (powers); i++)
1351 /* At each iteration N is greater than the *subsequent* power.
1352 That way N/1024.0 produces a decimal number in the units of
1354 if ((n >> 10) < 1024 || i == countof (powers) - 1)
1356 /* Must cast to long first because MS VC can't directly cast
1357 __int64 to double. (This is safe because N is known to
1359 double val = (double) (long) n / 1024.0;
1360 /* Print values smaller than 10 with one decimal digits, and
1361 others without any decimals. */
1362 snprintf (buf, sizeof (buf), "%.*f%c",
1363 val < 10 ? 1 : 0, val, powers[i]);
1368 return NULL; /* unreached */
1371 /* Count the digits in the provided number. Used to allocate space
1372 when printing numbers. */
1375 numdigit (wgint number)
1379 ++cnt; /* accomodate '-' */
1380 while ((number /= 10) != 0)
1385 #define ONE_DIGIT(figure) *p++ = n / (figure) + '0'
1386 #define ONE_DIGIT_ADVANCE(figure) (ONE_DIGIT (figure), n %= (figure))
1388 #define DIGITS_1(figure) ONE_DIGIT (figure)
1389 #define DIGITS_2(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_1 ((figure) / 10)
1390 #define DIGITS_3(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_2 ((figure) / 10)
1391 #define DIGITS_4(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_3 ((figure) / 10)
1392 #define DIGITS_5(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_4 ((figure) / 10)
1393 #define DIGITS_6(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_5 ((figure) / 10)
1394 #define DIGITS_7(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_6 ((figure) / 10)
1395 #define DIGITS_8(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_7 ((figure) / 10)
1396 #define DIGITS_9(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_8 ((figure) / 10)
1397 #define DIGITS_10(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_9 ((figure) / 10)
1399 /* DIGITS_<11-20> are only used on machines with 64-bit numbers. */
1401 #define DIGITS_11(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_10 ((figure) / 10)
1402 #define DIGITS_12(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_11 ((figure) / 10)
1403 #define DIGITS_13(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_12 ((figure) / 10)
1404 #define DIGITS_14(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_13 ((figure) / 10)
1405 #define DIGITS_15(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_14 ((figure) / 10)
1406 #define DIGITS_16(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_15 ((figure) / 10)
1407 #define DIGITS_17(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_16 ((figure) / 10)
1408 #define DIGITS_18(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_17 ((figure) / 10)
1409 #define DIGITS_19(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_18 ((figure) / 10)
1411 /* It is annoying that we have three different syntaxes for 64-bit constants:
1412 - nnnL for 64-bit systems, where they are of type long;
1413 - nnnLL for 32-bit systems that support long long;
1414 - nnnI64 for MS compiler on Windows, which doesn't support long long. */
1417 /* If long is large enough, use long constants. */
1418 # define C10000000000 10000000000L
1419 # define C100000000000 100000000000L
1420 # define C1000000000000 1000000000000L
1421 # define C10000000000000 10000000000000L
1422 # define C100000000000000 100000000000000L
1423 # define C1000000000000000 1000000000000000L
1424 # define C10000000000000000 10000000000000000L
1425 # define C100000000000000000 100000000000000000L
1426 # define C1000000000000000000 1000000000000000000L
1428 # if SIZEOF_LONG_LONG != 0
1429 /* Otherwise, if long long is available, use long long constants. */
1430 # define C10000000000 10000000000LL
1431 # define C100000000000 100000000000LL
1432 # define C1000000000000 1000000000000LL
1433 # define C10000000000000 10000000000000LL
1434 # define C100000000000000 100000000000000LL
1435 # define C1000000000000000 1000000000000000LL
1436 # define C10000000000000000 10000000000000000LL
1437 # define C100000000000000000 100000000000000000LL
1438 # define C1000000000000000000 1000000000000000000LL
1440 # if defined(WINDOWS)
1441 /* Use __int64 constants under Windows. */
1442 # define C10000000000 10000000000I64
1443 # define C100000000000 100000000000I64
1444 # define C1000000000000 1000000000000I64
1445 # define C10000000000000 10000000000000I64
1446 # define C100000000000000 100000000000000I64
1447 # define C1000000000000000 1000000000000000I64
1448 # define C10000000000000000 10000000000000000I64
1449 # define C100000000000000000 100000000000000000I64
1450 # define C1000000000000000000 1000000000000000000I64
1455 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1456 cases and to portably support strange sizes of wgint. */
1457 #if SIZEOF_LONG >= SIZEOF_WGINT
1458 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%ld", (long) (n))
1460 # if SIZEOF_LONG_LONG >= SIZEOF_WGINT
1461 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%lld", (long long) (n))
1464 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%I64", (__int64) (n))
1469 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1470 `sprintf(buffer, "%lld", (long long) number)', only typically much
1471 faster and portable to machines without long long.
1473 The speedup may make a difference in programs that frequently
1474 convert numbers to strings. Some implementations of sprintf,
1475 particularly the one in GNU libc, have been known to be extremely
1476 slow when converting integers to strings.
1478 Return the pointer to the location where the terminating zero was
1479 printed. (Equivalent to calling buffer+strlen(buffer) after the
1482 BUFFER should be big enough to accept as many bytes as you expect
1483 the number to take up. On machines with 64-bit longs the maximum
1484 needed size is 24 bytes. That includes the digits needed for the
1485 largest 64-bit number, the `-' sign in case it's negative, and the
1486 terminating '\0'. */
1489 number_to_string (char *buffer, wgint number)
1494 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1495 /* We are running in a strange or misconfigured environment. Let
1496 sprintf cope with it. */
1497 SPRINTF_WGINT (buffer, n);
1498 p += strlen (buffer);
1499 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1505 /* We cannot print a '-' and assign -n to n because -n would
1506 overflow. Let sprintf deal with this border case. */
1507 SPRINTF_WGINT (buffer, n);
1508 p += strlen (buffer);
1516 if (n < 10) { DIGITS_1 (1); }
1517 else if (n < 100) { DIGITS_2 (10); }
1518 else if (n < 1000) { DIGITS_3 (100); }
1519 else if (n < 10000) { DIGITS_4 (1000); }
1520 else if (n < 100000) { DIGITS_5 (10000); }
1521 else if (n < 1000000) { DIGITS_6 (100000); }
1522 else if (n < 10000000) { DIGITS_7 (1000000); }
1523 else if (n < 100000000) { DIGITS_8 (10000000); }
1524 else if (n < 1000000000) { DIGITS_9 (100000000); }
1525 #if SIZEOF_WGINT == 4
1526 /* wgint is four bytes long: we're done. */
1527 /* ``if (1)'' serves only to preserve editor indentation. */
1528 else if (1) { DIGITS_10 (1000000000); }
1530 /* wgint is 64 bits long -- make sure to process all the digits. */
1531 else if (n < C10000000000) { DIGITS_10 (1000000000); }
1532 else if (n < C100000000000) { DIGITS_11 (C10000000000); }
1533 else if (n < C1000000000000) { DIGITS_12 (C100000000000); }
1534 else if (n < C10000000000000) { DIGITS_13 (C1000000000000); }
1535 else if (n < C100000000000000) { DIGITS_14 (C10000000000000); }
1536 else if (n < C1000000000000000) { DIGITS_15 (C100000000000000); }
1537 else if (n < C10000000000000000) { DIGITS_16 (C1000000000000000); }
1538 else if (n < C100000000000000000) { DIGITS_17 (C10000000000000000); }
1539 else if (n < C1000000000000000000) { DIGITS_18 (C100000000000000000); }
1540 else { DIGITS_19 (C1000000000000000000); }
1544 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1550 #undef ONE_DIGIT_ADVANCE
1574 /* Print NUMBER to a statically allocated string and return a pointer
1575 to the printed representation.
1577 This function is intended to be used in conjunction with printf.
1578 It is hard to portably print wgint values:
1579 a) you cannot use printf("%ld", number) because wgint can be long
1580 long on 32-bit machines with LFS.
1581 b) you cannot use printf("%lld", number) because NUMBER could be
1582 long on 32-bit machines without LFS, or on 64-bit machines,
1583 which do not require LFS. Also, Windows doesn't support %lld.
1584 c) you cannot use printf("%j", (int_max_t) number) because not all
1585 versions of printf support "%j", the most notable being the one
1587 d) you cannot #define WGINT_FMT to the appropriate format and use
1588 printf(WGINT_FMT, number) because that would break translations
1589 for user-visible messages, such as printf("Downloaded: %d
1592 What you should use instead is printf("%s", number_to_static_string
1595 CAVEAT: since the function returns pointers to static data, you
1596 must be careful to copy its result before calling it again.
1597 However, to make it more useful with printf, the function maintains
1598 an internal ring of static buffers to return. That way things like
1599 printf("%s %s", number_to_static_string (num1),
1600 number_to_static_string (num2)) work as expected. Three buffers
1601 are currently used, which means that "%s %s %s" will work, but "%s
1602 %s %s %s" won't. If you need to print more than three wgints,
1603 bump the RING_SIZE (or rethink your message.) */
1606 number_to_static_string (wgint number)
1608 static char ring[RING_SIZE][24];
1610 char *buf = ring[ringpos];
1611 number_to_string (buf, number);
1612 ringpos = (ringpos + 1) % RING_SIZE;
1616 /* Support for timers. */
1618 #undef TIMER_WINDOWS
1619 #undef TIMER_GETTIMEOFDAY
1622 /* Depending on the OS and availability of gettimeofday(), one and
1623 only one of the above constants will be defined. Virtually all
1624 modern Unix systems will define TIMER_GETTIMEOFDAY; Windows will
1625 use TIMER_WINDOWS. TIMER_TIME is a catch-all method for
1626 non-Windows systems without gettimeofday. */
1629 # define TIMER_WINDOWS
1630 #else /* not WINDOWS */
1631 # ifdef HAVE_GETTIMEOFDAY
1632 # define TIMER_GETTIMEOFDAY
1636 #endif /* not WINDOWS */
1638 #ifdef TIMER_GETTIMEOFDAY
1639 typedef struct timeval wget_sys_time;
1643 typedef time_t wget_sys_time;
1646 #ifdef TIMER_WINDOWS
1648 DWORD lores; /* In case GetTickCount is used */
1649 LARGE_INTEGER hires; /* In case high-resolution timer is used */
1654 /* Whether the start time has been initialized. */
1657 /* The starting point in time which, subtracted from the current
1658 time, yields elapsed time. */
1659 wget_sys_time start;
1661 /* The most recent elapsed time, calculated by wtimer_elapsed().
1662 Measured in milliseconds. */
1663 double elapsed_last;
1665 /* Approximately, the time elapsed between the true start of the
1666 measurement and the time represented by START. */
1667 double elapsed_pre_start;
1670 #ifdef TIMER_WINDOWS
1672 /* Whether high-resolution timers are used. Set by wtimer_initialize_once
1673 the first time wtimer_allocate is called. */
1674 static int using_hires_timers;
1676 /* Frequency of high-resolution timers -- number of updates per
1677 millisecond. Calculated the first time wtimer_allocate is called
1678 provided that high-resolution timers are available. */
1679 static double hires_millisec_freq;
1681 /* The first time a timer is created, determine whether to use
1682 high-resolution timers. */
1685 wtimer_initialize_once (void)
1687 static int init_done;
1693 QueryPerformanceFrequency (&freq);
1694 if (freq.QuadPart != 0)
1696 using_hires_timers = 1;
1697 hires_millisec_freq = (double) freq.QuadPart / 1000.0;
1701 #endif /* TIMER_WINDOWS */
1703 /* Allocate a timer. Calling wtimer_read on the timer will return
1704 zero. It is not legal to call wtimer_update with a freshly
1705 allocated timer -- use wtimer_reset first. */
1708 wtimer_allocate (void)
1710 struct wget_timer *wt = xnew (struct wget_timer);
1713 #ifdef TIMER_WINDOWS
1714 wtimer_initialize_once ();
1720 /* Allocate a new timer and reset it. Return the new timer. */
1725 struct wget_timer *wt = wtimer_allocate ();
1730 /* Free the resources associated with the timer. Its further use is
1734 wtimer_delete (struct wget_timer *wt)
1739 /* Store system time to WST. */
1742 wtimer_sys_set (wget_sys_time *wst)
1744 #ifdef TIMER_GETTIMEOFDAY
1745 gettimeofday (wst, NULL);
1752 #ifdef TIMER_WINDOWS
1753 if (using_hires_timers)
1755 QueryPerformanceCounter (&wst->hires);
1759 /* Where hires counters are not available, use GetTickCount rather
1760 GetSystemTime, because it is unaffected by clock skew and simpler
1761 to use. Note that overflows don't affect us because we never use
1762 absolute values of the ticker, only the differences. */
1763 wst->lores = GetTickCount ();
1768 /* Reset timer WT. This establishes the starting point from which
1769 wtimer_elapsed() will return the number of elapsed milliseconds.
1770 It is allowed to reset a previously used timer. */
1773 wtimer_reset (struct wget_timer *wt)
1775 /* Set the start time to the current time. */
1776 wtimer_sys_set (&wt->start);
1777 wt->elapsed_last = 0;
1778 wt->elapsed_pre_start = 0;
1779 wt->initialized = 1;
1783 wtimer_sys_diff (wget_sys_time *wst1, wget_sys_time *wst2)
1785 #ifdef TIMER_GETTIMEOFDAY
1786 return ((double)(wst1->tv_sec - wst2->tv_sec) * 1000
1787 + (double)(wst1->tv_usec - wst2->tv_usec) / 1000);
1791 return 1000 * (*wst1 - *wst2);
1795 if (using_hires_timers)
1796 return (wst1->hires.QuadPart - wst2->hires.QuadPart) / hires_millisec_freq;
1798 return wst1->lores - wst2->lores;
1802 /* Update the timer's elapsed interval. This function causes the
1803 timer to call gettimeofday (or time(), etc.) to update its idea of
1804 current time. To get the elapsed interval in milliseconds, use
1807 This function handles clock skew, i.e. time that moves backwards is
1811 wtimer_update (struct wget_timer *wt)
1816 assert (wt->initialized != 0);
1818 wtimer_sys_set (&now);
1819 elapsed = wt->elapsed_pre_start + wtimer_sys_diff (&now, &wt->start);
1821 /* Ideally we'd just return the difference between NOW and
1822 wt->start. However, the system timer can be set back, and we
1823 could return a value smaller than when we were last called, even
1824 a negative value. Both of these would confuse the callers, which
1825 expect us to return monotonically nondecreasing values.
1827 Therefore: if ELAPSED is smaller than its previous known value,
1828 we reset wt->start to the current time and effectively start
1829 measuring from this point. But since we don't want the elapsed
1830 value to start from zero, we set elapsed_pre_start to the last
1831 elapsed time and increment all future calculations by that
1834 if (elapsed < wt->elapsed_last)
1837 wt->elapsed_pre_start = wt->elapsed_last;
1838 elapsed = wt->elapsed_last;
1841 wt->elapsed_last = elapsed;
1844 /* Return the elapsed time in milliseconds between the last call to
1845 wtimer_reset and the last call to wtimer_update.
1847 A typical use of the timer interface would be:
1849 struct wtimer *timer = wtimer_new ();
1850 ... do something that takes a while ...
1852 double msecs = wtimer_read (); */
1855 wtimer_read (const struct wget_timer *wt)
1857 return wt->elapsed_last;
1860 /* Return the assessed granularity of the timer implementation, in
1861 milliseconds. This is used by code that tries to substitute a
1862 better value for timers that have returned zero. */
1865 wtimer_granularity (void)
1867 #ifdef TIMER_GETTIMEOFDAY
1868 /* Granularity of gettimeofday varies wildly between architectures.
1869 However, it appears that on modern machines it tends to be better
1870 than 1ms. Assume 100 usecs. (Perhaps the configure process
1871 could actually measure this?) */
1879 #ifdef TIMER_WINDOWS
1880 if (using_hires_timers)
1881 return 1.0 / hires_millisec_freq;
1883 return 10; /* according to MSDN */
1887 /* This should probably be at a better place, but it doesn't really
1888 fit into html-parse.c. */
1890 /* The function returns the pointer to the malloc-ed quoted version of
1891 string s. It will recognize and quote numeric and special graphic
1892 entities, as per RFC1866:
1900 No other entities are recognized or replaced. */
1902 html_quote_string (const char *s)
1908 /* Pass through the string, and count the new size. */
1909 for (i = 0; *s; s++, i++)
1912 i += 4; /* `amp;' */
1913 else if (*s == '<' || *s == '>')
1914 i += 3; /* `lt;' and `gt;' */
1915 else if (*s == '\"')
1916 i += 5; /* `quot;' */
1920 res = (char *)xmalloc (i + 1);
1922 for (p = res; *s; s++)
1935 *p++ = (*s == '<' ? 'l' : 'g');
1962 /* Determine the width of the terminal we're running on. If that's
1963 not possible, return 0. */
1966 determine_screen_width (void)
1968 /* If there's a way to get the terminal size using POSIX
1969 tcgetattr(), somebody please tell me. */
1974 if (opt.lfilename != NULL)
1977 fd = fileno (stderr);
1978 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1979 return 0; /* most likely ENOTTY */
1982 #else /* not TIOCGWINSZ */
1984 CONSOLE_SCREEN_BUFFER_INFO csbi;
1985 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1987 return csbi.dwSize.X;
1988 # else /* neither WINDOWS nor TIOCGWINSZ */
1990 #endif /* neither WINDOWS nor TIOCGWINSZ */
1991 #endif /* not TIOCGWINSZ */
1994 /* Return a random number between 0 and MAX-1, inclusive.
1996 If MAX is greater than the value of RAND_MAX+1 on the system, the
1997 returned value will be in the range [0, RAND_MAX]. This may be
1998 fixed in a future release.
2000 The random number generator is seeded automatically the first time
2003 This uses rand() for portability. It has been suggested that
2004 random() offers better randomness, but this is not required for
2005 Wget, so I chose to go for simplicity and use rand
2008 DO NOT use this for cryptographic purposes. It is only meant to be
2009 used in situations where quality of the random numbers returned
2010 doesn't really matter. */
2013 random_number (int max)
2021 srand (time (NULL));
2026 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
2027 and enforce that assumption by masking other bits. */
2029 # define RAND_MAX 32767
2033 /* This is equivalent to rand() % max, but uses the high-order bits
2034 for better randomness on architecture where rand() is implemented
2035 using a simple congruential generator. */
2037 bounded = (double)max * rnd / (RAND_MAX + 1.0);
2038 return (int)bounded;
2041 /* Return a random uniformly distributed floating point number in the
2042 [0, 1) range. The precision of returned numbers is 9 digits.
2044 Modify this to use erand48() where available! */
2049 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
2050 sure it's greater than 1000. */
2051 int rnd1 = random_number (1000);
2052 int rnd2 = random_number (1000);
2053 int rnd3 = random_number (1000);
2054 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
2057 /* Implementation of run_with_timeout, a generic timeout-forcing
2058 routine for systems with Unix-like signal handling. */
2060 #ifdef USE_SIGNAL_TIMEOUT
2061 # ifdef HAVE_SIGSETJMP
2062 # define SETJMP(env) sigsetjmp (env, 1)
2064 static sigjmp_buf run_with_timeout_env;
2067 abort_run_with_timeout (int sig)
2069 assert (sig == SIGALRM);
2070 siglongjmp (run_with_timeout_env, -1);
2072 # else /* not HAVE_SIGSETJMP */
2073 # define SETJMP(env) setjmp (env)
2075 static jmp_buf run_with_timeout_env;
2078 abort_run_with_timeout (int sig)
2080 assert (sig == SIGALRM);
2081 /* We don't have siglongjmp to preserve the set of blocked signals;
2082 if we longjumped out of the handler at this point, SIGALRM would
2083 remain blocked. We must unblock it manually. */
2084 int mask = siggetmask ();
2085 mask &= ~sigmask (SIGALRM);
2088 /* Now it's safe to longjump. */
2089 longjmp (run_with_timeout_env, -1);
2091 # endif /* not HAVE_SIGSETJMP */
2093 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
2094 setitimer where available, alarm otherwise.
2096 TIMEOUT should be non-zero. If the timeout value is so small that
2097 it would be rounded to zero, it is rounded to the least legal value
2098 instead (1us for setitimer, 1s for alarm). That ensures that
2099 SIGALRM will be delivered in all cases. */
2102 alarm_set (double timeout)
2105 /* Use the modern itimer interface. */
2106 struct itimerval itv;
2108 itv.it_value.tv_sec = (long) timeout;
2109 itv.it_value.tv_usec = 1000000L * (timeout - (long)timeout);
2110 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
2111 /* Ensure that we wait for at least the minimum interval.
2112 Specifying zero would mean "wait forever". */
2113 itv.it_value.tv_usec = 1;
2114 setitimer (ITIMER_REAL, &itv, NULL);
2115 #else /* not ITIMER_REAL */
2116 /* Use the old alarm() interface. */
2117 int secs = (int) timeout;
2119 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
2120 because alarm(0) means "never deliver the alarm", i.e. "wait
2121 forever", which is not what someone who specifies a 0.5s
2122 timeout would expect. */
2125 #endif /* not ITIMER_REAL */
2128 /* Cancel the alarm set with alarm_set. */
2134 struct itimerval disable;
2136 setitimer (ITIMER_REAL, &disable, NULL);
2137 #else /* not ITIMER_REAL */
2139 #endif /* not ITIMER_REAL */
2142 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
2143 seconds. Returns non-zero if the function was interrupted with a
2144 timeout, zero otherwise.
2146 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
2147 using setitimer() or alarm(). The timeout is enforced by
2148 longjumping out of the SIGALRM handler. This has several
2149 advantages compared to the traditional approach of relying on
2150 signals causing system calls to exit with EINTR:
2152 * The callback function is *forcibly* interrupted after the
2153 timeout expires, (almost) regardless of what it was doing and
2154 whether it was in a syscall. For example, a calculation that
2155 takes a long time is interrupted as reliably as an IO
2158 * It works with both SYSV and BSD signals because it doesn't
2159 depend on the default setting of SA_RESTART.
2161 * It doesn't special handler setup beyond a simple call to
2162 signal(). (It does use sigsetjmp/siglongjmp, but they're
2165 The only downside is that, if FUN allocates internal resources that
2166 are normally freed prior to exit from the functions, they will be
2167 lost in case of timeout. */
2170 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2180 signal (SIGALRM, abort_run_with_timeout);
2181 if (SETJMP (run_with_timeout_env) != 0)
2183 /* Longjumped out of FUN with a timeout. */
2184 signal (SIGALRM, SIG_DFL);
2187 alarm_set (timeout);
2190 /* Preserve errno in case alarm() or signal() modifies it. */
2191 saved_errno = errno;
2193 signal (SIGALRM, SIG_DFL);
2194 errno = saved_errno;
2199 #else /* not USE_SIGNAL_TIMEOUT */
2202 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2203 define it under Windows, because Windows has its own version of
2204 run_with_timeout that uses threads. */
2207 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2212 #endif /* not WINDOWS */
2213 #endif /* not USE_SIGNAL_TIMEOUT */
2217 /* Sleep the specified amount of seconds. On machines without
2218 nanosleep(), this may sleep shorter if interrupted by signals. */
2221 xsleep (double seconds)
2223 #ifdef HAVE_NANOSLEEP
2224 /* nanosleep is the preferred interface because it offers high
2225 accuracy and, more importantly, because it allows us to reliably
2226 restart receiving a signal such as SIGWINCH. (There was an
2227 actual Debian bug report about --limit-rate malfunctioning while
2228 the terminal was being resized.) */
2229 struct timespec sleep, remaining;
2230 sleep.tv_sec = (long) seconds;
2231 sleep.tv_nsec = 1000000000L * (seconds - (long) seconds);
2232 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2233 /* If nanosleep has been interrupted by a signal, adjust the
2234 sleeping period and return to sleep. */
2236 #else /* not HAVE_NANOSLEEP */
2238 /* If usleep is available, use it in preference to select. */
2241 /* On some systems, usleep cannot handle values larger than
2242 1,000,000. If the period is larger than that, use sleep
2243 first, then add usleep for subsecond accuracy. */
2245 seconds -= (long) seconds;
2247 usleep (seconds * 1000000L);
2248 #else /* not HAVE_USLEEP */
2250 struct timeval sleep;
2251 sleep.tv_sec = (long) seconds;
2252 sleep.tv_usec = 1000000L * (seconds - (long) seconds);
2253 select (0, NULL, NULL, NULL, &sleep);
2254 /* If select returns -1 and errno is EINTR, it means we were
2255 interrupted by a signal. But without knowing how long we've
2256 actually slept, we can't return to sleep. Using gettimeofday to
2257 track sleeps is slow and unreliable due to clock skew. */
2258 #else /* not HAVE_SELECT */
2260 #endif /* not HAVE_SELECT */
2261 #endif /* not HAVE_USLEEP */
2262 #endif /* not HAVE_NANOSLEEP */
2265 #endif /* not WINDOWS */