1 /* Various utility functions.
2 Copyright (C) 2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
36 #else /* not HAVE_STRING_H */
38 #endif /* not HAVE_STRING_H */
39 #include <sys/types.h>
44 # include <sys/mman.h>
55 #ifdef HAVE_SYS_UTIME_H
56 # include <sys/utime.h>
60 # include <libc.h> /* for access() */
64 #ifdef WGET_USE_STDARG
70 /* For TIOCGWINSZ and friends: */
71 #ifdef HAVE_SYS_IOCTL_H
72 # include <sys/ioctl.h>
78 /* Needed for run_with_timeout. */
79 #undef USE_SIGNAL_TIMEOUT
87 #ifndef HAVE_SIGSETJMP
88 /* If sigsetjmp is a macro, configure won't pick it up. */
90 # define HAVE_SIGSETJMP
95 # ifdef HAVE_SIGSETJMP
96 # define USE_SIGNAL_TIMEOUT
99 # define USE_SIGNAL_TIMEOUT
111 /* Utility function: like xstrdup(), but also lowercases S. */
114 xstrdup_lower (const char *s)
116 char *copy = xstrdup (s);
123 /* Copy the string formed by two pointers (one on the beginning, other
124 on the char after the last char) to a new, malloc-ed location.
127 strdupdelim (const char *beg, const char *end)
129 char *res = (char *)xmalloc (end - beg + 1);
130 memcpy (res, beg, end - beg);
131 res[end - beg] = '\0';
135 /* Parse a string containing comma-separated elements, and return a
136 vector of char pointers with the elements. Spaces following the
137 commas are ignored. */
139 sepstring (const char *s)
153 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
154 res[i] = strdupdelim (p, s);
157 /* Skip the blanks following the ','. */
165 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
166 res[i] = strdupdelim (p, s);
171 #ifdef WGET_USE_STDARG
172 # define VA_START(args, arg1) va_start (args, arg1)
174 # define VA_START(args, ignored) va_start (args)
177 /* Like sprintf, but allocates a string of sufficient size with malloc
178 and returns it. GNU libc has a similar function named asprintf,
179 which requires the pointer to the string to be passed. */
182 aprintf (const char *fmt, ...)
184 /* This function is implemented using vsnprintf, which we provide
185 for the systems that don't have it. Therefore, it should be 100%
189 char *str = xmalloc (size);
196 /* See log_vprintf_internal for explanation why it's OK to rely
197 on the return value of vsnprintf. */
199 VA_START (args, fmt);
200 n = vsnprintf (str, size, fmt, args);
203 /* If the printing worked, return the string. */
204 if (n > -1 && n < size)
207 /* Else try again with a larger buffer. */
208 if (n > -1) /* C99 */
209 size = n + 1; /* precisely what is needed */
211 size <<= 1; /* twice the old size */
212 str = xrealloc (str, size);
214 return NULL; /* unreached */
217 /* Concatenate the NULL-terminated list of string arguments into
218 freshly allocated space. */
221 concat_strings (const char *str0, ...)
224 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
227 const char *next_str;
228 int total_length = 0;
231 /* Calculate the length of and allocate the resulting string. */
234 VA_START (args, str0);
235 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
237 int len = strlen (next_str);
238 if (argcount < countof (saved_lengths))
239 saved_lengths[argcount++] = len;
243 p = ret = xmalloc (total_length + 1);
245 /* Copy the strings into the allocated space. */
248 VA_START (args, str0);
249 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
252 if (argcount < countof (saved_lengths))
253 len = saved_lengths[argcount++];
255 len = strlen (next_str);
256 memcpy (p, next_str, len);
265 /* Return pointer to a static char[] buffer in which zero-terminated
266 string-representation of TM (in form hh:mm:ss) is printed.
268 If TM is NULL, the current time will be used. */
271 time_str (time_t *tm)
273 static char output[15];
275 time_t secs = tm ? *tm : time (NULL);
279 /* In case of error, return the empty string. Maybe we should
280 just abort if this happens? */
284 ptm = localtime (&secs);
285 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
289 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
292 datetime_str (time_t *tm)
294 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
296 time_t secs = tm ? *tm : time (NULL);
300 /* In case of error, return the empty string. Maybe we should
301 just abort if this happens? */
305 ptm = localtime (&secs);
306 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
307 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
308 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
312 /* The Windows versions of the following two functions are defined in
317 fork_to_background (void)
320 /* Whether we arrange our own version of opt.lfilename here. */
321 int logfile_changed = 0;
325 /* We must create the file immediately to avoid either a race
326 condition (which arises from using unique_name and failing to
327 use fopen_excl) or lying to the user about the log file name
328 (which arises from using unique_name, printing the name, and
329 using fopen_excl later on.) */
330 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, 0, &opt.lfilename);
346 /* parent, no error */
347 printf (_("Continuing in background, pid %d.\n"), (int)pid);
349 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
350 exit (0); /* #### should we use _exit()? */
353 /* child: give up the privileges and keep running. */
355 freopen ("/dev/null", "r", stdin);
356 freopen ("/dev/null", "w", stdout);
357 freopen ("/dev/null", "w", stderr);
359 #endif /* not WINDOWS */
361 /* "Touch" FILE, i.e. make its atime and mtime equal to the time
362 specified with TM. */
364 touch (const char *file, time_t tm)
366 #ifdef HAVE_STRUCT_UTIMBUF
367 struct utimbuf times;
368 times.actime = times.modtime = tm;
371 times[0] = times[1] = tm;
374 if (utime (file, ×) == -1)
375 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
378 /* Checks if FILE is a symbolic link, and removes it if it is. Does
379 nothing under MS-Windows. */
381 remove_link (const char *file)
386 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
388 DEBUGP (("Unlinking %s (symlink).\n", file));
391 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
392 file, strerror (errno));
397 /* Does FILENAME exist? This is quite a lousy implementation, since
398 it supplies no error codes -- only a yes-or-no answer. Thus it
399 will return that a file does not exist if, e.g., the directory is
400 unreadable. I don't mind it too much currently, though. The
401 proper way should, of course, be to have a third, error state,
402 other than true/false, but that would introduce uncalled-for
403 additional complexity to the callers. */
405 file_exists_p (const char *filename)
408 return access (filename, F_OK) >= 0;
411 return stat (filename, &buf) >= 0;
415 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
416 Returns 0 on error. */
418 file_non_directory_p (const char *path)
421 /* Use lstat() rather than stat() so that symbolic links pointing to
422 directories can be identified correctly. */
423 if (lstat (path, &buf) != 0)
425 return S_ISDIR (buf.st_mode) ? 0 : 1;
428 /* Return the size of file named by FILENAME, or -1 if it cannot be
429 opened or seeked into. */
431 file_size (const char *filename)
433 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
435 /* We use fseek rather than stat to determine the file size because
436 that way we can also verify that the file is readable without
437 explicitly checking for permissions. Inspired by the POST patch
439 FILE *fp = fopen (filename, "rb");
442 fseeko (fp, 0, SEEK_END);
448 if (stat (filename, &st) < 0)
454 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
455 doesn't exist is found. Return a freshly allocated copy of the
459 unique_name_1 (const char *prefix)
462 int plen = strlen (prefix);
463 char *template = (char *)alloca (plen + 1 + 24);
464 char *template_tail = template + plen;
466 memcpy (template, prefix, plen);
467 *template_tail++ = '.';
470 number_to_string (template_tail, count++);
471 while (file_exists_p (template));
473 return xstrdup (template);
476 /* Return a unique file name, based on FILE.
478 More precisely, if FILE doesn't exist, it is returned unmodified.
479 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
480 file name that doesn't exist is returned.
482 The resulting file is not created, only verified that it didn't
483 exist at the point in time when the function was called.
484 Therefore, where security matters, don't rely that the file created
485 by this function exists until you open it with O_EXCL or
488 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
489 string. Otherwise, it may return FILE if the file doesn't exist
490 (and therefore doesn't need changing). */
493 unique_name (const char *file, int allow_passthrough)
495 /* If the FILE itself doesn't exist, return it without
497 if (!file_exists_p (file))
498 return allow_passthrough ? (char *)file : xstrdup (file);
500 /* Otherwise, find a numeric suffix that results in unused file name
502 return unique_name_1 (file);
505 /* Create a file based on NAME, except without overwriting an existing
506 file with that name. Providing O_EXCL is correctly implemented,
507 this function does not have the race condition associated with
508 opening the file returned by unique_name. */
511 unique_create (const char *name, int binary, char **opened_name)
513 /* unique file name, based on NAME */
514 char *uname = unique_name (name, 0);
516 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
519 uname = unique_name (name, 0);
521 if (opened_name && fp != NULL)
524 *opened_name = uname;
536 /* Open the file for writing, with the addition that the file is
537 opened "exclusively". This means that, if the file already exists,
538 this function will *fail* and errno will be set to EEXIST. If
539 BINARY is set, the file will be opened in binary mode, equivalent
542 If opening the file fails for any reason, including the file having
543 previously existed, this function returns NULL and sets errno
547 fopen_excl (const char *fname, int binary)
551 int flags = O_WRONLY | O_CREAT | O_EXCL;
556 fd = open (fname, flags, 0666);
559 return fdopen (fd, binary ? "wb" : "w");
560 #else /* not O_EXCL */
561 return fopen (fname, binary ? "wb" : "w");
562 #endif /* not O_EXCL */
565 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
566 are missing, create them first. In case any mkdir() call fails,
567 return its error status. Returns 0 on successful completion.
569 The behaviour of this function should be identical to the behaviour
570 of `mkdir -p' on systems where mkdir supports the `-p' option. */
572 make_directory (const char *directory)
574 int i, ret, quit = 0;
577 /* Make a copy of dir, to be able to write to it. Otherwise, the
578 function is unsafe if called with a read-only char *argument. */
579 STRDUP_ALLOCA (dir, directory);
581 /* If the first character of dir is '/', skip it (and thus enable
582 creation of absolute-pathname directories. */
583 for (i = (*dir == '/'); 1; ++i)
585 for (; dir[i] && dir[i] != '/'; i++)
590 /* Check whether the directory already exists. Allow creation of
591 of intermediate directories to fail, as the initial path components
592 are not necessarily directories! */
593 if (!file_exists_p (dir))
594 ret = mkdir (dir, 0777);
605 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
606 should be a file name.
608 file_merge("/foo/bar", "baz") => "/foo/baz"
609 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
610 file_merge("foo", "bar") => "bar"
612 In other words, it's a simpler and gentler version of uri_merge_1. */
615 file_merge (const char *base, const char *file)
618 const char *cut = (const char *)strrchr (base, '/');
621 return xstrdup (file);
623 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
624 memcpy (result, base, cut - base);
625 result[cut - base] = '/';
626 strcpy (result + (cut - base) + 1, file);
631 static int in_acclist PARAMS ((const char *const *, const char *, int));
633 /* Determine whether a file is acceptable to be followed, according to
634 lists of patterns to accept/reject. */
636 acceptable (const char *s)
640 while (l && s[l] != '/')
647 return (in_acclist ((const char *const *)opt.accepts, s, 1)
648 && !in_acclist ((const char *const *)opt.rejects, s, 1));
650 return in_acclist ((const char *const *)opt.accepts, s, 1);
652 else if (opt.rejects)
653 return !in_acclist ((const char *const *)opt.rejects, s, 1);
657 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
658 `/something', frontcmp() will return 1 only if S2 begins with
659 `/something'. Otherwise, 0 is returned. */
661 frontcmp (const char *s1, const char *s2)
663 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
667 /* Iterate through STRLIST, and return the first element that matches
668 S, through wildcards or front comparison (as appropriate). */
670 proclist (char **strlist, const char *s, enum accd flags)
674 for (x = strlist; *x; x++)
675 if (has_wildcards_p (*x))
677 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
682 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
689 /* Returns whether DIRECTORY is acceptable for download, wrt the
690 include/exclude lists.
692 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
693 and absolute paths may be freely intermixed. */
695 accdir (const char *directory, enum accd flags)
697 /* Remove starting '/'. */
698 if (flags & ALLABS && *directory == '/')
702 if (!proclist (opt.includes, directory, flags))
707 if (proclist (opt.excludes, directory, flags))
713 /* Return non-zero if STRING ends with TAIL. For instance:
715 match_tail ("abc", "bc", 0) -> 1
716 match_tail ("abc", "ab", 0) -> 0
717 match_tail ("abc", "abc", 0) -> 1
719 If FOLD_CASE_P is non-zero, the comparison will be
723 match_tail (const char *string, const char *tail, int fold_case_p)
727 /* We want this to be fast, so we code two loops, one with
728 case-folding, one without. */
732 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
733 if (string[i] != tail[j])
738 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
739 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
743 /* If the tail was exhausted, the match was succesful. */
750 /* Checks whether string S matches each element of ACCEPTS. A list
751 element are matched either with fnmatch() or match_tail(),
752 according to whether the element contains wildcards or not.
754 If the BACKWARD is 0, don't do backward comparison -- just compare
757 in_acclist (const char *const *accepts, const char *s, int backward)
759 for (; *accepts; accepts++)
761 if (has_wildcards_p (*accepts))
763 /* fnmatch returns 0 if the pattern *does* match the
765 if (fnmatch (*accepts, s, 0) == 0)
772 if (match_tail (s, *accepts, 0))
777 if (!strcmp (s, *accepts))
785 /* Return the location of STR's suffix (file extension). Examples:
786 suffix ("foo.bar") -> "bar"
787 suffix ("foo.bar.baz") -> "baz"
788 suffix ("/foo/bar") -> NULL
789 suffix ("/foo.bar/baz") -> NULL */
791 suffix (const char *str)
795 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
799 return (char *)str + i;
804 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
808 has_wildcards_p (const char *s)
811 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
816 /* Return non-zero if FNAME ends with a typical HTML suffix. The
817 following (case-insensitive) suffixes are presumed to be HTML files:
821 ?html (`?' matches one character)
823 #### CAVEAT. This is not necessarily a good indication that FNAME
824 refers to a file that contains HTML! */
826 has_html_suffix_p (const char *fname)
830 if ((suf = suffix (fname)) == NULL)
832 if (!strcasecmp (suf, "html"))
834 if (!strcasecmp (suf, "htm"))
836 if (suf[0] && !strcasecmp (suf + 1, "html"))
841 /* Read a line from FP and return the pointer to freshly allocated
842 storage. The storage space is obtained through malloc() and should
843 be freed with free() when it is no longer needed.
845 The length of the line is not limited, except by available memory.
846 The newline character at the end of line is retained. The line is
847 terminated with a zero character.
849 After end-of-file is encountered without anything being read, NULL
850 is returned. NULL is also returned on error. To distinguish
851 between these two cases, use the stdio function ferror(). */
854 read_whole_line (FILE *fp)
858 char *line = (char *)xmalloc (bufsize);
860 while (fgets (line + length, bufsize - length, fp))
862 length += strlen (line + length);
864 /* Possible for example when reading from a binary file where
865 a line begins with \0. */
868 if (line[length - 1] == '\n')
871 /* fgets() guarantees to read the whole line, or to use up the
872 space we've given it. We can double the buffer
875 line = xrealloc (line, bufsize);
877 if (length == 0 || ferror (fp))
882 if (length + 1 < bufsize)
883 /* Relieve the memory from our exponential greediness. We say
884 `length + 1' because the terminating \0 is not included in
885 LENGTH. We don't need to zero-terminate the string ourselves,
886 though, because fgets() does that. */
887 line = xrealloc (line, length + 1);
891 /* Read FILE into memory. A pointer to `struct file_memory' are
892 returned; use struct element `content' to access file contents, and
893 the element `length' to know the file length. `content' is *not*
894 zero-terminated, and you should *not* read or write beyond the [0,
895 length) range of characters.
897 After you are done with the file contents, call read_file_free to
900 Depending on the operating system and the type of file that is
901 being read, read_file() either mmap's the file into memory, or
902 reads the file into the core using read().
904 If file is named "-", fileno(stdin) is used for reading instead.
905 If you want to read from a real file named "-", use "./-" instead. */
908 read_file (const char *file)
911 struct file_memory *fm;
913 int inhibit_close = 0;
915 /* Some magic in the finest tradition of Perl and its kin: if FILE
916 is "-", just use stdin. */
921 /* Note that we don't inhibit mmap() in this case. If stdin is
922 redirected from a regular file, mmap() will still work. */
925 fd = open (file, O_RDONLY);
928 fm = xnew (struct file_memory);
933 if (fstat (fd, &buf) < 0)
935 fm->length = buf.st_size;
936 /* NOTE: As far as I know, the callers of this function never
937 modify the file text. Relying on this would enable us to
938 specify PROT_READ and MAP_SHARED for a marginal gain in
939 efficiency, but at some cost to generality. */
940 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
942 if (fm->content == (char *)MAP_FAILED)
952 /* The most common reason why mmap() fails is that FD does not point
953 to a plain file. However, it's also possible that mmap() doesn't
954 work for a particular type of file. Therefore, whenever mmap()
955 fails, we just fall back to the regular method. */
956 #endif /* HAVE_MMAP */
959 size = 512; /* number of bytes fm->contents can
960 hold at any given time. */
961 fm->content = xmalloc (size);
965 if (fm->length > size / 2)
967 /* #### I'm not sure whether the whole exponential-growth
968 thing makes sense with kernel read. On Linux at least,
969 read() refuses to read more than 4K from a file at a
970 single chunk anyway. But other Unixes might optimize it
971 better, and it doesn't *hurt* anything, so I'm leaving
974 /* Normally, we grow SIZE exponentially to make the number
975 of calls to read() and realloc() logarithmic in relation
976 to file size. However, read() can read an amount of data
977 smaller than requested, and it would be unreasonable to
978 double SIZE every time *something* was read. Therefore,
979 we double SIZE only when the length exceeds half of the
980 entire allocated size. */
982 fm->content = xrealloc (fm->content, size);
984 nread = read (fd, fm->content + fm->length, size - fm->length);
986 /* Successful read. */
997 if (size > fm->length && fm->length != 0)
998 /* Due to exponential growth of fm->content, the allocated region
999 might be much larger than what is actually needed. */
1000 fm->content = xrealloc (fm->content, fm->length);
1007 xfree (fm->content);
1012 /* Release the resources held by FM. Specifically, this calls
1013 munmap() or xfree() on fm->content, depending whether mmap or
1014 malloc/read were used to read in the file. It also frees the
1015 memory needed to hold the FM structure itself. */
1018 read_file_free (struct file_memory *fm)
1023 munmap (fm->content, fm->length);
1028 xfree (fm->content);
1033 /* Free the pointers in a NULL-terminated vector of pointers, then
1034 free the pointer itself. */
1036 free_vec (char **vec)
1047 /* Append vector V2 to vector V1. The function frees V2 and
1048 reallocates V1 (thus you may not use the contents of neither
1049 pointer after the call). If V1 is NULL, V2 is returned. */
1051 merge_vecs (char **v1, char **v2)
1061 /* To avoid j == 0 */
1066 for (i = 0; v1[i]; i++);
1068 for (j = 0; v2[j]; j++);
1069 /* Reallocate v1. */
1070 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1071 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1076 /* A set of simple-minded routines to store strings in a linked list.
1077 This used to also be used for searching, but now we have hash
1080 /* It's a shame that these simple things like linked lists and hash
1081 tables (see hash.c) need to be implemented over and over again. It
1082 would be nice to be able to use the routines from glib -- see
1083 www.gtk.org for details. However, that would make Wget depend on
1084 glib, and I want to avoid dependencies to external libraries for
1085 reasons of convenience and portability (I suspect Wget is more
1086 portable than anything ever written for Gnome). */
1088 /* Append an element to the list. If the list has a huge number of
1089 elements, this can get slow because it has to find the list's
1090 ending. If you think you have to call slist_append in a loop,
1091 think about calling slist_prepend() followed by slist_nreverse(). */
1094 slist_append (slist *l, const char *s)
1096 slist *newel = xnew (slist);
1099 newel->string = xstrdup (s);
1104 /* Find the last element. */
1111 /* Prepend S to the list. Unlike slist_append(), this is O(1). */
1114 slist_prepend (slist *l, const char *s)
1116 slist *newel = xnew (slist);
1117 newel->string = xstrdup (s);
1122 /* Destructively reverse L. */
1125 slist_nreverse (slist *l)
1130 slist *next = l->next;
1138 /* Is there a specific entry in the list? */
1140 slist_contains (slist *l, const char *s)
1142 for (; l; l = l->next)
1143 if (!strcmp (l->string, s))
1148 /* Free the whole slist. */
1150 slist_free (slist *l)
1161 /* Sometimes it's useful to create "sets" of strings, i.e. special
1162 hash tables where you want to store strings as keys and merely
1163 query for their existence. Here is a set of utility routines that
1164 makes that transparent. */
1167 string_set_add (struct hash_table *ht, const char *s)
1169 /* First check whether the set element already exists. If it does,
1170 do nothing so that we don't have to free() the old element and
1171 then strdup() a new one. */
1172 if (hash_table_contains (ht, s))
1175 /* We use "1" as value. It provides us a useful and clear arbitrary
1176 value, and it consumes no memory -- the pointers to the same
1177 string "1" will be shared by all the key-value pairs in all `set'
1179 hash_table_put (ht, xstrdup (s), "1");
1182 /* Synonym for hash_table_contains... */
1185 string_set_contains (struct hash_table *ht, const char *s)
1187 return hash_table_contains (ht, s);
1191 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1198 string_set_free (struct hash_table *ht)
1200 hash_table_map (ht, string_set_free_mapper, NULL);
1201 hash_table_destroy (ht);
1205 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1212 /* Another utility function: call free() on all keys and values of HT. */
1215 free_keys_and_values (struct hash_table *ht)
1217 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1221 /* Engine for legible and legible_large_int; add thousand separators
1222 to numbers printed in strings. */
1225 legible_1 (const char *repr)
1227 static char outbuf[48];
1232 /* Reset the pointers. */
1236 /* Ignore the sign for the purpose of adding thousand
1243 /* How many digits before the first separator? */
1244 mod = strlen (inptr) % 3;
1246 for (i = 0; i < mod; i++)
1247 *outptr++ = inptr[i];
1248 /* Now insert the rest of them, putting separator before every
1250 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1252 if (i % 3 == 0 && i1 != 0)
1254 *outptr++ = inptr[i1];
1256 /* Zero-terminate the string. */
1261 /* Legible -- return a static pointer to the legibly printed wgint. */
1267 /* Print the number into the buffer. */
1268 number_to_string (inbuf, l);
1269 return legible_1 (inbuf);
1272 /* Write a string representation of LARGE_INT NUMBER into the provided
1273 buffer. The buffer should be able to accept 24 characters,
1274 including the terminating zero.
1276 It would be dangerous to use sprintf, because the code wouldn't
1277 work on a machine with gcc-provided long long support, but without
1278 libc support for "%lld". However, such platforms will typically
1279 not have snprintf and will use our version, which does support
1280 "%lld" where long longs are available. */
1283 large_int_to_string (char *buffer, LARGE_INT number)
1285 snprintf (buffer, 24, LARGE_INT_FMT, number);
1288 /* The same as legible(), but works on LARGE_INT. */
1291 legible_large_int (LARGE_INT l)
1294 large_int_to_string (inbuf, l);
1295 return legible_1 (inbuf);
1298 /* Count the digits in an integer number. */
1300 numdigit (wgint number)
1308 while ((number /= 10) > 0)
1313 #define ONE_DIGIT(figure) *p++ = n / (figure) + '0'
1314 #define ONE_DIGIT_ADVANCE(figure) (ONE_DIGIT (figure), n %= (figure))
1316 #define DIGITS_1(figure) ONE_DIGIT (figure)
1317 #define DIGITS_2(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_1 ((figure) / 10)
1318 #define DIGITS_3(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_2 ((figure) / 10)
1319 #define DIGITS_4(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_3 ((figure) / 10)
1320 #define DIGITS_5(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_4 ((figure) / 10)
1321 #define DIGITS_6(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_5 ((figure) / 10)
1322 #define DIGITS_7(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_6 ((figure) / 10)
1323 #define DIGITS_8(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_7 ((figure) / 10)
1324 #define DIGITS_9(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_8 ((figure) / 10)
1325 #define DIGITS_10(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_9 ((figure) / 10)
1327 /* DIGITS_<11-20> are only used on machines with 64-bit numbers. */
1329 #define DIGITS_11(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_10 ((figure) / 10)
1330 #define DIGITS_12(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_11 ((figure) / 10)
1331 #define DIGITS_13(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_12 ((figure) / 10)
1332 #define DIGITS_14(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_13 ((figure) / 10)
1333 #define DIGITS_15(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_14 ((figure) / 10)
1334 #define DIGITS_16(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_15 ((figure) / 10)
1335 #define DIGITS_17(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_16 ((figure) / 10)
1336 #define DIGITS_18(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_17 ((figure) / 10)
1337 #define DIGITS_19(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_18 ((figure) / 10)
1339 /* It is annoying that we have three different syntaxes for 64-bit constants:
1340 - nnnL for 64-bit systems, where they are of type long;
1341 - nnnLL for 32-bit systems that support long long;
1342 - nnnI64 for MS compiler on Windows, which doesn't support long long. */
1345 /* If long is large enough, use long constants. */
1346 # define C10000000000 10000000000L
1347 # define C100000000000 100000000000L
1348 # define C1000000000000 1000000000000L
1349 # define C10000000000000 10000000000000L
1350 # define C100000000000000 100000000000000L
1351 # define C1000000000000000 1000000000000000L
1352 # define C10000000000000000 10000000000000000L
1353 # define C100000000000000000 100000000000000000L
1354 # define C1000000000000000000 1000000000000000000L
1356 # if SIZEOF_LONG_LONG != 0
1357 /* Otherwise, if long long is available, use long long constants. */
1358 # define C10000000000 10000000000LL
1359 # define C100000000000 100000000000LL
1360 # define C1000000000000 1000000000000LL
1361 # define C10000000000000 10000000000000LL
1362 # define C100000000000000 100000000000000LL
1363 # define C1000000000000000 1000000000000000LL
1364 # define C10000000000000000 10000000000000000LL
1365 # define C100000000000000000 100000000000000000LL
1366 # define C1000000000000000000 1000000000000000000LL
1368 # if defined(WINDOWS)
1369 /* Use __int64 constants under Windows. */
1370 # define C10000000000 10000000000I64
1371 # define C100000000000 100000000000I64
1372 # define C1000000000000 1000000000000I64
1373 # define C10000000000000 10000000000000I64
1374 # define C100000000000000 100000000000000I64
1375 # define C1000000000000000 1000000000000000I64
1376 # define C10000000000000000 10000000000000000I64
1377 # define C100000000000000000 100000000000000000I64
1378 # define C1000000000000000000 1000000000000000000I64
1383 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1384 cases and to portably support strange sizes of wgint. */
1385 #if SIZEOF_LONG >= SIZEOF_WGINT
1386 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%ld", (long) (n))
1388 # if SIZEOF_LONG_LONG >= SIZEOF_WGINT
1389 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%lld", (long long) (n))
1392 # define SPRINTF_WGINT(buf, n) sprintf(buf, "%I64", (__int64) (n))
1397 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1398 `sprintf(buffer, "%lld", (long long) number)', only much faster and
1399 portable to machines without long long.
1401 The speedup may make a difference in programs that frequently
1402 convert numbers to strings. Some implementations of sprintf,
1403 particularly the one in GNU libc, have been known to be extremely
1404 slow when converting integers to strings.
1406 Return the pointer to the location where the terminating zero was
1407 printed. (Equivalent to calling buffer+strlen(buffer) after the
1410 BUFFER should be big enough to accept as many bytes as you expect
1411 the number to take up. On machines with 64-bit longs the maximum
1412 needed size is 24 bytes. That includes the digits needed for the
1413 largest 64-bit number, the `-' sign in case it's negative, and the
1414 terminating '\0'. */
1417 number_to_string (char *buffer, wgint number)
1422 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1423 /* We are running in a strange or misconfigured environment. Let
1424 sprintf cope with it. */
1425 SPRINTF_WGINT (buffer, n);
1426 p += strlen (buffer);
1427 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1433 /* We cannot print a '-' and assign -n to n because -n would
1434 overflow. Let sprintf deal with this border case. */
1435 SPRINTF_WGINT (buffer, n);
1436 p += strlen (buffer);
1444 if (n < 10) { DIGITS_1 (1); }
1445 else if (n < 100) { DIGITS_2 (10); }
1446 else if (n < 1000) { DIGITS_3 (100); }
1447 else if (n < 10000) { DIGITS_4 (1000); }
1448 else if (n < 100000) { DIGITS_5 (10000); }
1449 else if (n < 1000000) { DIGITS_6 (100000); }
1450 else if (n < 10000000) { DIGITS_7 (1000000); }
1451 else if (n < 100000000) { DIGITS_8 (10000000); }
1452 else if (n < 1000000000) { DIGITS_9 (100000000); }
1453 #if SIZEOF_WGINT == 4
1454 /* wgint is four bytes long: we're done. */
1455 /* ``if (1)'' serves only to preserve editor indentation. */
1456 else if (1) { DIGITS_10 (1000000000); }
1458 /* wgint is 64 bits long -- make sure to process all the digits. */
1459 else if (n < C10000000000) { DIGITS_10 (1000000000); }
1460 else if (n < C100000000000) { DIGITS_11 (C10000000000); }
1461 else if (n < C1000000000000) { DIGITS_12 (C100000000000); }
1462 else if (n < C10000000000000) { DIGITS_13 (C1000000000000); }
1463 else if (n < C100000000000000) { DIGITS_14 (C10000000000000); }
1464 else if (n < C1000000000000000) { DIGITS_15 (C100000000000000); }
1465 else if (n < C10000000000000000) { DIGITS_16 (C1000000000000000); }
1466 else if (n < C100000000000000000) { DIGITS_17 (C10000000000000000); }
1467 else if (n < C1000000000000000000) { DIGITS_18 (C100000000000000000); }
1468 else { DIGITS_19 (C1000000000000000000); }
1472 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1478 #undef ONE_DIGIT_ADVANCE
1502 /* Print NUMBER to a statically allocated string and return a pointer
1503 to the printed representation.
1505 This function is intended to be used in conjunction with printf.
1506 It is hard to portably print wgint values:
1507 a) you cannot use printf("%ld", number) because wgint can be long
1508 long on 32-bit machines with LFS.
1509 b) you cannot use printf("%lld", number) because NUMBER could be
1510 long on 32-bit machines without LFS, or on 64-bit machines,
1511 which do not require LFS. Also, Windows doesn't support %lld.
1512 c) you cannot use printf("%j", (int_max_t) number) because not all
1513 versions of printf support "%j", the most notable being the one
1515 d) you cannot #define WGINT_FMT to the appropriate format and use
1516 printf(WGINT_FMT, number) because that would break translations
1517 for user-visible messages, such as printf("Downloaded: %d
1520 What you should use instead is printf("%s", number_to_static_string
1523 CAVEAT: since the function returns pointers to static data, you
1524 must be careful to copy its result before calling it again.
1525 However, to make it more useful with printf, the function maintains
1526 an internal ring of static buffers to return. That way things like
1527 printf("%s %s", number_to_static_string (num1),
1528 number_to_static_string (num2)) work as expected. Three buffers
1529 are currently used, which means that "%s %s %s" will work, but "%s
1530 %s %s %s" won't. If you need to print more than three wgints,
1531 bump the RING_SIZE (or rethink your message.) */
1534 number_to_static_string (wgint number)
1536 static char ring[RING_SIZE][24];
1538 char *buf = ring[ringpos];
1539 number_to_string (buf, number);
1540 ringpos = (ringpos + 1) % RING_SIZE;
1544 /* Support for timers. */
1546 #undef TIMER_WINDOWS
1547 #undef TIMER_GETTIMEOFDAY
1550 /* Depending on the OS and availability of gettimeofday(), one and
1551 only one of the above constants will be defined. Virtually all
1552 modern Unix systems will define TIMER_GETTIMEOFDAY; Windows will
1553 use TIMER_WINDOWS. TIMER_TIME is a catch-all method for
1554 non-Windows systems without gettimeofday. */
1557 # define TIMER_WINDOWS
1558 #else /* not WINDOWS */
1559 # ifdef HAVE_GETTIMEOFDAY
1560 # define TIMER_GETTIMEOFDAY
1564 #endif /* not WINDOWS */
1566 #ifdef TIMER_GETTIMEOFDAY
1567 typedef struct timeval wget_sys_time;
1571 typedef time_t wget_sys_time;
1574 #ifdef TIMER_WINDOWS
1576 DWORD lores; /* In case GetTickCount is used */
1577 LARGE_INTEGER hires; /* In case high-resolution timer is used */
1582 /* Whether the start time has been initialized. */
1585 /* The starting point in time which, subtracted from the current
1586 time, yields elapsed time. */
1587 wget_sys_time start;
1589 /* The most recent elapsed time, calculated by wtimer_elapsed().
1590 Measured in milliseconds. */
1591 double elapsed_last;
1593 /* Approximately, the time elapsed between the true start of the
1594 measurement and the time represented by START. */
1595 double elapsed_pre_start;
1598 #ifdef TIMER_WINDOWS
1600 /* Whether high-resolution timers are used. Set by wtimer_initialize_once
1601 the first time wtimer_allocate is called. */
1602 static int using_hires_timers;
1604 /* Frequency of high-resolution timers -- number of updates per
1605 millisecond. Calculated the first time wtimer_allocate is called
1606 provided that high-resolution timers are available. */
1607 static double hires_millisec_freq;
1609 /* The first time a timer is created, determine whether to use
1610 high-resolution timers. */
1613 wtimer_initialize_once (void)
1615 static int init_done;
1621 QueryPerformanceFrequency (&freq);
1622 if (freq.QuadPart != 0)
1624 using_hires_timers = 1;
1625 hires_millisec_freq = (double) freq.QuadPart / 1000.0;
1629 #endif /* TIMER_WINDOWS */
1631 /* Allocate a timer. Calling wtimer_read on the timer will return
1632 zero. It is not legal to call wtimer_update with a freshly
1633 allocated timer -- use wtimer_reset first. */
1636 wtimer_allocate (void)
1638 struct wget_timer *wt = xnew (struct wget_timer);
1641 #ifdef TIMER_WINDOWS
1642 wtimer_initialize_once ();
1648 /* Allocate a new timer and reset it. Return the new timer. */
1653 struct wget_timer *wt = wtimer_allocate ();
1658 /* Free the resources associated with the timer. Its further use is
1662 wtimer_delete (struct wget_timer *wt)
1667 /* Store system time to WST. */
1670 wtimer_sys_set (wget_sys_time *wst)
1672 #ifdef TIMER_GETTIMEOFDAY
1673 gettimeofday (wst, NULL);
1680 #ifdef TIMER_WINDOWS
1681 if (using_hires_timers)
1683 QueryPerformanceCounter (&wst->hires);
1687 /* Where hires counters are not available, use GetTickCount rather
1688 GetSystemTime, because it is unaffected by clock skew and simpler
1689 to use. Note that overflows don't affect us because we never use
1690 absolute values of the ticker, only the differences. */
1691 wst->lores = GetTickCount ();
1696 /* Reset timer WT. This establishes the starting point from which
1697 wtimer_elapsed() will return the number of elapsed milliseconds.
1698 It is allowed to reset a previously used timer. */
1701 wtimer_reset (struct wget_timer *wt)
1703 /* Set the start time to the current time. */
1704 wtimer_sys_set (&wt->start);
1705 wt->elapsed_last = 0;
1706 wt->elapsed_pre_start = 0;
1707 wt->initialized = 1;
1711 wtimer_sys_diff (wget_sys_time *wst1, wget_sys_time *wst2)
1713 #ifdef TIMER_GETTIMEOFDAY
1714 return ((double)(wst1->tv_sec - wst2->tv_sec) * 1000
1715 + (double)(wst1->tv_usec - wst2->tv_usec) / 1000);
1719 return 1000 * (*wst1 - *wst2);
1723 if (using_hires_timers)
1724 return (wst1->hires.QuadPart - wst2->hires.QuadPart) / hires_millisec_freq;
1726 return wst1->lores - wst2->lores;
1730 /* Update the timer's elapsed interval. This function causes the
1731 timer to call gettimeofday (or time(), etc.) to update its idea of
1732 current time. To get the elapsed interval in milliseconds, use
1735 This function handles clock skew, i.e. time that moves backwards is
1739 wtimer_update (struct wget_timer *wt)
1744 assert (wt->initialized != 0);
1746 wtimer_sys_set (&now);
1747 elapsed = wt->elapsed_pre_start + wtimer_sys_diff (&now, &wt->start);
1749 /* Ideally we'd just return the difference between NOW and
1750 wt->start. However, the system timer can be set back, and we
1751 could return a value smaller than when we were last called, even
1752 a negative value. Both of these would confuse the callers, which
1753 expect us to return monotonically nondecreasing values.
1755 Therefore: if ELAPSED is smaller than its previous known value,
1756 we reset wt->start to the current time and effectively start
1757 measuring from this point. But since we don't want the elapsed
1758 value to start from zero, we set elapsed_pre_start to the last
1759 elapsed time and increment all future calculations by that
1762 if (elapsed < wt->elapsed_last)
1765 wt->elapsed_pre_start = wt->elapsed_last;
1766 elapsed = wt->elapsed_last;
1769 wt->elapsed_last = elapsed;
1772 /* Return the elapsed time in milliseconds between the last call to
1773 wtimer_reset and the last call to wtimer_update.
1775 A typical use of the timer interface would be:
1777 struct wtimer *timer = wtimer_new ();
1778 ... do something that takes a while ...
1780 double msecs = wtimer_read (); */
1783 wtimer_read (const struct wget_timer *wt)
1785 return wt->elapsed_last;
1788 /* Return the assessed granularity of the timer implementation, in
1789 milliseconds. This is used by code that tries to substitute a
1790 better value for timers that have returned zero. */
1793 wtimer_granularity (void)
1795 #ifdef TIMER_GETTIMEOFDAY
1796 /* Granularity of gettimeofday varies wildly between architectures.
1797 However, it appears that on modern machines it tends to be better
1798 than 1ms. Assume 100 usecs. (Perhaps the configure process
1799 could actually measure this?) */
1807 #ifdef TIMER_WINDOWS
1808 if (using_hires_timers)
1809 return 1.0 / hires_millisec_freq;
1811 return 10; /* according to MSDN */
1815 /* This should probably be at a better place, but it doesn't really
1816 fit into html-parse.c. */
1818 /* The function returns the pointer to the malloc-ed quoted version of
1819 string s. It will recognize and quote numeric and special graphic
1820 entities, as per RFC1866:
1828 No other entities are recognized or replaced. */
1830 html_quote_string (const char *s)
1836 /* Pass through the string, and count the new size. */
1837 for (i = 0; *s; s++, i++)
1840 i += 4; /* `amp;' */
1841 else if (*s == '<' || *s == '>')
1842 i += 3; /* `lt;' and `gt;' */
1843 else if (*s == '\"')
1844 i += 5; /* `quot;' */
1848 res = (char *)xmalloc (i + 1);
1850 for (p = res; *s; s++)
1863 *p++ = (*s == '<' ? 'l' : 'g');
1890 /* Determine the width of the terminal we're running on. If that's
1891 not possible, return 0. */
1894 determine_screen_width (void)
1896 /* If there's a way to get the terminal size using POSIX
1897 tcgetattr(), somebody please tell me. */
1902 if (opt.lfilename != NULL)
1905 fd = fileno (stderr);
1906 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1907 return 0; /* most likely ENOTTY */
1910 #else /* not TIOCGWINSZ */
1912 CONSOLE_SCREEN_BUFFER_INFO csbi;
1913 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1915 return csbi.dwSize.X;
1916 # else /* neither WINDOWS nor TIOCGWINSZ */
1918 #endif /* neither WINDOWS nor TIOCGWINSZ */
1919 #endif /* not TIOCGWINSZ */
1922 /* Return a random number between 0 and MAX-1, inclusive.
1924 If MAX is greater than the value of RAND_MAX+1 on the system, the
1925 returned value will be in the range [0, RAND_MAX]. This may be
1926 fixed in a future release.
1928 The random number generator is seeded automatically the first time
1931 This uses rand() for portability. It has been suggested that
1932 random() offers better randomness, but this is not required for
1933 Wget, so I chose to go for simplicity and use rand
1936 DO NOT use this for cryptographic purposes. It is only meant to be
1937 used in situations where quality of the random numbers returned
1938 doesn't really matter. */
1941 random_number (int max)
1949 srand (time (NULL));
1954 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1955 and enforce that assumption by masking other bits. */
1957 # define RAND_MAX 32767
1961 /* This is equivalent to rand() % max, but uses the high-order bits
1962 for better randomness on architecture where rand() is implemented
1963 using a simple congruential generator. */
1965 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1966 return (int)bounded;
1969 /* Return a random uniformly distributed floating point number in the
1970 [0, 1) range. The precision of returned numbers is 9 digits.
1972 Modify this to use erand48() where available! */
1977 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1978 sure it's greater than 1000. */
1979 int rnd1 = random_number (1000);
1980 int rnd2 = random_number (1000);
1981 int rnd3 = random_number (1000);
1982 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1985 /* Implementation of run_with_timeout, a generic timeout-forcing
1986 routine for systems with Unix-like signal handling. */
1988 #ifdef USE_SIGNAL_TIMEOUT
1989 # ifdef HAVE_SIGSETJMP
1990 # define SETJMP(env) sigsetjmp (env, 1)
1992 static sigjmp_buf run_with_timeout_env;
1995 abort_run_with_timeout (int sig)
1997 assert (sig == SIGALRM);
1998 siglongjmp (run_with_timeout_env, -1);
2000 # else /* not HAVE_SIGSETJMP */
2001 # define SETJMP(env) setjmp (env)
2003 static jmp_buf run_with_timeout_env;
2006 abort_run_with_timeout (int sig)
2008 assert (sig == SIGALRM);
2009 /* We don't have siglongjmp to preserve the set of blocked signals;
2010 if we longjumped out of the handler at this point, SIGALRM would
2011 remain blocked. We must unblock it manually. */
2012 int mask = siggetmask ();
2013 mask &= ~sigmask (SIGALRM);
2016 /* Now it's safe to longjump. */
2017 longjmp (run_with_timeout_env, -1);
2019 # endif /* not HAVE_SIGSETJMP */
2021 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
2022 setitimer where available, alarm otherwise.
2024 TIMEOUT should be non-zero. If the timeout value is so small that
2025 it would be rounded to zero, it is rounded to the least legal value
2026 instead (1us for setitimer, 1s for alarm). That ensures that
2027 SIGALRM will be delivered in all cases. */
2030 alarm_set (double timeout)
2033 /* Use the modern itimer interface. */
2034 struct itimerval itv;
2036 itv.it_value.tv_sec = (long) timeout;
2037 itv.it_value.tv_usec = 1000000L * (timeout - (long)timeout);
2038 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
2039 /* Ensure that we wait for at least the minimum interval.
2040 Specifying zero would mean "wait forever". */
2041 itv.it_value.tv_usec = 1;
2042 setitimer (ITIMER_REAL, &itv, NULL);
2043 #else /* not ITIMER_REAL */
2044 /* Use the old alarm() interface. */
2045 int secs = (int) timeout;
2047 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
2048 because alarm(0) means "never deliver the alarm", i.e. "wait
2049 forever", which is not what someone who specifies a 0.5s
2050 timeout would expect. */
2053 #endif /* not ITIMER_REAL */
2056 /* Cancel the alarm set with alarm_set. */
2062 struct itimerval disable;
2064 setitimer (ITIMER_REAL, &disable, NULL);
2065 #else /* not ITIMER_REAL */
2067 #endif /* not ITIMER_REAL */
2070 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
2071 seconds. Returns non-zero if the function was interrupted with a
2072 timeout, zero otherwise.
2074 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
2075 using setitimer() or alarm(). The timeout is enforced by
2076 longjumping out of the SIGALRM handler. This has several
2077 advantages compared to the traditional approach of relying on
2078 signals causing system calls to exit with EINTR:
2080 * The callback function is *forcibly* interrupted after the
2081 timeout expires, (almost) regardless of what it was doing and
2082 whether it was in a syscall. For example, a calculation that
2083 takes a long time is interrupted as reliably as an IO
2086 * It works with both SYSV and BSD signals because it doesn't
2087 depend on the default setting of SA_RESTART.
2089 * It doesn't special handler setup beyond a simple call to
2090 signal(). (It does use sigsetjmp/siglongjmp, but they're
2093 The only downside is that, if FUN allocates internal resources that
2094 are normally freed prior to exit from the functions, they will be
2095 lost in case of timeout. */
2098 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2108 signal (SIGALRM, abort_run_with_timeout);
2109 if (SETJMP (run_with_timeout_env) != 0)
2111 /* Longjumped out of FUN with a timeout. */
2112 signal (SIGALRM, SIG_DFL);
2115 alarm_set (timeout);
2118 /* Preserve errno in case alarm() or signal() modifies it. */
2119 saved_errno = errno;
2121 signal (SIGALRM, SIG_DFL);
2122 errno = saved_errno;
2127 #else /* not USE_SIGNAL_TIMEOUT */
2130 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2131 define it under Windows, because Windows has its own version of
2132 run_with_timeout that uses threads. */
2135 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2140 #endif /* not WINDOWS */
2141 #endif /* not USE_SIGNAL_TIMEOUT */
2145 /* Sleep the specified amount of seconds. On machines without
2146 nanosleep(), this may sleep shorter if interrupted by signals. */
2149 xsleep (double seconds)
2151 #ifdef HAVE_NANOSLEEP
2152 /* nanosleep is the preferred interface because it offers high
2153 accuracy and, more importantly, because it allows us to reliably
2154 restart receiving a signal such as SIGWINCH. (There was an
2155 actual Debian bug report about --limit-rate malfunctioning while
2156 the terminal was being resized.) */
2157 struct timespec sleep, remaining;
2158 sleep.tv_sec = (long) seconds;
2159 sleep.tv_nsec = 1000000000L * (seconds - (long) seconds);
2160 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2161 /* If nanosleep has been interrupted by a signal, adjust the
2162 sleeping period and return to sleep. */
2164 #else /* not HAVE_NANOSLEEP */
2166 /* If usleep is available, use it in preference to select. */
2169 /* On some systems, usleep cannot handle values larger than
2170 1,000,000. If the period is larger than that, use sleep
2171 first, then add usleep for subsecond accuracy. */
2173 seconds -= (long) seconds;
2175 usleep (seconds * 1000000L);
2176 #else /* not HAVE_USLEEP */
2178 struct timeval sleep;
2179 sleep.tv_sec = (long) seconds;
2180 sleep.tv_usec = 1000000L * (seconds - (long) seconds);
2181 select (0, NULL, NULL, NULL, &sleep);
2182 /* If select returns -1 and errno is EINTR, it means we were
2183 interrupted by a signal. But without knowing how long we've
2184 actually slept, we can't return to sleep. Using gettimeofday to
2185 track sleeps is slow and unreliable due to clock skew. */
2186 #else /* not HAVE_SELECT */
2188 #endif /* not HAVE_SELECT */
2189 #endif /* not HAVE_USLEEP */
2190 #endif /* not HAVE_NANOSLEEP */
2193 #endif /* not WINDOWS */