1 /* Various utility functions.
2 Copyright (C) 2003 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 /* Return a count of how many times CHR occurs in STRING. */
126 count_char (const char *string, char chr)
130 for (p = string; *p; p++)
136 /* Copy the string formed by two pointers (one on the beginning, other
137 on the char after the last char) to a new, malloc-ed location.
140 strdupdelim (const char *beg, const char *end)
142 char *res = (char *)xmalloc (end - beg + 1);
143 memcpy (res, beg, end - beg);
144 res[end - beg] = '\0';
148 /* Parse a string containing comma-separated elements, and return a
149 vector of char pointers with the elements. Spaces following the
150 commas are ignored. */
152 sepstring (const char *s)
166 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
167 res[i] = strdupdelim (p, s);
170 /* Skip the blanks following the ','. */
178 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
179 res[i] = strdupdelim (p, s);
184 #ifdef WGET_USE_STDARG
185 # define VA_START(args, arg1) va_start (args, arg1)
187 # define VA_START(args, ignored) va_start (args)
190 /* Like sprintf, but allocates a string of sufficient size with malloc
191 and returns it. GNU libc has a similar function named asprintf,
192 which requires the pointer to the string to be passed. */
195 aprintf (const char *fmt, ...)
197 /* This function is implemented using vsnprintf, which we provide
198 for the systems that don't have it. Therefore, it should be 100%
202 char *str = xmalloc (size);
209 /* See log_vprintf_internal for explanation why it's OK to rely
210 on the return value of vsnprintf. */
212 VA_START (args, fmt);
213 n = vsnprintf (str, size, fmt, args);
216 /* If the printing worked, return the string. */
217 if (n > -1 && n < size)
220 /* Else try again with a larger buffer. */
221 if (n > -1) /* C99 */
222 size = n + 1; /* precisely what is needed */
224 size <<= 1; /* twice the old size */
225 str = xrealloc (str, size);
227 return NULL; /* unreached */
230 /* Return pointer to a static char[] buffer in which zero-terminated
231 string-representation of TM (in form hh:mm:ss) is printed.
233 If TM is NULL, the current time will be used. */
236 time_str (time_t *tm)
238 static char output[15];
240 time_t secs = tm ? *tm : time (NULL);
244 /* In case of error, return the empty string. Maybe we should
245 just abort if this happens? */
249 ptm = localtime (&secs);
250 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
254 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
257 datetime_str (time_t *tm)
259 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
261 time_t secs = tm ? *tm : time (NULL);
265 /* In case of error, return the empty string. Maybe we should
266 just abort if this happens? */
270 ptm = localtime (&secs);
271 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
272 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
273 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
277 /* The Windows versions of the following two functions are defined in
282 fork_to_background (void)
285 /* Whether we arrange our own version of opt.lfilename here. */
286 int logfile_changed = 0;
290 /* We must create the file immediately to avoid either a race
291 condition (which arises from using unique_name and failing to
292 use fopen_excl) or lying to the user about the log file name
293 (which arises from using unique_name, printing the name, and
294 using fopen_excl later on.) */
295 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, 0, &opt.lfilename);
311 /* parent, no error */
312 printf (_("Continuing in background, pid %d.\n"), (int)pid);
314 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
315 exit (0); /* #### should we use _exit()? */
318 /* child: give up the privileges and keep running. */
320 freopen ("/dev/null", "r", stdin);
321 freopen ("/dev/null", "w", stdout);
322 freopen ("/dev/null", "w", stderr);
324 #endif /* not WINDOWS */
326 /* "Touch" FILE, i.e. make its atime and mtime equal to the time
327 specified with TM. */
329 touch (const char *file, time_t tm)
331 #ifdef HAVE_STRUCT_UTIMBUF
332 struct utimbuf times;
333 times.actime = times.modtime = tm;
336 times[0] = times[1] = tm;
339 if (utime (file, ×) == -1)
340 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
343 /* Checks if FILE is a symbolic link, and removes it if it is. Does
344 nothing under MS-Windows. */
346 remove_link (const char *file)
351 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
353 DEBUGP (("Unlinking %s (symlink).\n", file));
356 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
357 file, strerror (errno));
362 /* Does FILENAME exist? This is quite a lousy implementation, since
363 it supplies no error codes -- only a yes-or-no answer. Thus it
364 will return that a file does not exist if, e.g., the directory is
365 unreadable. I don't mind it too much currently, though. The
366 proper way should, of course, be to have a third, error state,
367 other than true/false, but that would introduce uncalled-for
368 additional complexity to the callers. */
370 file_exists_p (const char *filename)
373 return access (filename, F_OK) >= 0;
376 return stat (filename, &buf) >= 0;
380 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
381 Returns 0 on error. */
383 file_non_directory_p (const char *path)
386 /* Use lstat() rather than stat() so that symbolic links pointing to
387 directories can be identified correctly. */
388 if (lstat (path, &buf) != 0)
390 return S_ISDIR (buf.st_mode) ? 0 : 1;
393 /* Return the size of file named by FILENAME, or -1 if it cannot be
394 opened or seeked into. */
396 file_size (const char *filename)
398 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
400 /* We use fseek rather than stat to determine the file size because
401 that way we can also verify that the file is readable without
402 explicitly checking for permissions. Inspired by the POST patch
404 FILE *fp = fopen (filename, "rb");
407 fseeko (fp, 0, SEEK_END);
413 if (stat (filename, &st) < 0)
419 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
420 doesn't exist is found. Return a freshly allocated copy of the
424 unique_name_1 (const char *prefix)
427 int plen = strlen (prefix);
428 char *template = (char *)alloca (plen + 1 + 24);
429 char *template_tail = template + plen;
431 memcpy (template, prefix, plen);
432 *template_tail++ = '.';
435 number_to_string (template_tail, count++);
436 while (file_exists_p (template));
438 return xstrdup (template);
441 /* Return a unique file name, based on FILE.
443 More precisely, if FILE doesn't exist, it is returned unmodified.
444 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
445 file name that doesn't exist is returned.
447 The resulting file is not created, only verified that it didn't
448 exist at the point in time when the function was called.
449 Therefore, where security matters, don't rely that the file created
450 by this function exists until you open it with O_EXCL or
453 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
454 string. Otherwise, it may return FILE if the file doesn't exist
455 (and therefore doesn't need changing). */
458 unique_name (const char *file, int allow_passthrough)
460 /* If the FILE itself doesn't exist, return it without
462 if (!file_exists_p (file))
463 return allow_passthrough ? (char *)file : xstrdup (file);
465 /* Otherwise, find a numeric suffix that results in unused file name
467 return unique_name_1 (file);
470 /* Create a file based on NAME, except without overwriting an existing
471 file with that name. Providing O_EXCL is correctly implemented,
472 this function does not have the race condition associated with
473 opening the file returned by unique_name. */
476 unique_create (const char *name, int binary, char **opened_name)
478 /* unique file name, based on NAME */
479 char *uname = unique_name (name, 0);
481 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
484 uname = unique_name (name, 0);
486 if (opened_name && fp != NULL)
489 *opened_name = uname;
501 /* Open the file for writing, with the addition that the file is
502 opened "exclusively". This means that, if the file already exists,
503 this function will *fail* and errno will be set to EEXIST. If
504 BINARY is set, the file will be opened in binary mode, equivalent
507 If opening the file fails for any reason, including the file having
508 previously existed, this function returns NULL and sets errno
512 fopen_excl (const char *fname, int binary)
515 int flags = O_WRONLY | O_CREAT | O_EXCL;
520 int fd = open (fname, flags, 0666);
523 return fdopen (fd, binary ? "wb" : "w");
524 #else /* not O_EXCL */
525 return fopen (fname, binary ? "wb" : "w");
526 #endif /* not O_EXCL */
529 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
530 are missing, create them first. In case any mkdir() call fails,
531 return its error status. Returns 0 on successful completion.
533 The behaviour of this function should be identical to the behaviour
534 of `mkdir -p' on systems where mkdir supports the `-p' option. */
536 make_directory (const char *directory)
543 /* Make a copy of dir, to be able to write to it. Otherwise, the
544 function is unsafe if called with a read-only char *argument. */
545 STRDUP_ALLOCA (dir, directory);
547 /* If the first character of dir is '/', skip it (and thus enable
548 creation of absolute-pathname directories. */
549 for (i = (*dir == '/'); 1; ++i)
551 for (; dir[i] && dir[i] != '/'; i++)
556 /* Check whether the directory already exists. Allow creation of
557 of intermediate directories to fail, as the initial path components
558 are not necessarily directories! */
559 if (!file_exists_p (dir))
560 ret = mkdir (dir, 0777);
571 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
572 should be a file name.
574 file_merge("/foo/bar", "baz") => "/foo/baz"
575 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
576 file_merge("foo", "bar") => "bar"
578 In other words, it's a simpler and gentler version of uri_merge_1. */
581 file_merge (const char *base, const char *file)
584 const char *cut = (const char *)strrchr (base, '/');
587 return xstrdup (file);
589 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
590 memcpy (result, base, cut - base);
591 result[cut - base] = '/';
592 strcpy (result + (cut - base) + 1, file);
597 static int in_acclist PARAMS ((const char *const *, const char *, int));
599 /* Determine whether a file is acceptable to be followed, according to
600 lists of patterns to accept/reject. */
602 acceptable (const char *s)
606 while (l && s[l] != '/')
613 return (in_acclist ((const char *const *)opt.accepts, s, 1)
614 && !in_acclist ((const char *const *)opt.rejects, s, 1));
616 return in_acclist ((const char *const *)opt.accepts, s, 1);
618 else if (opt.rejects)
619 return !in_acclist ((const char *const *)opt.rejects, s, 1);
623 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
624 `/something', frontcmp() will return 1 only if S2 begins with
625 `/something'. Otherwise, 0 is returned. */
627 frontcmp (const char *s1, const char *s2)
629 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
633 /* Iterate through STRLIST, and return the first element that matches
634 S, through wildcards or front comparison (as appropriate). */
636 proclist (char **strlist, const char *s, enum accd flags)
640 for (x = strlist; *x; x++)
641 if (has_wildcards_p (*x))
643 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
648 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
655 /* Returns whether DIRECTORY is acceptable for download, wrt the
656 include/exclude lists.
658 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
659 and absolute paths may be freely intermixed. */
661 accdir (const char *directory, enum accd flags)
663 /* Remove starting '/'. */
664 if (flags & ALLABS && *directory == '/')
668 if (!proclist (opt.includes, directory, flags))
673 if (proclist (opt.excludes, directory, flags))
679 /* Return non-zero if STRING ends with TAIL. For instance:
681 match_tail ("abc", "bc", 0) -> 1
682 match_tail ("abc", "ab", 0) -> 0
683 match_tail ("abc", "abc", 0) -> 1
685 If FOLD_CASE_P is non-zero, the comparison will be
689 match_tail (const char *string, const char *tail, int fold_case_p)
693 /* We want this to be fast, so we code two loops, one with
694 case-folding, one without. */
698 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
699 if (string[i] != tail[j])
704 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
705 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
709 /* If the tail was exhausted, the match was succesful. */
716 /* Checks whether string S matches each element of ACCEPTS. A list
717 element are matched either with fnmatch() or match_tail(),
718 according to whether the element contains wildcards or not.
720 If the BACKWARD is 0, don't do backward comparison -- just compare
723 in_acclist (const char *const *accepts, const char *s, int backward)
725 for (; *accepts; accepts++)
727 if (has_wildcards_p (*accepts))
729 /* fnmatch returns 0 if the pattern *does* match the
731 if (fnmatch (*accepts, s, 0) == 0)
738 if (match_tail (s, *accepts, 0))
743 if (!strcmp (s, *accepts))
751 /* Return the location of STR's suffix (file extension). Examples:
752 suffix ("foo.bar") -> "bar"
753 suffix ("foo.bar.baz") -> "baz"
754 suffix ("/foo/bar") -> NULL
755 suffix ("/foo.bar/baz") -> NULL */
757 suffix (const char *str)
761 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
765 return (char *)str + i;
770 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
774 has_wildcards_p (const char *s)
777 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
782 /* Return non-zero if FNAME ends with a typical HTML suffix. The
783 following (case-insensitive) suffixes are presumed to be HTML files:
787 ?html (`?' matches one character)
789 #### CAVEAT. This is not necessarily a good indication that FNAME
790 refers to a file that contains HTML! */
792 has_html_suffix_p (const char *fname)
796 if ((suf = suffix (fname)) == NULL)
798 if (!strcasecmp (suf, "html"))
800 if (!strcasecmp (suf, "htm"))
802 if (suf[0] && !strcasecmp (suf + 1, "html"))
807 /* Read a line from FP and return the pointer to freshly allocated
808 storage. The storage space is obtained through malloc() and should
809 be freed with free() when it is no longer needed.
811 The length of the line is not limited, except by available memory.
812 The newline character at the end of line is retained. The line is
813 terminated with a zero character.
815 After end-of-file is encountered without anything being read, NULL
816 is returned. NULL is also returned on error. To distinguish
817 between these two cases, use the stdio function ferror(). */
820 read_whole_line (FILE *fp)
824 char *line = (char *)xmalloc (bufsize);
826 while (fgets (line + length, bufsize - length, fp))
828 length += strlen (line + length);
830 /* Possible for example when reading from a binary file where
831 a line begins with \0. */
834 if (line[length - 1] == '\n')
837 /* fgets() guarantees to read the whole line, or to use up the
838 space we've given it. We can double the buffer
841 line = xrealloc (line, bufsize);
843 if (length == 0 || ferror (fp))
848 if (length + 1 < bufsize)
849 /* Relieve the memory from our exponential greediness. We say
850 `length + 1' because the terminating \0 is not included in
851 LENGTH. We don't need to zero-terminate the string ourselves,
852 though, because fgets() does that. */
853 line = xrealloc (line, length + 1);
857 /* Read FILE into memory. A pointer to `struct file_memory' are
858 returned; use struct element `content' to access file contents, and
859 the element `length' to know the file length. `content' is *not*
860 zero-terminated, and you should *not* read or write beyond the [0,
861 length) range of characters.
863 After you are done with the file contents, call read_file_free to
866 Depending on the operating system and the type of file that is
867 being read, read_file() either mmap's the file into memory, or
868 reads the file into the core using read().
870 If file is named "-", fileno(stdin) is used for reading instead.
871 If you want to read from a real file named "-", use "./-" instead. */
874 read_file (const char *file)
877 struct file_memory *fm;
879 int inhibit_close = 0;
881 /* Some magic in the finest tradition of Perl and its kin: if FILE
882 is "-", just use stdin. */
887 /* Note that we don't inhibit mmap() in this case. If stdin is
888 redirected from a regular file, mmap() will still work. */
891 fd = open (file, O_RDONLY);
894 fm = xnew (struct file_memory);
899 if (fstat (fd, &buf) < 0)
901 fm->length = buf.st_size;
902 /* NOTE: As far as I know, the callers of this function never
903 modify the file text. Relying on this would enable us to
904 specify PROT_READ and MAP_SHARED for a marginal gain in
905 efficiency, but at some cost to generality. */
906 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
908 if (fm->content == (char *)MAP_FAILED)
918 /* The most common reason why mmap() fails is that FD does not point
919 to a plain file. However, it's also possible that mmap() doesn't
920 work for a particular type of file. Therefore, whenever mmap()
921 fails, we just fall back to the regular method. */
922 #endif /* HAVE_MMAP */
925 size = 512; /* number of bytes fm->contents can
926 hold at any given time. */
927 fm->content = xmalloc (size);
931 if (fm->length > size / 2)
933 /* #### I'm not sure whether the whole exponential-growth
934 thing makes sense with kernel read. On Linux at least,
935 read() refuses to read more than 4K from a file at a
936 single chunk anyway. But other Unixes might optimize it
937 better, and it doesn't *hurt* anything, so I'm leaving
940 /* Normally, we grow SIZE exponentially to make the number
941 of calls to read() and realloc() logarithmic in relation
942 to file size. However, read() can read an amount of data
943 smaller than requested, and it would be unreasonable to
944 double SIZE every time *something* was read. Therefore,
945 we double SIZE only when the length exceeds half of the
946 entire allocated size. */
948 fm->content = xrealloc (fm->content, size);
950 nread = read (fd, fm->content + fm->length, size - fm->length);
952 /* Successful read. */
963 if (size > fm->length && fm->length != 0)
964 /* Due to exponential growth of fm->content, the allocated region
965 might be much larger than what is actually needed. */
966 fm->content = xrealloc (fm->content, fm->length);
978 /* Release the resources held by FM. Specifically, this calls
979 munmap() or xfree() on fm->content, depending whether mmap or
980 malloc/read were used to read in the file. It also frees the
981 memory needed to hold the FM structure itself. */
984 read_file_free (struct file_memory *fm)
989 munmap (fm->content, fm->length);
999 /* Free the pointers in a NULL-terminated vector of pointers, then
1000 free the pointer itself. */
1002 free_vec (char **vec)
1013 /* Append vector V2 to vector V1. The function frees V2 and
1014 reallocates V1 (thus you may not use the contents of neither
1015 pointer after the call). If V1 is NULL, V2 is returned. */
1017 merge_vecs (char **v1, char **v2)
1027 /* To avoid j == 0 */
1032 for (i = 0; v1[i]; i++);
1034 for (j = 0; v2[j]; j++);
1035 /* Reallocate v1. */
1036 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1037 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1042 /* A set of simple-minded routines to store strings in a linked list.
1043 This used to also be used for searching, but now we have hash
1046 /* It's a shame that these simple things like linked lists and hash
1047 tables (see hash.c) need to be implemented over and over again. It
1048 would be nice to be able to use the routines from glib -- see
1049 www.gtk.org for details. However, that would make Wget depend on
1050 glib, and I want to avoid dependencies to external libraries for
1051 reasons of convenience and portability (I suspect Wget is more
1052 portable than anything ever written for Gnome). */
1054 /* Append an element to the list. If the list has a huge number of
1055 elements, this can get slow because it has to find the list's
1056 ending. If you think you have to call slist_append in a loop,
1057 think about calling slist_prepend() followed by slist_nreverse(). */
1060 slist_append (slist *l, const char *s)
1062 slist *newel = xnew (slist);
1065 newel->string = xstrdup (s);
1070 /* Find the last element. */
1077 /* Prepend S to the list. Unlike slist_append(), this is O(1). */
1080 slist_prepend (slist *l, const char *s)
1082 slist *newel = xnew (slist);
1083 newel->string = xstrdup (s);
1088 /* Destructively reverse L. */
1091 slist_nreverse (slist *l)
1096 slist *next = l->next;
1104 /* Is there a specific entry in the list? */
1106 slist_contains (slist *l, const char *s)
1108 for (; l; l = l->next)
1109 if (!strcmp (l->string, s))
1114 /* Free the whole slist. */
1116 slist_free (slist *l)
1127 /* Sometimes it's useful to create "sets" of strings, i.e. special
1128 hash tables where you want to store strings as keys and merely
1129 query for their existence. Here is a set of utility routines that
1130 makes that transparent. */
1133 string_set_add (struct hash_table *ht, const char *s)
1135 /* First check whether the set element already exists. If it does,
1136 do nothing so that we don't have to free() the old element and
1137 then strdup() a new one. */
1138 if (hash_table_contains (ht, s))
1141 /* We use "1" as value. It provides us a useful and clear arbitrary
1142 value, and it consumes no memory -- the pointers to the same
1143 string "1" will be shared by all the key-value pairs in all `set'
1145 hash_table_put (ht, xstrdup (s), "1");
1148 /* Synonym for hash_table_contains... */
1151 string_set_contains (struct hash_table *ht, const char *s)
1153 return hash_table_contains (ht, s);
1157 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1164 string_set_free (struct hash_table *ht)
1166 hash_table_map (ht, string_set_free_mapper, NULL);
1167 hash_table_destroy (ht);
1171 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1178 /* Another utility function: call free() on all keys and values of HT. */
1181 free_keys_and_values (struct hash_table *ht)
1183 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1187 /* Engine for legible and legible_large_int; add thousand separators
1188 to numbers printed in strings. */
1191 legible_1 (const char *repr)
1193 static char outbuf[48];
1198 /* Reset the pointers. */
1202 /* Ignore the sign for the purpose of adding thousand
1209 /* How many digits before the first separator? */
1210 mod = strlen (inptr) % 3;
1212 for (i = 0; i < mod; i++)
1213 *outptr++ = inptr[i];
1214 /* Now insert the rest of them, putting separator before every
1216 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1218 if (i % 3 == 0 && i1 != 0)
1220 *outptr++ = inptr[i1];
1222 /* Zero-terminate the string. */
1227 /* Legible -- return a static pointer to the legibly printed wgint. */
1233 /* Print the number into the buffer. */
1234 number_to_string (inbuf, l);
1235 return legible_1 (inbuf);
1238 /* Write a string representation of LARGE_INT NUMBER into the provided
1239 buffer. The buffer should be able to accept 24 characters,
1240 including the terminating zero.
1242 It would be dangerous to use sprintf, because the code wouldn't
1243 work on a machine with gcc-provided long long support, but without
1244 libc support for "%lld". However, such platforms will typically
1245 not have snprintf and will use our version, which does support
1246 "%lld" where long longs are available. */
1249 large_int_to_string (char *buffer, LARGE_INT number)
1251 snprintf (buffer, 24, LARGE_INT_FMT, number);
1254 /* The same as legible(), but works on LARGE_INT. */
1257 legible_large_int (LARGE_INT l)
1260 large_int_to_string (inbuf, l);
1261 return legible_1 (inbuf);
1264 /* Count the digits in an integer number. */
1266 numdigit (wgint number)
1274 while ((number /= 10) > 0)
1279 #define ONE_DIGIT(figure) *p++ = n / (figure) + '0'
1280 #define ONE_DIGIT_ADVANCE(figure) (ONE_DIGIT (figure), n %= (figure))
1282 #define DIGITS_1(figure) ONE_DIGIT (figure)
1283 #define DIGITS_2(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_1 ((figure) / 10)
1284 #define DIGITS_3(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_2 ((figure) / 10)
1285 #define DIGITS_4(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_3 ((figure) / 10)
1286 #define DIGITS_5(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_4 ((figure) / 10)
1287 #define DIGITS_6(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_5 ((figure) / 10)
1288 #define DIGITS_7(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_6 ((figure) / 10)
1289 #define DIGITS_8(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_7 ((figure) / 10)
1290 #define DIGITS_9(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_8 ((figure) / 10)
1291 #define DIGITS_10(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_9 ((figure) / 10)
1293 /* DIGITS_<11-20> are only used on machines with 64-bit numbers. */
1295 #define DIGITS_11(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_10 ((figure) / 10)
1296 #define DIGITS_12(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_11 ((figure) / 10)
1297 #define DIGITS_13(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_12 ((figure) / 10)
1298 #define DIGITS_14(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_13 ((figure) / 10)
1299 #define DIGITS_15(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_14 ((figure) / 10)
1300 #define DIGITS_16(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_15 ((figure) / 10)
1301 #define DIGITS_17(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_16 ((figure) / 10)
1302 #define DIGITS_18(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_17 ((figure) / 10)
1303 #define DIGITS_19(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_18 ((figure) / 10)
1305 /* It is annoying that we have three different syntaxes for 64-bit constants:
1306 - nnnL for 64-bit systems, where they are of type long;
1307 - nnnLL for 32-bit systems that support long long;
1308 - nnnI64 for MS compiler on Windows, which doesn't support long long. */
1311 /* If long is large enough, use long constants. */
1312 # define C10000000000 10000000000L
1313 # define C100000000000 100000000000L
1314 # define C1000000000000 1000000000000L
1315 # define C10000000000000 10000000000000L
1316 # define C100000000000000 100000000000000L
1317 # define C1000000000000000 1000000000000000L
1318 # define C10000000000000000 10000000000000000L
1319 # define C100000000000000000 100000000000000000L
1320 # define C1000000000000000000 1000000000000000000L
1322 # if SIZEOF_LONG_LONG != 0
1323 /* Otherwise, if long long is available, use long long constants. */
1324 # define C10000000000 10000000000LL
1325 # define C100000000000 100000000000LL
1326 # define C1000000000000 1000000000000LL
1327 # define C10000000000000 10000000000000LL
1328 # define C100000000000000 100000000000000LL
1329 # define C1000000000000000 1000000000000000LL
1330 # define C10000000000000000 10000000000000000LL
1331 # define C100000000000000000 100000000000000000LL
1332 # define C1000000000000000000 1000000000000000000LL
1334 # if defined(WINDOWS)
1335 /* Use __int64 constants under Windows. */
1336 # define C10000000000 10000000000I64
1337 # define C100000000000 100000000000I64
1338 # define C1000000000000 1000000000000I64
1339 # define C10000000000000 10000000000000I64
1340 # define C100000000000000 100000000000000I64
1341 # define C1000000000000000 1000000000000000I64
1342 # define C10000000000000000 10000000000000000I64
1343 # define C100000000000000000 100000000000000000I64
1344 # define C1000000000000000000 1000000000000000000I64
1349 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1350 cases and to portably support strange sizes of wgint. */
1351 #if SIZEOF_LONG >= SIZEOF_WGINT
1352 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%ld", (long) (n))
1354 # if SIZEOF_LONG_LONG >= SIZEOF_WGINT
1355 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%lld", (long long) (n))
1358 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%I64", (__int64) (n))
1363 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1364 `sprintf(buffer, "%lld", (long long) number)', only much faster and
1365 portable to machines without long long.
1367 The speedup may make a difference in programs that frequently
1368 convert numbers to strings. Some implementations of sprintf,
1369 particularly the one in GNU libc, have been known to be extremely
1370 slow when converting integers to strings.
1372 Return the pointer to the location where the terminating zero was
1373 printed. (Equivalent to calling buffer+strlen(buffer) after the
1376 BUFFER should be big enough to accept as many bytes as you expect
1377 the number to take up. On machines with 64-bit longs the maximum
1378 needed size is 24 bytes. That includes the digits needed for the
1379 largest 64-bit number, the `-' sign in case it's negative, and the
1380 terminating '\0'. */
1383 number_to_string (char *buffer, wgint number)
1388 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1389 /* We are running in a strange or misconfigured environment. Let
1390 sprintf cope with it. */
1391 SPRINTF_WGINT (buffer, n);
1392 p += strlen (buffer);
1393 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1399 /* We cannot print a '-' and assign -n to n because -n would
1400 overflow. Let sprintf deal with this border case. */
1401 SPRINTF_WGINT (buffer, n);
1402 p += strlen (buffer);
1410 if (n < 10) { DIGITS_1 (1); }
1411 else if (n < 100) { DIGITS_2 (10); }
1412 else if (n < 1000) { DIGITS_3 (100); }
1413 else if (n < 10000) { DIGITS_4 (1000); }
1414 else if (n < 100000) { DIGITS_5 (10000); }
1415 else if (n < 1000000) { DIGITS_6 (100000); }
1416 else if (n < 10000000) { DIGITS_7 (1000000); }
1417 else if (n < 100000000) { DIGITS_8 (10000000); }
1418 else if (n < 1000000000) { DIGITS_9 (100000000); }
1419 #if SIZEOF_WGINT == 4
1420 /* wgint is four bytes long: we're done. */
1421 /* ``if (1)'' serves only to preserve editor indentation. */
1422 else if (1) { DIGITS_10 (1000000000); }
1424 /* wgint is 64 bits long -- make sure to process all the digits. */
1425 else if (n < C10000000000) { DIGITS_10 (1000000000); }
1426 else if (n < C100000000000) { DIGITS_11 (C10000000000); }
1427 else if (n < C1000000000000) { DIGITS_12 (C100000000000); }
1428 else if (n < C10000000000000) { DIGITS_13 (C1000000000000); }
1429 else if (n < C100000000000000) { DIGITS_14 (C10000000000000); }
1430 else if (n < C1000000000000000) { DIGITS_15 (C100000000000000); }
1431 else if (n < C10000000000000000) { DIGITS_16 (C1000000000000000); }
1432 else if (n < C100000000000000000) { DIGITS_17 (C10000000000000000); }
1433 else if (n < C1000000000000000000) { DIGITS_18 (C100000000000000000); }
1434 else { DIGITS_19 (C1000000000000000000); }
1438 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1444 #undef ONE_DIGIT_ADVANCE
1468 /* Print NUMBER to a statically allocated string and return a pointer
1469 to the printed representation.
1471 This function is intended to be used in conjunction with printf.
1472 It is hard to portably print wgint values:
1473 a) you cannot use printf("%ld", number) because wgint can be long
1474 long on 32-bit machines with LFS.
1475 b) you cannot use printf("%lld", number) because NUMBER could be
1476 long on 32-bit machines without LFS, or on 64-bit machines,
1477 which do not require LFS. Also, Windows doesn't support %lld.
1478 c) you cannot use printf("%j", (int_max_t) number) because not all
1479 versions of printf support "%j", the most notable being the one
1481 d) you cannot #define WGINT_FMT to the appropriate format and use
1482 printf(WGINT_FMT, number) because that would break translations
1483 for user-visible messages, such as printf("Downloaded: %d
1486 What you should use instead is printf("%s", number_to_static_string
1489 CAVEAT: since the function returns pointers to static data, you
1490 must be careful to copy its result before calling it again.
1491 However, to make it more useful with printf, the function maintains
1492 an internal ring of static buffers to return. That way things like
1493 printf("%s %s", number_to_static_string (num1),
1494 number_to_static_string (num2)) work as expected. Three buffers
1495 are currently used, which means that "%s %s %s" will work, but "%s
1496 %s %s %s" won't. If you need to print more than three wgints,
1497 bump the RING_SIZE (or rethink your message.) */
1500 number_to_static_string (wgint number)
1502 static char ring[RING_SIZE][24];
1504 char *buf = ring[ringpos];
1505 number_to_string (buf, number);
1506 ringpos = (ringpos + 1) % RING_SIZE;
1510 /* Support for timers. */
1512 #undef TIMER_WINDOWS
1513 #undef TIMER_GETTIMEOFDAY
1516 /* Depending on the OS and availability of gettimeofday(), one and
1517 only one of the above constants will be defined. Virtually all
1518 modern Unix systems will define TIMER_GETTIMEOFDAY; Windows will
1519 use TIMER_WINDOWS. TIMER_TIME is a catch-all method for
1520 non-Windows systems without gettimeofday.
1522 #### Perhaps we should also support ftime(), which exists on old
1523 BSD 4.2-influenced systems? (It also existed under MS DOS Borland
1524 C, if memory serves me.) */
1527 # define TIMER_WINDOWS
1528 #else /* not WINDOWS */
1529 # ifdef HAVE_GETTIMEOFDAY
1530 # define TIMER_GETTIMEOFDAY
1534 #endif /* not WINDOWS */
1536 #ifdef TIMER_GETTIMEOFDAY
1537 typedef struct timeval wget_sys_time;
1541 typedef time_t wget_sys_time;
1544 #ifdef TIMER_WINDOWS
1545 typedef ULARGE_INTEGER wget_sys_time;
1549 /* Whether the start time has been initialized. */
1552 /* The starting point in time which, subtracted from the current
1553 time, yields elapsed time. */
1554 wget_sys_time start;
1556 /* The most recent elapsed time, calculated by wtimer_elapsed().
1557 Measured in milliseconds. */
1558 double elapsed_last;
1560 /* Approximately, the time elapsed between the true start of the
1561 measurement and the time represented by START. */
1562 double elapsed_pre_start;
1565 /* Allocate a timer. Calling wtimer_read on the timer will return
1566 zero. It is not legal to call wtimer_update with a freshly
1567 allocated timer -- use wtimer_reset first. */
1570 wtimer_allocate (void)
1572 struct wget_timer *wt = xnew (struct wget_timer);
1577 /* Allocate a new timer and reset it. Return the new timer. */
1582 struct wget_timer *wt = wtimer_allocate ();
1587 /* Free the resources associated with the timer. Its further use is
1591 wtimer_delete (struct wget_timer *wt)
1596 /* Store system time to WST. */
1599 wtimer_sys_set (wget_sys_time *wst)
1601 #ifdef TIMER_GETTIMEOFDAY
1602 gettimeofday (wst, NULL);
1609 #ifdef TIMER_WINDOWS
1610 /* We use GetSystemTime to get the elapsed time. MSDN warns that
1611 system clock adjustments can skew the output of GetSystemTime
1612 when used as a timer and gives preference to GetTickCount and
1613 high-resolution timers. But GetTickCount can overflow, and hires
1614 timers are typically used for profiling, not for regular time
1615 measurement. Since we handle clock skew anyway, we just use
1619 GetSystemTime (&st);
1621 /* As recommended by MSDN, we convert SYSTEMTIME to FILETIME, copy
1622 FILETIME to ULARGE_INTEGER, and use regular 64-bit integer
1623 arithmetic on that. */
1624 SystemTimeToFileTime (&st, &ft);
1625 wst->HighPart = ft.dwHighDateTime;
1626 wst->LowPart = ft.dwLowDateTime;
1630 /* Reset timer WT. This establishes the starting point from which
1631 wtimer_elapsed() will return the number of elapsed milliseconds.
1632 It is allowed to reset a previously used timer.
1634 If a non-zero value is used as START, the timer's values will be
1638 wtimer_reset (struct wget_timer *wt)
1640 /* Set the start time to the current time. */
1641 wtimer_sys_set (&wt->start);
1642 wt->elapsed_last = 0;
1643 wt->elapsed_pre_start = 0;
1644 wt->initialized = 1;
1648 wtimer_sys_diff (wget_sys_time *wst1, wget_sys_time *wst2)
1650 #ifdef TIMER_GETTIMEOFDAY
1651 return ((double)(wst1->tv_sec - wst2->tv_sec) * 1000
1652 + (double)(wst1->tv_usec - wst2->tv_usec) / 1000);
1656 return 1000 * (*wst1 - *wst2);
1660 /* VC++ 6 doesn't support direct cast of uint64 to double. To work
1661 around this, we subtract, then convert to signed, then finally to
1663 return (double)(signed __int64)(wst1->QuadPart - wst2->QuadPart) / 10000;
1667 /* Update the timer's elapsed interval. This function causes the
1668 timer to call gettimeofday (or time(), etc.) to update its idea of
1669 current time. To get the elapsed interval in milliseconds, use
1672 This function handles clock skew, i.e. time that moves backwards is
1676 wtimer_update (struct wget_timer *wt)
1681 assert (wt->initialized != 0);
1683 wtimer_sys_set (&now);
1684 elapsed = wt->elapsed_pre_start + wtimer_sys_diff (&now, &wt->start);
1686 /* Ideally we'd just return the difference between NOW and
1687 wt->start. However, the system timer can be set back, and we
1688 could return a value smaller than when we were last called, even
1689 a negative value. Both of these would confuse the callers, which
1690 expect us to return monotonically nondecreasing values.
1692 Therefore: if ELAPSED is smaller than its previous known value,
1693 we reset wt->start to the current time and effectively start
1694 measuring from this point. But since we don't want the elapsed
1695 value to start from zero, we set elapsed_pre_start to the last
1696 elapsed time and increment all future calculations by that
1699 if (elapsed < wt->elapsed_last)
1702 wt->elapsed_pre_start = wt->elapsed_last;
1703 elapsed = wt->elapsed_last;
1706 wt->elapsed_last = elapsed;
1709 /* Return the elapsed time in milliseconds between the last call to
1710 wtimer_reset and the last call to wtimer_update.
1712 A typical use of the timer interface would be:
1714 struct wtimer *timer = wtimer_new ();
1715 ... do something that takes a while ...
1717 double msecs = wtimer_read (); */
1720 wtimer_read (const struct wget_timer *wt)
1722 return wt->elapsed_last;
1725 /* Return the assessed granularity of the timer implementation, in
1726 milliseconds. This is used by code that tries to substitute a
1727 better value for timers that have returned zero. */
1730 wtimer_granularity (void)
1732 #ifdef TIMER_GETTIMEOFDAY
1733 /* Granularity of gettimeofday varies wildly between architectures.
1734 However, it appears that on modern machines it tends to be better
1735 than 1ms. Assume 100 usecs. (Perhaps the configure process
1736 could actually measure this?) */
1744 #ifdef TIMER_WINDOWS
1745 /* According to MSDN, GetSystemTime returns a broken-down time
1746 structure the smallest member of which are milliseconds. */
1751 /* This should probably be at a better place, but it doesn't really
1752 fit into html-parse.c. */
1754 /* The function returns the pointer to the malloc-ed quoted version of
1755 string s. It will recognize and quote numeric and special graphic
1756 entities, as per RFC1866:
1764 No other entities are recognized or replaced. */
1766 html_quote_string (const char *s)
1772 /* Pass through the string, and count the new size. */
1773 for (i = 0; *s; s++, i++)
1776 i += 4; /* `amp;' */
1777 else if (*s == '<' || *s == '>')
1778 i += 3; /* `lt;' and `gt;' */
1779 else if (*s == '\"')
1780 i += 5; /* `quot;' */
1784 res = (char *)xmalloc (i + 1);
1786 for (p = res; *s; s++)
1799 *p++ = (*s == '<' ? 'l' : 'g');
1826 /* Determine the width of the terminal we're running on. If that's
1827 not possible, return 0. */
1830 determine_screen_width (void)
1832 /* If there's a way to get the terminal size using POSIX
1833 tcgetattr(), somebody please tell me. */
1838 if (opt.lfilename != NULL)
1841 fd = fileno (stderr);
1842 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1843 return 0; /* most likely ENOTTY */
1846 #else /* not TIOCGWINSZ */
1848 CONSOLE_SCREEN_BUFFER_INFO csbi;
1849 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1851 return csbi.dwSize.X;
1852 # else /* neither WINDOWS nor TIOCGWINSZ */
1854 #endif /* neither WINDOWS nor TIOCGWINSZ */
1855 #endif /* not TIOCGWINSZ */
1858 /* Return a random number between 0 and MAX-1, inclusive.
1860 If MAX is greater than the value of RAND_MAX+1 on the system, the
1861 returned value will be in the range [0, RAND_MAX]. This may be
1862 fixed in a future release.
1864 The random number generator is seeded automatically the first time
1867 This uses rand() for portability. It has been suggested that
1868 random() offers better randomness, but this is not required for
1869 Wget, so I chose to go for simplicity and use rand
1872 DO NOT use this for cryptographic purposes. It is only meant to be
1873 used in situations where quality of the random numbers returned
1874 doesn't really matter. */
1877 random_number (int max)
1885 srand (time (NULL));
1890 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1891 and enforce that assumption by masking other bits. */
1893 # define RAND_MAX 32767
1897 /* This is equivalent to rand() % max, but uses the high-order bits
1898 for better randomness on architecture where rand() is implemented
1899 using a simple congruential generator. */
1901 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1902 return (int)bounded;
1905 /* Return a random uniformly distributed floating point number in the
1906 [0, 1) range. The precision of returned numbers is 9 digits.
1908 Modify this to use erand48() where available! */
1913 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1914 sure it's greater than 1000. */
1915 int rnd1 = random_number (1000);
1916 int rnd2 = random_number (1000);
1917 int rnd3 = random_number (1000);
1918 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1922 /* A debugging function for checking whether an MD5 library works. */
1924 #include "gen-md5.h"
1927 debug_test_md5 (char *buf)
1929 unsigned char raw[16];
1930 static char res[33];
1934 ALLOCA_MD5_CONTEXT (ctx);
1937 gen_md5_update ((unsigned char *)buf, strlen (buf), ctx);
1938 gen_md5_finish (ctx, raw);
1945 *p2++ = XNUM_TO_digit (*p1 >> 4);
1946 *p2++ = XNUM_TO_digit (*p1 & 0xf);
1955 /* Implementation of run_with_timeout, a generic timeout-forcing
1956 routine for systems with Unix-like signal handling. */
1958 #ifdef USE_SIGNAL_TIMEOUT
1959 # ifdef HAVE_SIGSETJMP
1960 # define SETJMP(env) sigsetjmp (env, 1)
1962 static sigjmp_buf run_with_timeout_env;
1965 abort_run_with_timeout (int sig)
1967 assert (sig == SIGALRM);
1968 siglongjmp (run_with_timeout_env, -1);
1970 # else /* not HAVE_SIGSETJMP */
1971 # define SETJMP(env) setjmp (env)
1973 static jmp_buf run_with_timeout_env;
1976 abort_run_with_timeout (int sig)
1978 assert (sig == SIGALRM);
1979 /* We don't have siglongjmp to preserve the set of blocked signals;
1980 if we longjumped out of the handler at this point, SIGALRM would
1981 remain blocked. We must unblock it manually. */
1982 int mask = siggetmask ();
1983 mask &= ~sigmask (SIGALRM);
1986 /* Now it's safe to longjump. */
1987 longjmp (run_with_timeout_env, -1);
1989 # endif /* not HAVE_SIGSETJMP */
1991 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1992 setitimer where available, alarm otherwise.
1994 TIMEOUT should be non-zero. If the timeout value is so small that
1995 it would be rounded to zero, it is rounded to the least legal value
1996 instead (1us for setitimer, 1s for alarm). That ensures that
1997 SIGALRM will be delivered in all cases. */
2000 alarm_set (double timeout)
2003 /* Use the modern itimer interface. */
2004 struct itimerval itv;
2006 itv.it_value.tv_sec = (long) timeout;
2007 itv.it_value.tv_usec = 1000000L * (timeout - (long)timeout);
2008 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
2009 /* Ensure that we wait for at least the minimum interval.
2010 Specifying zero would mean "wait forever". */
2011 itv.it_value.tv_usec = 1;
2012 setitimer (ITIMER_REAL, &itv, NULL);
2013 #else /* not ITIMER_REAL */
2014 /* Use the old alarm() interface. */
2015 int secs = (int) timeout;
2017 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
2018 because alarm(0) means "never deliver the alarm", i.e. "wait
2019 forever", which is not what someone who specifies a 0.5s
2020 timeout would expect. */
2023 #endif /* not ITIMER_REAL */
2026 /* Cancel the alarm set with alarm_set. */
2032 struct itimerval disable;
2034 setitimer (ITIMER_REAL, &disable, NULL);
2035 #else /* not ITIMER_REAL */
2037 #endif /* not ITIMER_REAL */
2040 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
2041 seconds. Returns non-zero if the function was interrupted with a
2042 timeout, zero otherwise.
2044 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
2045 using setitimer() or alarm(). The timeout is enforced by
2046 longjumping out of the SIGALRM handler. This has several
2047 advantages compared to the traditional approach of relying on
2048 signals causing system calls to exit with EINTR:
2050 * The callback function is *forcibly* interrupted after the
2051 timeout expires, (almost) regardless of what it was doing and
2052 whether it was in a syscall. For example, a calculation that
2053 takes a long time is interrupted as reliably as an IO
2056 * It works with both SYSV and BSD signals because it doesn't
2057 depend on the default setting of SA_RESTART.
2059 * It doesn't special handler setup beyond a simple call to
2060 signal(). (It does use sigsetjmp/siglongjmp, but they're
2063 The only downside is that, if FUN allocates internal resources that
2064 are normally freed prior to exit from the functions, they will be
2065 lost in case of timeout. */
2068 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2078 signal (SIGALRM, abort_run_with_timeout);
2079 if (SETJMP (run_with_timeout_env) != 0)
2081 /* Longjumped out of FUN with a timeout. */
2082 signal (SIGALRM, SIG_DFL);
2085 alarm_set (timeout);
2088 /* Preserve errno in case alarm() or signal() modifies it. */
2089 saved_errno = errno;
2091 signal (SIGALRM, SIG_DFL);
2092 errno = saved_errno;
2097 #else /* not USE_SIGNAL_TIMEOUT */
2100 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2101 define it under Windows, because Windows has its own version of
2102 run_with_timeout that uses threads. */
2105 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2110 #endif /* not WINDOWS */
2111 #endif /* not USE_SIGNAL_TIMEOUT */
2115 /* Sleep the specified amount of seconds. On machines without
2116 nanosleep(), this may sleep shorter if interrupted by signals. */
2119 xsleep (double seconds)
2121 #ifdef HAVE_NANOSLEEP
2122 /* nanosleep is the preferred interface because it offers high
2123 accuracy and, more importantly, because it allows us to reliably
2124 restart after having been interrupted by a signal such as
2126 struct timespec sleep, remaining;
2127 sleep.tv_sec = (long) seconds;
2128 sleep.tv_nsec = 1000000000L * (seconds - (long) seconds);
2129 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2130 /* If nanosleep has been interrupted by a signal, adjust the
2131 sleeping period and return to sleep. */
2133 #else /* not HAVE_NANOSLEEP */
2135 /* If usleep is available, use it in preference to select. */
2138 /* On some systems, usleep cannot handle values larger than
2139 1,000,000. If the period is larger than that, use sleep
2140 first, then add usleep for subsecond accuracy. */
2142 seconds -= (long) seconds;
2144 usleep (seconds * 1000000L);
2145 #else /* not HAVE_USLEEP */
2147 struct timeval sleep;
2148 sleep.tv_sec = (long) seconds;
2149 sleep.tv_usec = 1000000L * (seconds - (long) seconds);
2150 select (0, NULL, NULL, NULL, &sleep);
2151 /* If select returns -1 and errno is EINTR, it means we were
2152 interrupted by a signal. But without knowing how long we've
2153 actually slept, we can't return to sleep. Using gettimeofday to
2154 track sleeps is slow and unreliable due to clock skew. */
2155 #else /* not HAVE_SELECT */
2157 #endif /* not HAVE_SELECT */
2158 #endif /* not HAVE_USLEEP */
2159 #endif /* not HAVE_NANOSLEEP */
2162 #endif /* not WINDOWS */