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 /* Concatenate the NULL-terminated list of string arguments into
231 freshly allocated space. */
234 concat_strings (const char *str0, ...)
237 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
240 const char *next_str;
241 int total_length = 0;
244 /* Calculate the length of and allocate the resulting string. */
247 VA_START (args, str0);
248 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
250 int len = strlen (next_str);
251 if (argcount < countof (saved_lengths))
252 saved_lengths[argcount++] = len;
256 p = ret = xmalloc (total_length + 1);
258 /* Copy the strings into the allocated space. */
261 VA_START (args, str0);
262 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
265 if (argcount < countof (saved_lengths))
266 len = saved_lengths[argcount++];
268 len = strlen (next_str);
269 memcpy (p, next_str, len);
278 /* Return pointer to a static char[] buffer in which zero-terminated
279 string-representation of TM (in form hh:mm:ss) is printed.
281 If TM is NULL, the current time will be used. */
284 time_str (time_t *tm)
286 static char output[15];
288 time_t secs = tm ? *tm : time (NULL);
292 /* In case of error, return the empty string. Maybe we should
293 just abort if this happens? */
297 ptm = localtime (&secs);
298 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
302 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
305 datetime_str (time_t *tm)
307 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
309 time_t secs = tm ? *tm : time (NULL);
313 /* In case of error, return the empty string. Maybe we should
314 just abort if this happens? */
318 ptm = localtime (&secs);
319 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
320 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
321 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
325 /* The Windows versions of the following two functions are defined in
330 fork_to_background (void)
333 /* Whether we arrange our own version of opt.lfilename here. */
334 int logfile_changed = 0;
338 /* We must create the file immediately to avoid either a race
339 condition (which arises from using unique_name and failing to
340 use fopen_excl) or lying to the user about the log file name
341 (which arises from using unique_name, printing the name, and
342 using fopen_excl later on.) */
343 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, 0, &opt.lfilename);
359 /* parent, no error */
360 printf (_("Continuing in background, pid %d.\n"), (int)pid);
362 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
363 exit (0); /* #### should we use _exit()? */
366 /* child: give up the privileges and keep running. */
368 freopen ("/dev/null", "r", stdin);
369 freopen ("/dev/null", "w", stdout);
370 freopen ("/dev/null", "w", stderr);
372 #endif /* not WINDOWS */
374 /* "Touch" FILE, i.e. make its atime and mtime equal to the time
375 specified with TM. */
377 touch (const char *file, time_t tm)
379 #ifdef HAVE_STRUCT_UTIMBUF
380 struct utimbuf times;
381 times.actime = times.modtime = tm;
384 times[0] = times[1] = tm;
387 if (utime (file, ×) == -1)
388 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
391 /* Checks if FILE is a symbolic link, and removes it if it is. Does
392 nothing under MS-Windows. */
394 remove_link (const char *file)
399 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
401 DEBUGP (("Unlinking %s (symlink).\n", file));
404 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
405 file, strerror (errno));
410 /* Does FILENAME exist? This is quite a lousy implementation, since
411 it supplies no error codes -- only a yes-or-no answer. Thus it
412 will return that a file does not exist if, e.g., the directory is
413 unreadable. I don't mind it too much currently, though. The
414 proper way should, of course, be to have a third, error state,
415 other than true/false, but that would introduce uncalled-for
416 additional complexity to the callers. */
418 file_exists_p (const char *filename)
421 return access (filename, F_OK) >= 0;
424 return stat (filename, &buf) >= 0;
428 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
429 Returns 0 on error. */
431 file_non_directory_p (const char *path)
434 /* Use lstat() rather than stat() so that symbolic links pointing to
435 directories can be identified correctly. */
436 if (lstat (path, &buf) != 0)
438 return S_ISDIR (buf.st_mode) ? 0 : 1;
441 /* Return the size of file named by FILENAME, or -1 if it cannot be
442 opened or seeked into. */
444 file_size (const char *filename)
446 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
448 /* We use fseek rather than stat to determine the file size because
449 that way we can also verify that the file is readable without
450 explicitly checking for permissions. Inspired by the POST patch
452 FILE *fp = fopen (filename, "rb");
455 fseeko (fp, 0, SEEK_END);
461 if (stat (filename, &st) < 0)
467 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
468 doesn't exist is found. Return a freshly allocated copy of the
472 unique_name_1 (const char *prefix)
475 int plen = strlen (prefix);
476 char *template = (char *)alloca (plen + 1 + 24);
477 char *template_tail = template + plen;
479 memcpy (template, prefix, plen);
480 *template_tail++ = '.';
483 number_to_string (template_tail, count++);
484 while (file_exists_p (template));
486 return xstrdup (template);
489 /* Return a unique file name, based on FILE.
491 More precisely, if FILE doesn't exist, it is returned unmodified.
492 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
493 file name that doesn't exist is returned.
495 The resulting file is not created, only verified that it didn't
496 exist at the point in time when the function was called.
497 Therefore, where security matters, don't rely that the file created
498 by this function exists until you open it with O_EXCL or
501 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
502 string. Otherwise, it may return FILE if the file doesn't exist
503 (and therefore doesn't need changing). */
506 unique_name (const char *file, int allow_passthrough)
508 /* If the FILE itself doesn't exist, return it without
510 if (!file_exists_p (file))
511 return allow_passthrough ? (char *)file : xstrdup (file);
513 /* Otherwise, find a numeric suffix that results in unused file name
515 return unique_name_1 (file);
518 /* Create a file based on NAME, except without overwriting an existing
519 file with that name. Providing O_EXCL is correctly implemented,
520 this function does not have the race condition associated with
521 opening the file returned by unique_name. */
524 unique_create (const char *name, int binary, char **opened_name)
526 /* unique file name, based on NAME */
527 char *uname = unique_name (name, 0);
529 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
532 uname = unique_name (name, 0);
534 if (opened_name && fp != NULL)
537 *opened_name = uname;
549 /* Open the file for writing, with the addition that the file is
550 opened "exclusively". This means that, if the file already exists,
551 this function will *fail* and errno will be set to EEXIST. If
552 BINARY is set, the file will be opened in binary mode, equivalent
555 If opening the file fails for any reason, including the file having
556 previously existed, this function returns NULL and sets errno
560 fopen_excl (const char *fname, int binary)
564 int flags = O_WRONLY | O_CREAT | O_EXCL;
569 fd = open (fname, flags, 0666);
572 return fdopen (fd, binary ? "wb" : "w");
573 #else /* not O_EXCL */
574 return fopen (fname, binary ? "wb" : "w");
575 #endif /* not O_EXCL */
578 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
579 are missing, create them first. In case any mkdir() call fails,
580 return its error status. Returns 0 on successful completion.
582 The behaviour of this function should be identical to the behaviour
583 of `mkdir -p' on systems where mkdir supports the `-p' option. */
585 make_directory (const char *directory)
587 int i, ret, quit = 0;
590 /* Make a copy of dir, to be able to write to it. Otherwise, the
591 function is unsafe if called with a read-only char *argument. */
592 STRDUP_ALLOCA (dir, directory);
594 /* If the first character of dir is '/', skip it (and thus enable
595 creation of absolute-pathname directories. */
596 for (i = (*dir == '/'); 1; ++i)
598 for (; dir[i] && dir[i] != '/'; i++)
603 /* Check whether the directory already exists. Allow creation of
604 of intermediate directories to fail, as the initial path components
605 are not necessarily directories! */
606 if (!file_exists_p (dir))
607 ret = mkdir (dir, 0777);
618 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
619 should be a file name.
621 file_merge("/foo/bar", "baz") => "/foo/baz"
622 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
623 file_merge("foo", "bar") => "bar"
625 In other words, it's a simpler and gentler version of uri_merge_1. */
628 file_merge (const char *base, const char *file)
631 const char *cut = (const char *)strrchr (base, '/');
634 return xstrdup (file);
636 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
637 memcpy (result, base, cut - base);
638 result[cut - base] = '/';
639 strcpy (result + (cut - base) + 1, file);
644 static int in_acclist PARAMS ((const char *const *, const char *, int));
646 /* Determine whether a file is acceptable to be followed, according to
647 lists of patterns to accept/reject. */
649 acceptable (const char *s)
653 while (l && s[l] != '/')
660 return (in_acclist ((const char *const *)opt.accepts, s, 1)
661 && !in_acclist ((const char *const *)opt.rejects, s, 1));
663 return in_acclist ((const char *const *)opt.accepts, s, 1);
665 else if (opt.rejects)
666 return !in_acclist ((const char *const *)opt.rejects, s, 1);
670 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
671 `/something', frontcmp() will return 1 only if S2 begins with
672 `/something'. Otherwise, 0 is returned. */
674 frontcmp (const char *s1, const char *s2)
676 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
680 /* Iterate through STRLIST, and return the first element that matches
681 S, through wildcards or front comparison (as appropriate). */
683 proclist (char **strlist, const char *s, enum accd flags)
687 for (x = strlist; *x; x++)
688 if (has_wildcards_p (*x))
690 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
695 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
702 /* Returns whether DIRECTORY is acceptable for download, wrt the
703 include/exclude lists.
705 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
706 and absolute paths may be freely intermixed. */
708 accdir (const char *directory, enum accd flags)
710 /* Remove starting '/'. */
711 if (flags & ALLABS && *directory == '/')
715 if (!proclist (opt.includes, directory, flags))
720 if (proclist (opt.excludes, directory, flags))
726 /* Return non-zero if STRING ends with TAIL. For instance:
728 match_tail ("abc", "bc", 0) -> 1
729 match_tail ("abc", "ab", 0) -> 0
730 match_tail ("abc", "abc", 0) -> 1
732 If FOLD_CASE_P is non-zero, the comparison will be
736 match_tail (const char *string, const char *tail, int fold_case_p)
740 /* We want this to be fast, so we code two loops, one with
741 case-folding, one without. */
745 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
746 if (string[i] != tail[j])
751 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
752 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
756 /* If the tail was exhausted, the match was succesful. */
763 /* Checks whether string S matches each element of ACCEPTS. A list
764 element are matched either with fnmatch() or match_tail(),
765 according to whether the element contains wildcards or not.
767 If the BACKWARD is 0, don't do backward comparison -- just compare
770 in_acclist (const char *const *accepts, const char *s, int backward)
772 for (; *accepts; accepts++)
774 if (has_wildcards_p (*accepts))
776 /* fnmatch returns 0 if the pattern *does* match the
778 if (fnmatch (*accepts, s, 0) == 0)
785 if (match_tail (s, *accepts, 0))
790 if (!strcmp (s, *accepts))
798 /* Return the location of STR's suffix (file extension). Examples:
799 suffix ("foo.bar") -> "bar"
800 suffix ("foo.bar.baz") -> "baz"
801 suffix ("/foo/bar") -> NULL
802 suffix ("/foo.bar/baz") -> NULL */
804 suffix (const char *str)
808 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
812 return (char *)str + i;
817 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
821 has_wildcards_p (const char *s)
824 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
829 /* Return non-zero if FNAME ends with a typical HTML suffix. The
830 following (case-insensitive) suffixes are presumed to be HTML files:
834 ?html (`?' matches one character)
836 #### CAVEAT. This is not necessarily a good indication that FNAME
837 refers to a file that contains HTML! */
839 has_html_suffix_p (const char *fname)
843 if ((suf = suffix (fname)) == NULL)
845 if (!strcasecmp (suf, "html"))
847 if (!strcasecmp (suf, "htm"))
849 if (suf[0] && !strcasecmp (suf + 1, "html"))
854 /* Read a line from FP and return the pointer to freshly allocated
855 storage. The storage space is obtained through malloc() and should
856 be freed with free() when it is no longer needed.
858 The length of the line is not limited, except by available memory.
859 The newline character at the end of line is retained. The line is
860 terminated with a zero character.
862 After end-of-file is encountered without anything being read, NULL
863 is returned. NULL is also returned on error. To distinguish
864 between these two cases, use the stdio function ferror(). */
867 read_whole_line (FILE *fp)
871 char *line = (char *)xmalloc (bufsize);
873 while (fgets (line + length, bufsize - length, fp))
875 length += strlen (line + length);
877 /* Possible for example when reading from a binary file where
878 a line begins with \0. */
881 if (line[length - 1] == '\n')
884 /* fgets() guarantees to read the whole line, or to use up the
885 space we've given it. We can double the buffer
888 line = xrealloc (line, bufsize);
890 if (length == 0 || ferror (fp))
895 if (length + 1 < bufsize)
896 /* Relieve the memory from our exponential greediness. We say
897 `length + 1' because the terminating \0 is not included in
898 LENGTH. We don't need to zero-terminate the string ourselves,
899 though, because fgets() does that. */
900 line = xrealloc (line, length + 1);
904 /* Read FILE into memory. A pointer to `struct file_memory' are
905 returned; use struct element `content' to access file contents, and
906 the element `length' to know the file length. `content' is *not*
907 zero-terminated, and you should *not* read or write beyond the [0,
908 length) range of characters.
910 After you are done with the file contents, call read_file_free to
913 Depending on the operating system and the type of file that is
914 being read, read_file() either mmap's the file into memory, or
915 reads the file into the core using read().
917 If file is named "-", fileno(stdin) is used for reading instead.
918 If you want to read from a real file named "-", use "./-" instead. */
921 read_file (const char *file)
924 struct file_memory *fm;
926 int inhibit_close = 0;
928 /* Some magic in the finest tradition of Perl and its kin: if FILE
929 is "-", just use stdin. */
934 /* Note that we don't inhibit mmap() in this case. If stdin is
935 redirected from a regular file, mmap() will still work. */
938 fd = open (file, O_RDONLY);
941 fm = xnew (struct file_memory);
946 if (fstat (fd, &buf) < 0)
948 fm->length = buf.st_size;
949 /* NOTE: As far as I know, the callers of this function never
950 modify the file text. Relying on this would enable us to
951 specify PROT_READ and MAP_SHARED for a marginal gain in
952 efficiency, but at some cost to generality. */
953 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
955 if (fm->content == (char *)MAP_FAILED)
965 /* The most common reason why mmap() fails is that FD does not point
966 to a plain file. However, it's also possible that mmap() doesn't
967 work for a particular type of file. Therefore, whenever mmap()
968 fails, we just fall back to the regular method. */
969 #endif /* HAVE_MMAP */
972 size = 512; /* number of bytes fm->contents can
973 hold at any given time. */
974 fm->content = xmalloc (size);
978 if (fm->length > size / 2)
980 /* #### I'm not sure whether the whole exponential-growth
981 thing makes sense with kernel read. On Linux at least,
982 read() refuses to read more than 4K from a file at a
983 single chunk anyway. But other Unixes might optimize it
984 better, and it doesn't *hurt* anything, so I'm leaving
987 /* Normally, we grow SIZE exponentially to make the number
988 of calls to read() and realloc() logarithmic in relation
989 to file size. However, read() can read an amount of data
990 smaller than requested, and it would be unreasonable to
991 double SIZE every time *something* was read. Therefore,
992 we double SIZE only when the length exceeds half of the
993 entire allocated size. */
995 fm->content = xrealloc (fm->content, size);
997 nread = read (fd, fm->content + fm->length, size - fm->length);
999 /* Successful read. */
1000 fm->length += nread;
1010 if (size > fm->length && fm->length != 0)
1011 /* Due to exponential growth of fm->content, the allocated region
1012 might be much larger than what is actually needed. */
1013 fm->content = xrealloc (fm->content, fm->length);
1020 xfree (fm->content);
1025 /* Release the resources held by FM. Specifically, this calls
1026 munmap() or xfree() on fm->content, depending whether mmap or
1027 malloc/read were used to read in the file. It also frees the
1028 memory needed to hold the FM structure itself. */
1031 read_file_free (struct file_memory *fm)
1036 munmap (fm->content, fm->length);
1041 xfree (fm->content);
1046 /* Free the pointers in a NULL-terminated vector of pointers, then
1047 free the pointer itself. */
1049 free_vec (char **vec)
1060 /* Append vector V2 to vector V1. The function frees V2 and
1061 reallocates V1 (thus you may not use the contents of neither
1062 pointer after the call). If V1 is NULL, V2 is returned. */
1064 merge_vecs (char **v1, char **v2)
1074 /* To avoid j == 0 */
1079 for (i = 0; v1[i]; i++);
1081 for (j = 0; v2[j]; j++);
1082 /* Reallocate v1. */
1083 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1084 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1089 /* A set of simple-minded routines to store strings in a linked list.
1090 This used to also be used for searching, but now we have hash
1093 /* It's a shame that these simple things like linked lists and hash
1094 tables (see hash.c) need to be implemented over and over again. It
1095 would be nice to be able to use the routines from glib -- see
1096 www.gtk.org for details. However, that would make Wget depend on
1097 glib, and I want to avoid dependencies to external libraries for
1098 reasons of convenience and portability (I suspect Wget is more
1099 portable than anything ever written for Gnome). */
1101 /* Append an element to the list. If the list has a huge number of
1102 elements, this can get slow because it has to find the list's
1103 ending. If you think you have to call slist_append in a loop,
1104 think about calling slist_prepend() followed by slist_nreverse(). */
1107 slist_append (slist *l, const char *s)
1109 slist *newel = xnew (slist);
1112 newel->string = xstrdup (s);
1117 /* Find the last element. */
1124 /* Prepend S to the list. Unlike slist_append(), this is O(1). */
1127 slist_prepend (slist *l, const char *s)
1129 slist *newel = xnew (slist);
1130 newel->string = xstrdup (s);
1135 /* Destructively reverse L. */
1138 slist_nreverse (slist *l)
1143 slist *next = l->next;
1151 /* Is there a specific entry in the list? */
1153 slist_contains (slist *l, const char *s)
1155 for (; l; l = l->next)
1156 if (!strcmp (l->string, s))
1161 /* Free the whole slist. */
1163 slist_free (slist *l)
1174 /* Sometimes it's useful to create "sets" of strings, i.e. special
1175 hash tables where you want to store strings as keys and merely
1176 query for their existence. Here is a set of utility routines that
1177 makes that transparent. */
1180 string_set_add (struct hash_table *ht, const char *s)
1182 /* First check whether the set element already exists. If it does,
1183 do nothing so that we don't have to free() the old element and
1184 then strdup() a new one. */
1185 if (hash_table_contains (ht, s))
1188 /* We use "1" as value. It provides us a useful and clear arbitrary
1189 value, and it consumes no memory -- the pointers to the same
1190 string "1" will be shared by all the key-value pairs in all `set'
1192 hash_table_put (ht, xstrdup (s), "1");
1195 /* Synonym for hash_table_contains... */
1198 string_set_contains (struct hash_table *ht, const char *s)
1200 return hash_table_contains (ht, s);
1204 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1211 string_set_free (struct hash_table *ht)
1213 hash_table_map (ht, string_set_free_mapper, NULL);
1214 hash_table_destroy (ht);
1218 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1225 /* Another utility function: call free() on all keys and values of HT. */
1228 free_keys_and_values (struct hash_table *ht)
1230 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1234 /* Engine for legible and legible_large_int; add thousand separators
1235 to numbers printed in strings. */
1238 legible_1 (const char *repr)
1240 static char outbuf[48];
1245 /* Reset the pointers. */
1249 /* Ignore the sign for the purpose of adding thousand
1256 /* How many digits before the first separator? */
1257 mod = strlen (inptr) % 3;
1259 for (i = 0; i < mod; i++)
1260 *outptr++ = inptr[i];
1261 /* Now insert the rest of them, putting separator before every
1263 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1265 if (i % 3 == 0 && i1 != 0)
1267 *outptr++ = inptr[i1];
1269 /* Zero-terminate the string. */
1274 /* Legible -- return a static pointer to the legibly printed wgint. */
1280 /* Print the number into the buffer. */
1281 number_to_string (inbuf, l);
1282 return legible_1 (inbuf);
1285 /* Write a string representation of LARGE_INT NUMBER into the provided
1286 buffer. The buffer should be able to accept 24 characters,
1287 including the terminating zero.
1289 It would be dangerous to use sprintf, because the code wouldn't
1290 work on a machine with gcc-provided long long support, but without
1291 libc support for "%lld". However, such platforms will typically
1292 not have snprintf and will use our version, which does support
1293 "%lld" where long longs are available. */
1296 large_int_to_string (char *buffer, LARGE_INT number)
1298 snprintf (buffer, 24, LARGE_INT_FMT, number);
1301 /* The same as legible(), but works on LARGE_INT. */
1304 legible_large_int (LARGE_INT l)
1307 large_int_to_string (inbuf, l);
1308 return legible_1 (inbuf);
1311 /* Count the digits in an integer number. */
1313 numdigit (wgint number)
1321 while ((number /= 10) > 0)
1326 #define ONE_DIGIT(figure) *p++ = n / (figure) + '0'
1327 #define ONE_DIGIT_ADVANCE(figure) (ONE_DIGIT (figure), n %= (figure))
1329 #define DIGITS_1(figure) ONE_DIGIT (figure)
1330 #define DIGITS_2(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_1 ((figure) / 10)
1331 #define DIGITS_3(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_2 ((figure) / 10)
1332 #define DIGITS_4(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_3 ((figure) / 10)
1333 #define DIGITS_5(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_4 ((figure) / 10)
1334 #define DIGITS_6(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_5 ((figure) / 10)
1335 #define DIGITS_7(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_6 ((figure) / 10)
1336 #define DIGITS_8(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_7 ((figure) / 10)
1337 #define DIGITS_9(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_8 ((figure) / 10)
1338 #define DIGITS_10(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_9 ((figure) / 10)
1340 /* DIGITS_<11-20> are only used on machines with 64-bit numbers. */
1342 #define DIGITS_11(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_10 ((figure) / 10)
1343 #define DIGITS_12(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_11 ((figure) / 10)
1344 #define DIGITS_13(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_12 ((figure) / 10)
1345 #define DIGITS_14(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_13 ((figure) / 10)
1346 #define DIGITS_15(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_14 ((figure) / 10)
1347 #define DIGITS_16(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_15 ((figure) / 10)
1348 #define DIGITS_17(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_16 ((figure) / 10)
1349 #define DIGITS_18(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_17 ((figure) / 10)
1350 #define DIGITS_19(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_18 ((figure) / 10)
1352 /* It is annoying that we have three different syntaxes for 64-bit constants:
1353 - nnnL for 64-bit systems, where they are of type long;
1354 - nnnLL for 32-bit systems that support long long;
1355 - nnnI64 for MS compiler on Windows, which doesn't support long long. */
1358 /* If long is large enough, use long constants. */
1359 # define C10000000000 10000000000L
1360 # define C100000000000 100000000000L
1361 # define C1000000000000 1000000000000L
1362 # define C10000000000000 10000000000000L
1363 # define C100000000000000 100000000000000L
1364 # define C1000000000000000 1000000000000000L
1365 # define C10000000000000000 10000000000000000L
1366 # define C100000000000000000 100000000000000000L
1367 # define C1000000000000000000 1000000000000000000L
1369 # if SIZEOF_LONG_LONG != 0
1370 /* Otherwise, if long long is available, use long long constants. */
1371 # define C10000000000 10000000000LL
1372 # define C100000000000 100000000000LL
1373 # define C1000000000000 1000000000000LL
1374 # define C10000000000000 10000000000000LL
1375 # define C100000000000000 100000000000000LL
1376 # define C1000000000000000 1000000000000000LL
1377 # define C10000000000000000 10000000000000000LL
1378 # define C100000000000000000 100000000000000000LL
1379 # define C1000000000000000000 1000000000000000000LL
1381 # if defined(WINDOWS)
1382 /* Use __int64 constants under Windows. */
1383 # define C10000000000 10000000000I64
1384 # define C100000000000 100000000000I64
1385 # define C1000000000000 1000000000000I64
1386 # define C10000000000000 10000000000000I64
1387 # define C100000000000000 100000000000000I64
1388 # define C1000000000000000 1000000000000000I64
1389 # define C10000000000000000 10000000000000000I64
1390 # define C100000000000000000 100000000000000000I64
1391 # define C1000000000000000000 1000000000000000000I64
1396 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1397 cases and to portably support strange sizes of wgint. */
1398 #if SIZEOF_LONG >= SIZEOF_WGINT
1399 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%ld", (long) (n))
1401 # if SIZEOF_LONG_LONG >= SIZEOF_WGINT
1402 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%lld", (long long) (n))
1405 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%I64", (__int64) (n))
1410 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1411 `sprintf(buffer, "%lld", (long long) number)', only much faster and
1412 portable to machines without long long.
1414 The speedup may make a difference in programs that frequently
1415 convert numbers to strings. Some implementations of sprintf,
1416 particularly the one in GNU libc, have been known to be extremely
1417 slow when converting integers to strings.
1419 Return the pointer to the location where the terminating zero was
1420 printed. (Equivalent to calling buffer+strlen(buffer) after the
1423 BUFFER should be big enough to accept as many bytes as you expect
1424 the number to take up. On machines with 64-bit longs the maximum
1425 needed size is 24 bytes. That includes the digits needed for the
1426 largest 64-bit number, the `-' sign in case it's negative, and the
1427 terminating '\0'. */
1430 number_to_string (char *buffer, wgint number)
1435 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1436 /* We are running in a strange or misconfigured environment. Let
1437 sprintf cope with it. */
1438 SPRINTF_WGINT (buffer, n);
1439 p += strlen (buffer);
1440 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1446 /* We cannot print a '-' and assign -n to n because -n would
1447 overflow. Let sprintf deal with this border case. */
1448 SPRINTF_WGINT (buffer, n);
1449 p += strlen (buffer);
1457 if (n < 10) { DIGITS_1 (1); }
1458 else if (n < 100) { DIGITS_2 (10); }
1459 else if (n < 1000) { DIGITS_3 (100); }
1460 else if (n < 10000) { DIGITS_4 (1000); }
1461 else if (n < 100000) { DIGITS_5 (10000); }
1462 else if (n < 1000000) { DIGITS_6 (100000); }
1463 else if (n < 10000000) { DIGITS_7 (1000000); }
1464 else if (n < 100000000) { DIGITS_8 (10000000); }
1465 else if (n < 1000000000) { DIGITS_9 (100000000); }
1466 #if SIZEOF_WGINT == 4
1467 /* wgint is four bytes long: we're done. */
1468 /* ``if (1)'' serves only to preserve editor indentation. */
1469 else if (1) { DIGITS_10 (1000000000); }
1471 /* wgint is 64 bits long -- make sure to process all the digits. */
1472 else if (n < C10000000000) { DIGITS_10 (1000000000); }
1473 else if (n < C100000000000) { DIGITS_11 (C10000000000); }
1474 else if (n < C1000000000000) { DIGITS_12 (C100000000000); }
1475 else if (n < C10000000000000) { DIGITS_13 (C1000000000000); }
1476 else if (n < C100000000000000) { DIGITS_14 (C10000000000000); }
1477 else if (n < C1000000000000000) { DIGITS_15 (C100000000000000); }
1478 else if (n < C10000000000000000) { DIGITS_16 (C1000000000000000); }
1479 else if (n < C100000000000000000) { DIGITS_17 (C10000000000000000); }
1480 else if (n < C1000000000000000000) { DIGITS_18 (C100000000000000000); }
1481 else { DIGITS_19 (C1000000000000000000); }
1485 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1491 #undef ONE_DIGIT_ADVANCE
1515 /* Print NUMBER to a statically allocated string and return a pointer
1516 to the printed representation.
1518 This function is intended to be used in conjunction with printf.
1519 It is hard to portably print wgint values:
1520 a) you cannot use printf("%ld", number) because wgint can be long
1521 long on 32-bit machines with LFS.
1522 b) you cannot use printf("%lld", number) because NUMBER could be
1523 long on 32-bit machines without LFS, or on 64-bit machines,
1524 which do not require LFS. Also, Windows doesn't support %lld.
1525 c) you cannot use printf("%j", (int_max_t) number) because not all
1526 versions of printf support "%j", the most notable being the one
1528 d) you cannot #define WGINT_FMT to the appropriate format and use
1529 printf(WGINT_FMT, number) because that would break translations
1530 for user-visible messages, such as printf("Downloaded: %d
1533 What you should use instead is printf("%s", number_to_static_string
1536 CAVEAT: since the function returns pointers to static data, you
1537 must be careful to copy its result before calling it again.
1538 However, to make it more useful with printf, the function maintains
1539 an internal ring of static buffers to return. That way things like
1540 printf("%s %s", number_to_static_string (num1),
1541 number_to_static_string (num2)) work as expected. Three buffers
1542 are currently used, which means that "%s %s %s" will work, but "%s
1543 %s %s %s" won't. If you need to print more than three wgints,
1544 bump the RING_SIZE (or rethink your message.) */
1547 number_to_static_string (wgint number)
1549 static char ring[RING_SIZE][24];
1551 char *buf = ring[ringpos];
1552 number_to_string (buf, number);
1553 ringpos = (ringpos + 1) % RING_SIZE;
1557 /* Support for timers. */
1559 #undef TIMER_WINDOWS
1560 #undef TIMER_GETTIMEOFDAY
1563 /* Depending on the OS and availability of gettimeofday(), one and
1564 only one of the above constants will be defined. Virtually all
1565 modern Unix systems will define TIMER_GETTIMEOFDAY; Windows will
1566 use TIMER_WINDOWS. TIMER_TIME is a catch-all method for
1567 non-Windows systems without gettimeofday.
1569 #### Perhaps we should also support ftime(), which exists on old
1570 BSD 4.2-influenced systems? (It also existed under MS DOS Borland
1571 C, if memory serves me.) */
1574 # define TIMER_WINDOWS
1575 #else /* not WINDOWS */
1576 # ifdef HAVE_GETTIMEOFDAY
1577 # define TIMER_GETTIMEOFDAY
1581 #endif /* not WINDOWS */
1583 #ifdef TIMER_GETTIMEOFDAY
1584 typedef struct timeval wget_sys_time;
1588 typedef time_t wget_sys_time;
1591 #ifdef TIMER_WINDOWS
1592 typedef ULARGE_INTEGER wget_sys_time;
1596 /* Whether the start time has been initialized. */
1599 /* The starting point in time which, subtracted from the current
1600 time, yields elapsed time. */
1601 wget_sys_time start;
1603 /* The most recent elapsed time, calculated by wtimer_elapsed().
1604 Measured in milliseconds. */
1605 double elapsed_last;
1607 /* Approximately, the time elapsed between the true start of the
1608 measurement and the time represented by START. */
1609 double elapsed_pre_start;
1612 /* Allocate a timer. Calling wtimer_read on the timer will return
1613 zero. It is not legal to call wtimer_update with a freshly
1614 allocated timer -- use wtimer_reset first. */
1617 wtimer_allocate (void)
1619 struct wget_timer *wt = xnew (struct wget_timer);
1624 /* Allocate a new timer and reset it. Return the new timer. */
1629 struct wget_timer *wt = wtimer_allocate ();
1634 /* Free the resources associated with the timer. Its further use is
1638 wtimer_delete (struct wget_timer *wt)
1643 /* Store system time to WST. */
1646 wtimer_sys_set (wget_sys_time *wst)
1648 #ifdef TIMER_GETTIMEOFDAY
1649 gettimeofday (wst, NULL);
1656 #ifdef TIMER_WINDOWS
1657 /* We use GetSystemTime to get the elapsed time. MSDN warns that
1658 system clock adjustments can skew the output of GetSystemTime
1659 when used as a timer and gives preference to GetTickCount and
1660 high-resolution timers. But GetTickCount can overflow, and hires
1661 timers are typically used for profiling, not for regular time
1662 measurement. Since we handle clock skew anyway, we just use
1666 GetSystemTime (&st);
1668 /* As recommended by MSDN, we convert SYSTEMTIME to FILETIME, copy
1669 FILETIME to ULARGE_INTEGER, and use regular 64-bit integer
1670 arithmetic on that. */
1671 SystemTimeToFileTime (&st, &ft);
1672 wst->HighPart = ft.dwHighDateTime;
1673 wst->LowPart = ft.dwLowDateTime;
1677 /* Reset timer WT. This establishes the starting point from which
1678 wtimer_elapsed() will return the number of elapsed milliseconds.
1679 It is allowed to reset a previously used timer.
1681 If a non-zero value is used as START, the timer's values will be
1685 wtimer_reset (struct wget_timer *wt)
1687 /* Set the start time to the current time. */
1688 wtimer_sys_set (&wt->start);
1689 wt->elapsed_last = 0;
1690 wt->elapsed_pre_start = 0;
1691 wt->initialized = 1;
1695 wtimer_sys_diff (wget_sys_time *wst1, wget_sys_time *wst2)
1697 #ifdef TIMER_GETTIMEOFDAY
1698 return ((double)(wst1->tv_sec - wst2->tv_sec) * 1000
1699 + (double)(wst1->tv_usec - wst2->tv_usec) / 1000);
1703 return 1000 * (*wst1 - *wst2);
1707 /* VC++ 6 doesn't support direct cast of uint64 to double. To work
1708 around this, we subtract, then convert to signed, then finally to
1710 return (double)(signed __int64)(wst1->QuadPart - wst2->QuadPart) / 10000;
1714 /* Update the timer's elapsed interval. This function causes the
1715 timer to call gettimeofday (or time(), etc.) to update its idea of
1716 current time. To get the elapsed interval in milliseconds, use
1719 This function handles clock skew, i.e. time that moves backwards is
1723 wtimer_update (struct wget_timer *wt)
1728 assert (wt->initialized != 0);
1730 wtimer_sys_set (&now);
1731 elapsed = wt->elapsed_pre_start + wtimer_sys_diff (&now, &wt->start);
1733 /* Ideally we'd just return the difference between NOW and
1734 wt->start. However, the system timer can be set back, and we
1735 could return a value smaller than when we were last called, even
1736 a negative value. Both of these would confuse the callers, which
1737 expect us to return monotonically nondecreasing values.
1739 Therefore: if ELAPSED is smaller than its previous known value,
1740 we reset wt->start to the current time and effectively start
1741 measuring from this point. But since we don't want the elapsed
1742 value to start from zero, we set elapsed_pre_start to the last
1743 elapsed time and increment all future calculations by that
1746 if (elapsed < wt->elapsed_last)
1749 wt->elapsed_pre_start = wt->elapsed_last;
1750 elapsed = wt->elapsed_last;
1753 wt->elapsed_last = elapsed;
1756 /* Return the elapsed time in milliseconds between the last call to
1757 wtimer_reset and the last call to wtimer_update.
1759 A typical use of the timer interface would be:
1761 struct wtimer *timer = wtimer_new ();
1762 ... do something that takes a while ...
1764 double msecs = wtimer_read (); */
1767 wtimer_read (const struct wget_timer *wt)
1769 return wt->elapsed_last;
1772 /* Return the assessed granularity of the timer implementation, in
1773 milliseconds. This is used by code that tries to substitute a
1774 better value for timers that have returned zero. */
1777 wtimer_granularity (void)
1779 #ifdef TIMER_GETTIMEOFDAY
1780 /* Granularity of gettimeofday varies wildly between architectures.
1781 However, it appears that on modern machines it tends to be better
1782 than 1ms. Assume 100 usecs. (Perhaps the configure process
1783 could actually measure this?) */
1791 #ifdef TIMER_WINDOWS
1792 /* According to MSDN, GetSystemTime returns a broken-down time
1793 structure the smallest member of which are milliseconds. */
1798 /* This should probably be at a better place, but it doesn't really
1799 fit into html-parse.c. */
1801 /* The function returns the pointer to the malloc-ed quoted version of
1802 string s. It will recognize and quote numeric and special graphic
1803 entities, as per RFC1866:
1811 No other entities are recognized or replaced. */
1813 html_quote_string (const char *s)
1819 /* Pass through the string, and count the new size. */
1820 for (i = 0; *s; s++, i++)
1823 i += 4; /* `amp;' */
1824 else if (*s == '<' || *s == '>')
1825 i += 3; /* `lt;' and `gt;' */
1826 else if (*s == '\"')
1827 i += 5; /* `quot;' */
1831 res = (char *)xmalloc (i + 1);
1833 for (p = res; *s; s++)
1846 *p++ = (*s == '<' ? 'l' : 'g');
1873 /* Determine the width of the terminal we're running on. If that's
1874 not possible, return 0. */
1877 determine_screen_width (void)
1879 /* If there's a way to get the terminal size using POSIX
1880 tcgetattr(), somebody please tell me. */
1885 if (opt.lfilename != NULL)
1888 fd = fileno (stderr);
1889 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1890 return 0; /* most likely ENOTTY */
1893 #else /* not TIOCGWINSZ */
1895 CONSOLE_SCREEN_BUFFER_INFO csbi;
1896 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1898 return csbi.dwSize.X;
1899 # else /* neither WINDOWS nor TIOCGWINSZ */
1901 #endif /* neither WINDOWS nor TIOCGWINSZ */
1902 #endif /* not TIOCGWINSZ */
1905 /* Return a random number between 0 and MAX-1, inclusive.
1907 If MAX is greater than the value of RAND_MAX+1 on the system, the
1908 returned value will be in the range [0, RAND_MAX]. This may be
1909 fixed in a future release.
1911 The random number generator is seeded automatically the first time
1914 This uses rand() for portability. It has been suggested that
1915 random() offers better randomness, but this is not required for
1916 Wget, so I chose to go for simplicity and use rand
1919 DO NOT use this for cryptographic purposes. It is only meant to be
1920 used in situations where quality of the random numbers returned
1921 doesn't really matter. */
1924 random_number (int max)
1932 srand (time (NULL));
1937 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1938 and enforce that assumption by masking other bits. */
1940 # define RAND_MAX 32767
1944 /* This is equivalent to rand() % max, but uses the high-order bits
1945 for better randomness on architecture where rand() is implemented
1946 using a simple congruential generator. */
1948 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1949 return (int)bounded;
1952 /* Return a random uniformly distributed floating point number in the
1953 [0, 1) range. The precision of returned numbers is 9 digits.
1955 Modify this to use erand48() where available! */
1960 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1961 sure it's greater than 1000. */
1962 int rnd1 = random_number (1000);
1963 int rnd2 = random_number (1000);
1964 int rnd3 = random_number (1000);
1965 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1969 /* A debugging function for checking whether an MD5 library works. */
1971 #include "gen-md5.h"
1974 debug_test_md5 (char *buf)
1976 unsigned char raw[16];
1977 static char res[33];
1981 ALLOCA_MD5_CONTEXT (ctx);
1984 gen_md5_update ((unsigned char *)buf, strlen (buf), ctx);
1985 gen_md5_finish (ctx, raw);
1992 *p2++ = XNUM_TO_digit (*p1 >> 4);
1993 *p2++ = XNUM_TO_digit (*p1 & 0xf);
2002 /* Implementation of run_with_timeout, a generic timeout-forcing
2003 routine for systems with Unix-like signal handling. */
2005 #ifdef USE_SIGNAL_TIMEOUT
2006 # ifdef HAVE_SIGSETJMP
2007 # define SETJMP(env) sigsetjmp (env, 1)
2009 static sigjmp_buf run_with_timeout_env;
2012 abort_run_with_timeout (int sig)
2014 assert (sig == SIGALRM);
2015 siglongjmp (run_with_timeout_env, -1);
2017 # else /* not HAVE_SIGSETJMP */
2018 # define SETJMP(env) setjmp (env)
2020 static jmp_buf run_with_timeout_env;
2023 abort_run_with_timeout (int sig)
2025 assert (sig == SIGALRM);
2026 /* We don't have siglongjmp to preserve the set of blocked signals;
2027 if we longjumped out of the handler at this point, SIGALRM would
2028 remain blocked. We must unblock it manually. */
2029 int mask = siggetmask ();
2030 mask &= ~sigmask (SIGALRM);
2033 /* Now it's safe to longjump. */
2034 longjmp (run_with_timeout_env, -1);
2036 # endif /* not HAVE_SIGSETJMP */
2038 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
2039 setitimer where available, alarm otherwise.
2041 TIMEOUT should be non-zero. If the timeout value is so small that
2042 it would be rounded to zero, it is rounded to the least legal value
2043 instead (1us for setitimer, 1s for alarm). That ensures that
2044 SIGALRM will be delivered in all cases. */
2047 alarm_set (double timeout)
2050 /* Use the modern itimer interface. */
2051 struct itimerval itv;
2053 itv.it_value.tv_sec = (long) timeout;
2054 itv.it_value.tv_usec = 1000000L * (timeout - (long)timeout);
2055 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
2056 /* Ensure that we wait for at least the minimum interval.
2057 Specifying zero would mean "wait forever". */
2058 itv.it_value.tv_usec = 1;
2059 setitimer (ITIMER_REAL, &itv, NULL);
2060 #else /* not ITIMER_REAL */
2061 /* Use the old alarm() interface. */
2062 int secs = (int) timeout;
2064 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
2065 because alarm(0) means "never deliver the alarm", i.e. "wait
2066 forever", which is not what someone who specifies a 0.5s
2067 timeout would expect. */
2070 #endif /* not ITIMER_REAL */
2073 /* Cancel the alarm set with alarm_set. */
2079 struct itimerval disable;
2081 setitimer (ITIMER_REAL, &disable, NULL);
2082 #else /* not ITIMER_REAL */
2084 #endif /* not ITIMER_REAL */
2087 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
2088 seconds. Returns non-zero if the function was interrupted with a
2089 timeout, zero otherwise.
2091 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
2092 using setitimer() or alarm(). The timeout is enforced by
2093 longjumping out of the SIGALRM handler. This has several
2094 advantages compared to the traditional approach of relying on
2095 signals causing system calls to exit with EINTR:
2097 * The callback function is *forcibly* interrupted after the
2098 timeout expires, (almost) regardless of what it was doing and
2099 whether it was in a syscall. For example, a calculation that
2100 takes a long time is interrupted as reliably as an IO
2103 * It works with both SYSV and BSD signals because it doesn't
2104 depend on the default setting of SA_RESTART.
2106 * It doesn't special handler setup beyond a simple call to
2107 signal(). (It does use sigsetjmp/siglongjmp, but they're
2110 The only downside is that, if FUN allocates internal resources that
2111 are normally freed prior to exit from the functions, they will be
2112 lost in case of timeout. */
2115 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2125 signal (SIGALRM, abort_run_with_timeout);
2126 if (SETJMP (run_with_timeout_env) != 0)
2128 /* Longjumped out of FUN with a timeout. */
2129 signal (SIGALRM, SIG_DFL);
2132 alarm_set (timeout);
2135 /* Preserve errno in case alarm() or signal() modifies it. */
2136 saved_errno = errno;
2138 signal (SIGALRM, SIG_DFL);
2139 errno = saved_errno;
2144 #else /* not USE_SIGNAL_TIMEOUT */
2147 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2148 define it under Windows, because Windows has its own version of
2149 run_with_timeout that uses threads. */
2152 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2157 #endif /* not WINDOWS */
2158 #endif /* not USE_SIGNAL_TIMEOUT */
2162 /* Sleep the specified amount of seconds. On machines without
2163 nanosleep(), this may sleep shorter if interrupted by signals. */
2166 xsleep (double seconds)
2168 #ifdef HAVE_NANOSLEEP
2169 /* nanosleep is the preferred interface because it offers high
2170 accuracy and, more importantly, because it allows us to reliably
2171 restart after having been interrupted by a signal such as
2173 struct timespec sleep, remaining;
2174 sleep.tv_sec = (long) seconds;
2175 sleep.tv_nsec = 1000000000L * (seconds - (long) seconds);
2176 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2177 /* If nanosleep has been interrupted by a signal, adjust the
2178 sleeping period and return to sleep. */
2180 #else /* not HAVE_NANOSLEEP */
2182 /* If usleep is available, use it in preference to select. */
2185 /* On some systems, usleep cannot handle values larger than
2186 1,000,000. If the period is larger than that, use sleep
2187 first, then add usleep for subsecond accuracy. */
2189 seconds -= (long) seconds;
2191 usleep (seconds * 1000000L);
2192 #else /* not HAVE_USLEEP */
2194 struct timeval sleep;
2195 sleep.tv_sec = (long) seconds;
2196 sleep.tv_usec = 1000000L * (seconds - (long) seconds);
2197 select (0, NULL, NULL, NULL, &sleep);
2198 /* If select returns -1 and errno is EINTR, it means we were
2199 interrupted by a signal. But without knowing how long we've
2200 actually slept, we can't return to sleep. Using gettimeofday to
2201 track sleeps is slow and unreliable due to clock skew. */
2202 #else /* not HAVE_SELECT */
2204 #endif /* not HAVE_SELECT */
2205 #endif /* not HAVE_USLEEP */
2206 #endif /* not HAVE_NANOSLEEP */
2209 #endif /* not WINDOWS */