1 /* Various utility functions.
2 Copyright (C) 2003 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
36 #else /* not HAVE_STRING_H */
38 #endif /* not HAVE_STRING_H */
39 #include <sys/types.h>
44 # include <sys/mman.h>
55 #ifdef HAVE_SYS_UTIME_H
56 # include <sys/utime.h>
60 # include <libc.h> /* for access() */
64 #ifdef WGET_USE_STDARG
70 /* For TIOCGWINSZ and friends: */
71 #ifdef HAVE_SYS_IOCTL_H
72 # include <sys/ioctl.h>
78 /* Needed for run_with_timeout. */
79 #undef USE_SIGNAL_TIMEOUT
87 #ifndef HAVE_SIGSETJMP
88 /* If sigsetjmp is a macro, configure won't pick it up. */
90 # define HAVE_SIGSETJMP
95 # ifdef HAVE_SIGSETJMP
96 # define USE_SIGNAL_TIMEOUT
99 # define USE_SIGNAL_TIMEOUT
111 /* Utility function: like xstrdup(), but also lowercases S. */
114 xstrdup_lower (const char *s)
116 char *copy = xstrdup (s);
123 /* Return a count of how many times CHR occurs in STRING. */
126 count_char (const char *string, char chr)
130 for (p = string; *p; p++)
136 /* Copy the string formed by two pointers (one on the beginning, other
137 on the char after the last char) to a new, malloc-ed location.
140 strdupdelim (const char *beg, const char *end)
142 char *res = (char *)xmalloc (end - beg + 1);
143 memcpy (res, beg, end - beg);
144 res[end - beg] = '\0';
148 /* Parse a string containing comma-separated elements, and return a
149 vector of char pointers with the elements. Spaces following the
150 commas are ignored. */
152 sepstring (const char *s)
166 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
167 res[i] = strdupdelim (p, s);
170 /* Skip the blanks following the ','. */
178 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
179 res[i] = strdupdelim (p, s);
184 #ifdef WGET_USE_STDARG
185 # define VA_START(args, arg1) va_start (args, arg1)
187 # define VA_START(args, ignored) va_start (args)
190 /* Like sprintf, but allocates a string of sufficient size with malloc
191 and returns it. GNU libc has a similar function named asprintf,
192 which requires the pointer to the string to be passed. */
195 aprintf (const char *fmt, ...)
197 /* This function is implemented using vsnprintf, which we provide
198 for the systems that don't have it. Therefore, it should be 100%
202 char *str = xmalloc (size);
209 /* See log_vprintf_internal for explanation why it's OK to rely
210 on the return value of vsnprintf. */
212 VA_START (args, fmt);
213 n = vsnprintf (str, size, fmt, args);
216 /* If the printing worked, return the string. */
217 if (n > -1 && n < size)
220 /* Else try again with a larger buffer. */
221 if (n > -1) /* C99 */
222 size = n + 1; /* precisely what is needed */
224 size <<= 1; /* twice the old size */
225 str = xrealloc (str, size);
227 return NULL; /* unreached */
230 /* Return pointer to a static char[] buffer in which zero-terminated
231 string-representation of TM (in form hh:mm:ss) is printed.
233 If TM is NULL, the current time will be used. */
236 time_str (time_t *tm)
238 static char output[15];
240 time_t secs = tm ? *tm : time (NULL);
244 /* In case of error, return the empty string. Maybe we should
245 just abort if this happens? */
249 ptm = localtime (&secs);
250 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
254 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
257 datetime_str (time_t *tm)
259 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
261 time_t secs = tm ? *tm : time (NULL);
265 /* In case of error, return the empty string. Maybe we should
266 just abort if this happens? */
270 ptm = localtime (&secs);
271 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
272 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
273 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
277 /* The Windows versions of the following two functions are defined in
282 fork_to_background (void)
285 /* Whether we arrange our own version of opt.lfilename here. */
290 opt.lfilename = unique_name (DEFAULT_LOGFILE, 0);
302 /* parent, no error */
303 printf (_("Continuing in background, pid %d.\n"), (int)pid);
305 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
306 exit (0); /* #### should we use _exit()? */
309 /* child: give up the privileges and keep running. */
311 freopen ("/dev/null", "r", stdin);
312 freopen ("/dev/null", "w", stdout);
313 freopen ("/dev/null", "w", stderr);
315 #endif /* not WINDOWS */
317 /* "Touch" FILE, i.e. make its atime and mtime equal to the time
318 specified with TM. */
320 touch (const char *file, time_t tm)
322 #ifdef HAVE_STRUCT_UTIMBUF
323 struct utimbuf times;
324 times.actime = times.modtime = tm;
327 times[0] = times[1] = tm;
330 if (utime (file, ×) == -1)
331 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
334 /* Checks if FILE is a symbolic link, and removes it if it is. Does
335 nothing under MS-Windows. */
337 remove_link (const char *file)
342 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
344 DEBUGP (("Unlinking %s (symlink).\n", file));
347 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
348 file, strerror (errno));
353 /* Does FILENAME exist? This is quite a lousy implementation, since
354 it supplies no error codes -- only a yes-or-no answer. Thus it
355 will return that a file does not exist if, e.g., the directory is
356 unreadable. I don't mind it too much currently, though. The
357 proper way should, of course, be to have a third, error state,
358 other than true/false, but that would introduce uncalled-for
359 additional complexity to the callers. */
361 file_exists_p (const char *filename)
364 return access (filename, F_OK) >= 0;
367 return stat (filename, &buf) >= 0;
371 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
372 Returns 0 on error. */
374 file_non_directory_p (const char *path)
377 /* Use lstat() rather than stat() so that symbolic links pointing to
378 directories can be identified correctly. */
379 if (lstat (path, &buf) != 0)
381 return S_ISDIR (buf.st_mode) ? 0 : 1;
384 /* Return the size of file named by FILENAME, or -1 if it cannot be
385 opened or seeked into. */
387 file_size (const char *filename)
390 /* We use fseek rather than stat to determine the file size because
391 that way we can also verify whether the file is readable.
392 Inspired by the POST patch by Arnaud Wylie. */
393 FILE *fp = fopen (filename, "rb");
396 fseek (fp, 0, SEEK_END);
402 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
403 doesn't exist is found. Return a freshly allocated copy of the
407 unique_name_1 (const char *prefix)
410 int plen = strlen (prefix);
411 char *template = (char *)alloca (plen + 1 + 24);
412 char *template_tail = template + plen;
414 memcpy (template, prefix, plen);
415 *template_tail++ = '.';
418 number_to_string (template_tail, count++);
419 while (file_exists_p (template));
421 return xstrdup (template);
424 /* Return a unique file name, based on FILE.
426 More precisely, if FILE doesn't exist, it is returned unmodified.
427 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
428 file name that doesn't exist is returned.
430 The resulting file is not created, only verified that it didn't
431 exist at the point in time when the function was called.
432 Therefore, where security matters, don't rely that the file created
433 by this function exists until you open it with O_EXCL or
436 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
437 string. Otherwise, it may return FILE if the file doesn't exist
438 (and therefore doesn't need changing). */
441 unique_name (const char *file, int allow_passthrough)
443 /* If the FILE itself doesn't exist, return it without
445 if (!file_exists_p (file))
446 return allow_passthrough ? (char *)file : xstrdup (file);
448 /* Otherwise, find a numeric suffix that results in unused file name
450 return unique_name_1 (file);
453 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
454 are missing, create them first. In case any mkdir() call fails,
455 return its error status. Returns 0 on successful completion.
457 The behaviour of this function should be identical to the behaviour
458 of `mkdir -p' on systems where mkdir supports the `-p' option. */
460 make_directory (const char *directory)
467 /* Make a copy of dir, to be able to write to it. Otherwise, the
468 function is unsafe if called with a read-only char *argument. */
469 STRDUP_ALLOCA (dir, directory);
471 /* If the first character of dir is '/', skip it (and thus enable
472 creation of absolute-pathname directories. */
473 for (i = (*dir == '/'); 1; ++i)
475 for (; dir[i] && dir[i] != '/'; i++)
480 /* Check whether the directory already exists. Allow creation of
481 of intermediate directories to fail, as the initial path components
482 are not necessarily directories! */
483 if (!file_exists_p (dir))
484 ret = mkdir (dir, 0777);
495 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
496 should be a file name.
498 file_merge("/foo/bar", "baz") => "/foo/baz"
499 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
500 file_merge("foo", "bar") => "bar"
502 In other words, it's a simpler and gentler version of uri_merge_1. */
505 file_merge (const char *base, const char *file)
508 const char *cut = (const char *)strrchr (base, '/');
511 return xstrdup (file);
513 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
514 memcpy (result, base, cut - base);
515 result[cut - base] = '/';
516 strcpy (result + (cut - base) + 1, file);
521 static int in_acclist PARAMS ((const char *const *, const char *, int));
523 /* Determine whether a file is acceptable to be followed, according to
524 lists of patterns to accept/reject. */
526 acceptable (const char *s)
530 while (l && s[l] != '/')
537 return (in_acclist ((const char *const *)opt.accepts, s, 1)
538 && !in_acclist ((const char *const *)opt.rejects, s, 1));
540 return in_acclist ((const char *const *)opt.accepts, s, 1);
542 else if (opt.rejects)
543 return !in_acclist ((const char *const *)opt.rejects, s, 1);
547 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
548 `/something', frontcmp() will return 1 only if S2 begins with
549 `/something'. Otherwise, 0 is returned. */
551 frontcmp (const char *s1, const char *s2)
553 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
557 /* Iterate through STRLIST, and return the first element that matches
558 S, through wildcards or front comparison (as appropriate). */
560 proclist (char **strlist, const char *s, enum accd flags)
564 for (x = strlist; *x; x++)
565 if (has_wildcards_p (*x))
567 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
572 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
579 /* Returns whether DIRECTORY is acceptable for download, wrt the
580 include/exclude lists.
582 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
583 and absolute paths may be freely intermixed. */
585 accdir (const char *directory, enum accd flags)
587 /* Remove starting '/'. */
588 if (flags & ALLABS && *directory == '/')
592 if (!proclist (opt.includes, directory, flags))
597 if (proclist (opt.excludes, directory, flags))
603 /* Return non-zero if STRING ends with TAIL. For instance:
605 match_tail ("abc", "bc", 0) -> 1
606 match_tail ("abc", "ab", 0) -> 0
607 match_tail ("abc", "abc", 0) -> 1
609 If FOLD_CASE_P is non-zero, the comparison will be
613 match_tail (const char *string, const char *tail, int fold_case_p)
617 /* We want this to be fast, so we code two loops, one with
618 case-folding, one without. */
622 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
623 if (string[i] != tail[j])
628 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
629 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
633 /* If the tail was exhausted, the match was succesful. */
640 /* Checks whether string S matches each element of ACCEPTS. A list
641 element are matched either with fnmatch() or match_tail(),
642 according to whether the element contains wildcards or not.
644 If the BACKWARD is 0, don't do backward comparison -- just compare
647 in_acclist (const char *const *accepts, const char *s, int backward)
649 for (; *accepts; accepts++)
651 if (has_wildcards_p (*accepts))
653 /* fnmatch returns 0 if the pattern *does* match the
655 if (fnmatch (*accepts, s, 0) == 0)
662 if (match_tail (s, *accepts, 0))
667 if (!strcmp (s, *accepts))
675 /* Return the location of STR's suffix (file extension). Examples:
676 suffix ("foo.bar") -> "bar"
677 suffix ("foo.bar.baz") -> "baz"
678 suffix ("/foo/bar") -> NULL
679 suffix ("/foo.bar/baz") -> NULL */
681 suffix (const char *str)
685 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
689 return (char *)str + i;
694 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
698 has_wildcards_p (const char *s)
701 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
706 /* Return non-zero if FNAME ends with a typical HTML suffix. The
707 following (case-insensitive) suffixes are presumed to be HTML files:
711 ?html (`?' matches one character)
713 #### CAVEAT. This is not necessarily a good indication that FNAME
714 refers to a file that contains HTML! */
716 has_html_suffix_p (const char *fname)
720 if ((suf = suffix (fname)) == NULL)
722 if (!strcasecmp (suf, "html"))
724 if (!strcasecmp (suf, "htm"))
726 if (suf[0] && !strcasecmp (suf + 1, "html"))
731 /* Read a line from FP and return the pointer to freshly allocated
732 storage. The storage space is obtained through malloc() and should
733 be freed with free() when it is no longer needed.
735 The length of the line is not limited, except by available memory.
736 The newline character at the end of line is retained. The line is
737 terminated with a zero character.
739 After end-of-file is encountered without anything being read, NULL
740 is returned. NULL is also returned on error. To distinguish
741 between these two cases, use the stdio function ferror(). */
744 read_whole_line (FILE *fp)
748 char *line = (char *)xmalloc (bufsize);
750 while (fgets (line + length, bufsize - length, fp))
752 length += strlen (line + length);
754 /* Possible for example when reading from a binary file where
755 a line begins with \0. */
758 if (line[length - 1] == '\n')
761 /* fgets() guarantees to read the whole line, or to use up the
762 space we've given it. We can double the buffer
765 line = xrealloc (line, bufsize);
767 if (length == 0 || ferror (fp))
772 if (length + 1 < bufsize)
773 /* Relieve the memory from our exponential greediness. We say
774 `length + 1' because the terminating \0 is not included in
775 LENGTH. We don't need to zero-terminate the string ourselves,
776 though, because fgets() does that. */
777 line = xrealloc (line, length + 1);
781 /* Read FILE into memory. A pointer to `struct file_memory' are
782 returned; use struct element `content' to access file contents, and
783 the element `length' to know the file length. `content' is *not*
784 zero-terminated, and you should *not* read or write beyond the [0,
785 length) range of characters.
787 After you are done with the file contents, call read_file_free to
790 Depending on the operating system and the type of file that is
791 being read, read_file() either mmap's the file into memory, or
792 reads the file into the core using read().
794 If file is named "-", fileno(stdin) is used for reading instead.
795 If you want to read from a real file named "-", use "./-" instead. */
798 read_file (const char *file)
801 struct file_memory *fm;
803 int inhibit_close = 0;
805 /* Some magic in the finest tradition of Perl and its kin: if FILE
806 is "-", just use stdin. */
811 /* Note that we don't inhibit mmap() in this case. If stdin is
812 redirected from a regular file, mmap() will still work. */
815 fd = open (file, O_RDONLY);
818 fm = xnew (struct file_memory);
823 if (fstat (fd, &buf) < 0)
825 fm->length = buf.st_size;
826 /* NOTE: As far as I know, the callers of this function never
827 modify the file text. Relying on this would enable us to
828 specify PROT_READ and MAP_SHARED for a marginal gain in
829 efficiency, but at some cost to generality. */
830 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
832 if (fm->content == (char *)MAP_FAILED)
842 /* The most common reason why mmap() fails is that FD does not point
843 to a plain file. However, it's also possible that mmap() doesn't
844 work for a particular type of file. Therefore, whenever mmap()
845 fails, we just fall back to the regular method. */
846 #endif /* HAVE_MMAP */
849 size = 512; /* number of bytes fm->contents can
850 hold at any given time. */
851 fm->content = xmalloc (size);
855 if (fm->length > size / 2)
857 /* #### I'm not sure whether the whole exponential-growth
858 thing makes sense with kernel read. On Linux at least,
859 read() refuses to read more than 4K from a file at a
860 single chunk anyway. But other Unixes might optimize it
861 better, and it doesn't *hurt* anything, so I'm leaving
864 /* Normally, we grow SIZE exponentially to make the number
865 of calls to read() and realloc() logarithmic in relation
866 to file size. However, read() can read an amount of data
867 smaller than requested, and it would be unreasonable to
868 double SIZE every time *something* was read. Therefore,
869 we double SIZE only when the length exceeds half of the
870 entire allocated size. */
872 fm->content = xrealloc (fm->content, size);
874 nread = read (fd, fm->content + fm->length, size - fm->length);
876 /* Successful read. */
887 if (size > fm->length && fm->length != 0)
888 /* Due to exponential growth of fm->content, the allocated region
889 might be much larger than what is actually needed. */
890 fm->content = xrealloc (fm->content, fm->length);
902 /* Release the resources held by FM. Specifically, this calls
903 munmap() or xfree() on fm->content, depending whether mmap or
904 malloc/read were used to read in the file. It also frees the
905 memory needed to hold the FM structure itself. */
908 read_file_free (struct file_memory *fm)
913 munmap (fm->content, fm->length);
923 /* Free the pointers in a NULL-terminated vector of pointers, then
924 free the pointer itself. */
926 free_vec (char **vec)
937 /* Append vector V2 to vector V1. The function frees V2 and
938 reallocates V1 (thus you may not use the contents of neither
939 pointer after the call). If V1 is NULL, V2 is returned. */
941 merge_vecs (char **v1, char **v2)
951 /* To avoid j == 0 */
956 for (i = 0; v1[i]; i++);
958 for (j = 0; v2[j]; j++);
960 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
961 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
966 /* A set of simple-minded routines to store strings in a linked list.
967 This used to also be used for searching, but now we have hash
970 /* It's a shame that these simple things like linked lists and hash
971 tables (see hash.c) need to be implemented over and over again. It
972 would be nice to be able to use the routines from glib -- see
973 www.gtk.org for details. However, that would make Wget depend on
974 glib, and I want to avoid dependencies to external libraries for
975 reasons of convenience and portability (I suspect Wget is more
976 portable than anything ever written for Gnome). */
978 /* Append an element to the list. If the list has a huge number of
979 elements, this can get slow because it has to find the list's
980 ending. If you think you have to call slist_append in a loop,
981 think about calling slist_prepend() followed by slist_nreverse(). */
984 slist_append (slist *l, const char *s)
986 slist *newel = xnew (slist);
989 newel->string = xstrdup (s);
994 /* Find the last element. */
1001 /* Prepend S to the list. Unlike slist_append(), this is O(1). */
1004 slist_prepend (slist *l, const char *s)
1006 slist *newel = xnew (slist);
1007 newel->string = xstrdup (s);
1012 /* Destructively reverse L. */
1015 slist_nreverse (slist *l)
1020 slist *next = l->next;
1028 /* Is there a specific entry in the list? */
1030 slist_contains (slist *l, const char *s)
1032 for (; l; l = l->next)
1033 if (!strcmp (l->string, s))
1038 /* Free the whole slist. */
1040 slist_free (slist *l)
1051 /* Sometimes it's useful to create "sets" of strings, i.e. special
1052 hash tables where you want to store strings as keys and merely
1053 query for their existence. Here is a set of utility routines that
1054 makes that transparent. */
1057 string_set_add (struct hash_table *ht, const char *s)
1059 /* First check whether the set element already exists. If it does,
1060 do nothing so that we don't have to free() the old element and
1061 then strdup() a new one. */
1062 if (hash_table_contains (ht, s))
1065 /* We use "1" as value. It provides us a useful and clear arbitrary
1066 value, and it consumes no memory -- the pointers to the same
1067 string "1" will be shared by all the key-value pairs in all `set'
1069 hash_table_put (ht, xstrdup (s), "1");
1072 /* Synonym for hash_table_contains... */
1075 string_set_contains (struct hash_table *ht, const char *s)
1077 return hash_table_contains (ht, s);
1081 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1088 string_set_free (struct hash_table *ht)
1090 hash_table_map (ht, string_set_free_mapper, NULL);
1091 hash_table_destroy (ht);
1095 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1102 /* Another utility function: call free() on all keys and values of HT. */
1105 free_keys_and_values (struct hash_table *ht)
1107 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1111 /* Engine for legible and legible_large_int; add thousand separators
1112 to numbers printed in strings. */
1115 legible_1 (const char *repr)
1117 static char outbuf[48];
1122 /* Reset the pointers. */
1126 /* Ignore the sign for the purpose of adding thousand
1133 /* How many digits before the first separator? */
1134 mod = strlen (inptr) % 3;
1136 for (i = 0; i < mod; i++)
1137 *outptr++ = inptr[i];
1138 /* Now insert the rest of them, putting separator before every
1140 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1142 if (i % 3 == 0 && i1 != 0)
1144 *outptr++ = inptr[i1];
1146 /* Zero-terminate the string. */
1151 /* Legible -- return a static pointer to the legibly printed long. */
1157 /* Print the number into the buffer. */
1158 number_to_string (inbuf, l);
1159 return legible_1 (inbuf);
1162 /* Write a string representation of LARGE_INT NUMBER into the provided
1163 buffer. The buffer should be able to accept 24 characters,
1164 including the terminating zero.
1166 It would be dangerous to use sprintf, because the code wouldn't
1167 work on a machine with gcc-provided long long support, but without
1168 libc support for "%lld". However, such platforms will typically
1169 not have snprintf and will use our version, which does support
1170 "%lld" where long longs are available. */
1173 large_int_to_string (char *buffer, LARGE_INT number)
1175 snprintf (buffer, 24, LARGE_INT_FMT, number);
1178 /* The same as legible(), but works on LARGE_INT. */
1181 legible_large_int (LARGE_INT l)
1184 large_int_to_string (inbuf, l);
1185 return legible_1 (inbuf);
1188 /* Count the digits in a (long) integer. */
1190 numdigit (long number)
1198 while ((number /= 10) > 0)
1203 /* Attempt to calculate INT_MAX on machines that don't bother to
1209 # define INT_MAX ((int) ~((unsigned)1 << CHAR_BIT * sizeof (int) - 1))
1212 #define ONE_DIGIT(figure) *p++ = n / (figure) + '0'
1213 #define ONE_DIGIT_ADVANCE(figure) (ONE_DIGIT (figure), n %= (figure))
1215 #define DIGITS_1(figure) ONE_DIGIT (figure)
1216 #define DIGITS_2(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_1 ((figure) / 10)
1217 #define DIGITS_3(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_2 ((figure) / 10)
1218 #define DIGITS_4(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_3 ((figure) / 10)
1219 #define DIGITS_5(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_4 ((figure) / 10)
1220 #define DIGITS_6(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_5 ((figure) / 10)
1221 #define DIGITS_7(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_6 ((figure) / 10)
1222 #define DIGITS_8(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_7 ((figure) / 10)
1223 #define DIGITS_9(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_8 ((figure) / 10)
1224 #define DIGITS_10(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_9 ((figure) / 10)
1226 /* DIGITS_<11-20> are only used on machines with 64-bit longs. */
1228 #define DIGITS_11(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_10 ((figure) / 10)
1229 #define DIGITS_12(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_11 ((figure) / 10)
1230 #define DIGITS_13(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_12 ((figure) / 10)
1231 #define DIGITS_14(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_13 ((figure) / 10)
1232 #define DIGITS_15(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_14 ((figure) / 10)
1233 #define DIGITS_16(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_15 ((figure) / 10)
1234 #define DIGITS_17(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_16 ((figure) / 10)
1235 #define DIGITS_18(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_17 ((figure) / 10)
1236 #define DIGITS_19(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_18 ((figure) / 10)
1238 /* Print NUMBER to BUFFER in base 10. This should be completely
1239 equivalent to `sprintf(buffer, "%ld", number)', only much faster.
1241 The speedup may make a difference in programs that frequently
1242 convert numbers to strings. Some implementations of sprintf,
1243 particularly the one in GNU libc, have been known to be extremely
1244 slow compared to this function.
1246 Return the pointer to the location where the terminating zero was
1247 printed. (Equivalent to calling buffer+strlen(buffer) after the
1250 BUFFER should be big enough to accept as many bytes as you expect
1251 the number to take up. On machines with 64-bit longs the maximum
1252 needed size is 24 bytes. That includes the digits needed for the
1253 largest 64-bit number, the `-' sign in case it's negative, and the
1254 terminating '\0'. */
1257 number_to_string (char *buffer, long number)
1262 #if (SIZEOF_LONG != 4) && (SIZEOF_LONG != 8)
1263 /* We are running in a strange or misconfigured environment. Let
1264 sprintf cope with it. */
1265 sprintf (buffer, "%ld", n);
1266 p += strlen (buffer);
1267 #else /* (SIZEOF_LONG == 4) || (SIZEOF_LONG == 8) */
1273 /* We cannot print a '-' and assign -n to n because -n would
1274 overflow. Let sprintf deal with this border case. */
1275 sprintf (buffer, "%ld", n);
1276 p += strlen (buffer);
1284 if (n < 10) { DIGITS_1 (1); }
1285 else if (n < 100) { DIGITS_2 (10); }
1286 else if (n < 1000) { DIGITS_3 (100); }
1287 else if (n < 10000) { DIGITS_4 (1000); }
1288 else if (n < 100000) { DIGITS_5 (10000); }
1289 else if (n < 1000000) { DIGITS_6 (100000); }
1290 else if (n < 10000000) { DIGITS_7 (1000000); }
1291 else if (n < 100000000) { DIGITS_8 (10000000); }
1292 else if (n < 1000000000) { DIGITS_9 (100000000); }
1293 #if SIZEOF_LONG == 4
1294 /* ``if (1)'' serves only to preserve editor indentation. */
1295 else if (1) { DIGITS_10 (1000000000); }
1296 #else /* SIZEOF_LONG != 4 */
1297 else if (n < 10000000000L) { DIGITS_10 (1000000000L); }
1298 else if (n < 100000000000L) { DIGITS_11 (10000000000L); }
1299 else if (n < 1000000000000L) { DIGITS_12 (100000000000L); }
1300 else if (n < 10000000000000L) { DIGITS_13 (1000000000000L); }
1301 else if (n < 100000000000000L) { DIGITS_14 (10000000000000L); }
1302 else if (n < 1000000000000000L) { DIGITS_15 (100000000000000L); }
1303 else if (n < 10000000000000000L) { DIGITS_16 (1000000000000000L); }
1304 else if (n < 100000000000000000L) { DIGITS_17 (10000000000000000L); }
1305 else if (n < 1000000000000000000L) { DIGITS_18 (100000000000000000L); }
1306 else { DIGITS_19 (1000000000000000000L); }
1307 #endif /* SIZEOF_LONG != 4 */
1310 #endif /* (SIZEOF_LONG == 4) || (SIZEOF_LONG == 8) */
1316 #undef ONE_DIGIT_ADVANCE
1338 /* Support for timers. */
1340 #undef TIMER_WINDOWS
1341 #undef TIMER_GETTIMEOFDAY
1344 /* Depending on the OS and availability of gettimeofday(), one and
1345 only one of the above constants will be defined. Virtually all
1346 modern Unix systems will define TIMER_GETTIMEOFDAY; Windows will
1347 use TIMER_WINDOWS. TIMER_TIME is a catch-all method for
1348 non-Windows systems without gettimeofday.
1350 #### Perhaps we should also support ftime(), which exists on old
1351 BSD 4.2-influenced systems? (It also existed under MS DOS Borland
1352 C, if memory serves me.) */
1355 # define TIMER_WINDOWS
1356 #else /* not WINDOWS */
1357 # ifdef HAVE_GETTIMEOFDAY
1358 # define TIMER_GETTIMEOFDAY
1362 #endif /* not WINDOWS */
1364 #ifdef TIMER_GETTIMEOFDAY
1365 typedef struct timeval wget_sys_time;
1369 typedef time_t wget_sys_time;
1372 #ifdef TIMER_WINDOWS
1373 typedef ULARGE_INTEGER wget_sys_time;
1377 /* Whether the start time has been initialized. */
1380 /* The starting point in time which, subtracted from the current
1381 time, yields elapsed time. */
1382 wget_sys_time start;
1384 /* The most recent elapsed time, calculated by wtimer_elapsed().
1385 Measured in milliseconds. */
1386 double elapsed_last;
1388 /* Approximately, the time elapsed between the true start of the
1389 measurement and the time represented by START. */
1390 double elapsed_pre_start;
1393 /* Allocate a timer. Calling wtimer_read on the timer will return
1394 zero. It is not legal to call wtimer_update with a freshly
1395 allocated timer -- use wtimer_reset first. */
1398 wtimer_allocate (void)
1400 struct wget_timer *wt = xnew (struct wget_timer);
1405 /* Allocate a new timer and reset it. Return the new timer. */
1410 struct wget_timer *wt = wtimer_allocate ();
1415 /* Free the resources associated with the timer. Its further use is
1419 wtimer_delete (struct wget_timer *wt)
1424 /* Store system time to WST. */
1427 wtimer_sys_set (wget_sys_time *wst)
1429 #ifdef TIMER_GETTIMEOFDAY
1430 gettimeofday (wst, NULL);
1437 #ifdef TIMER_WINDOWS
1438 /* We use GetSystemTime to get the elapsed time. MSDN warns that
1439 system clock adjustments can skew the output of GetSystemTime
1440 when used as a timer and gives preference to GetTickCount and
1441 high-resolution timers. But GetTickCount can overflow, and hires
1442 timers are typically used for profiling, not for regular time
1443 measurement. Since we handle clock skew anyway, we just use
1447 GetSystemTime (&st);
1449 /* As recommended by MSDN, we convert SYSTEMTIME to FILETIME, copy
1450 FILETIME to ULARGE_INTEGER, and use regular 64-bit integer
1451 arithmetic on that. */
1452 SystemTimeToFileTime (&st, &ft);
1453 wst->HighPart = ft.dwHighDateTime;
1454 wst->LowPart = ft.dwLowDateTime;
1458 /* Reset timer WT. This establishes the starting point from which
1459 wtimer_elapsed() will return the number of elapsed milliseconds.
1460 It is allowed to reset a previously used timer.
1462 If a non-zero value is used as START, the timer's values will be
1466 wtimer_reset (struct wget_timer *wt)
1468 /* Set the start time to the current time. */
1469 wtimer_sys_set (&wt->start);
1470 wt->elapsed_last = 0;
1471 wt->elapsed_pre_start = 0;
1472 wt->initialized = 1;
1476 wtimer_sys_diff (wget_sys_time *wst1, wget_sys_time *wst2)
1478 #ifdef TIMER_GETTIMEOFDAY
1479 return ((double)(wst1->tv_sec - wst2->tv_sec) * 1000
1480 + (double)(wst1->tv_usec - wst2->tv_usec) / 1000);
1484 return 1000 * (*wst1 - *wst2);
1488 /* VC++ 6 doesn't support direct cast of uint64 to double. To work
1489 around this, we subtract, then convert to signed, then finally to
1491 return (double)(signed __int64)(wst1->QuadPart - wst2->QuadPart) / 10000;
1495 /* Update the timer's elapsed interval. This function causes the
1496 timer to call gettimeofday (or time(), etc.) to update its idea of
1497 current time. To get the elapsed interval in milliseconds, use
1500 This function handles clock skew, i.e. time that moves backwards is
1504 wtimer_update (struct wget_timer *wt)
1509 assert (wt->initialized != 0);
1511 wtimer_sys_set (&now);
1512 elapsed = wt->elapsed_pre_start + wtimer_sys_diff (&now, &wt->start);
1514 /* Ideally we'd just return the difference between NOW and
1515 wt->start. However, the system timer can be set back, and we
1516 could return a value smaller than when we were last called, even
1517 a negative value. Both of these would confuse the callers, which
1518 expect us to return monotonically nondecreasing values.
1520 Therefore: if ELAPSED is smaller than its previous known value,
1521 we reset wt->start to the current time and effectively start
1522 measuring from this point. But since we don't want the elapsed
1523 value to start from zero, we set elapsed_pre_start to the last
1524 elapsed time and increment all future calculations by that
1527 if (elapsed < wt->elapsed_last)
1530 wt->elapsed_pre_start = wt->elapsed_last;
1531 elapsed = wt->elapsed_last;
1534 wt->elapsed_last = elapsed;
1537 /* Return the elapsed time in milliseconds between the last call to
1538 wtimer_reset and the last call to wtimer_update.
1540 A typical use of the timer interface would be:
1542 struct wtimer *timer = wtimer_new ();
1543 ... do something that takes a while ...
1545 double msecs = wtimer_read (); */
1548 wtimer_read (const struct wget_timer *wt)
1550 return wt->elapsed_last;
1553 /* Return the assessed granularity of the timer implementation, in
1554 milliseconds. This is used by code that tries to substitute a
1555 better value for timers that have returned zero. */
1558 wtimer_granularity (void)
1560 #ifdef TIMER_GETTIMEOFDAY
1561 /* Granularity of gettimeofday varies wildly between architectures.
1562 However, it appears that on modern machines it tends to be better
1563 than 1ms. Assume 100 usecs. (Perhaps the configure process
1564 could actually measure this?) */
1572 #ifdef TIMER_WINDOWS
1573 /* According to MSDN, GetSystemTime returns a broken-down time
1574 structure the smallest member of which are milliseconds. */
1579 /* This should probably be at a better place, but it doesn't really
1580 fit into html-parse.c. */
1582 /* The function returns the pointer to the malloc-ed quoted version of
1583 string s. It will recognize and quote numeric and special graphic
1584 entities, as per RFC1866:
1592 No other entities are recognized or replaced. */
1594 html_quote_string (const char *s)
1600 /* Pass through the string, and count the new size. */
1601 for (i = 0; *s; s++, i++)
1604 i += 4; /* `amp;' */
1605 else if (*s == '<' || *s == '>')
1606 i += 3; /* `lt;' and `gt;' */
1607 else if (*s == '\"')
1608 i += 5; /* `quot;' */
1612 res = (char *)xmalloc (i + 1);
1614 for (p = res; *s; s++)
1627 *p++ = (*s == '<' ? 'l' : 'g');
1654 /* Determine the width of the terminal we're running on. If that's
1655 not possible, return 0. */
1658 determine_screen_width (void)
1660 /* If there's a way to get the terminal size using POSIX
1661 tcgetattr(), somebody please tell me. */
1664 #else /* TIOCGWINSZ */
1668 if (opt.lfilename != NULL)
1671 fd = fileno (stderr);
1672 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1673 return 0; /* most likely ENOTTY */
1676 #endif /* TIOCGWINSZ */
1679 /* Return a random number between 0 and MAX-1, inclusive.
1681 If MAX is greater than the value of RAND_MAX+1 on the system, the
1682 returned value will be in the range [0, RAND_MAX]. This may be
1683 fixed in a future release.
1685 The random number generator is seeded automatically the first time
1688 This uses rand() for portability. It has been suggested that
1689 random() offers better randomness, but this is not required for
1690 Wget, so I chose to go for simplicity and use rand
1693 DO NOT use this for cryptographic purposes. It is only meant to be
1694 used in situations where quality of the random numbers returned
1695 doesn't really matter. */
1698 random_number (int max)
1706 srand (time (NULL));
1711 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1712 and enforce that assumption by masking other bits. */
1714 # define RAND_MAX 32767
1718 /* This is equivalent to rand() % max, but uses the high-order bits
1719 for better randomness on architecture where rand() is implemented
1720 using a simple congruential generator. */
1722 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1723 return (int)bounded;
1726 /* Return a random uniformly distributed floating point number in the
1727 [0, 1) range. The precision of returned numbers is 9 digits.
1729 Modify this to use erand48() where available! */
1734 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1735 sure it's greater than 1000. */
1736 int rnd1 = random_number (1000);
1737 int rnd2 = random_number (1000);
1738 int rnd3 = random_number (1000);
1739 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1743 /* A debugging function for checking whether an MD5 library works. */
1745 #include "gen-md5.h"
1748 debug_test_md5 (char *buf)
1750 unsigned char raw[16];
1751 static char res[33];
1755 ALLOCA_MD5_CONTEXT (ctx);
1758 gen_md5_update ((unsigned char *)buf, strlen (buf), ctx);
1759 gen_md5_finish (ctx, raw);
1766 *p2++ = XNUM_TO_digit (*p1 >> 4);
1767 *p2++ = XNUM_TO_digit (*p1 & 0xf);
1776 /* Implementation of run_with_timeout, a generic timeout-forcing
1777 routine for systems with Unix-like signal handling. */
1779 #ifdef USE_SIGNAL_TIMEOUT
1780 # ifdef HAVE_SIGSETJMP
1781 # define SETJMP(env) sigsetjmp (env, 1)
1783 static sigjmp_buf run_with_timeout_env;
1786 abort_run_with_timeout (int sig)
1788 assert (sig == SIGALRM);
1789 siglongjmp (run_with_timeout_env, -1);
1791 # else /* not HAVE_SIGSETJMP */
1792 # define SETJMP(env) setjmp (env)
1794 static jmp_buf run_with_timeout_env;
1797 abort_run_with_timeout (int sig)
1799 assert (sig == SIGALRM);
1800 /* We don't have siglongjmp to preserve the set of blocked signals;
1801 if we longjumped out of the handler at this point, SIGALRM would
1802 remain blocked. We must unblock it manually. */
1803 int mask = siggetmask ();
1804 mask &= ~sigmask (SIGALRM);
1807 /* Now it's safe to longjump. */
1808 longjmp (run_with_timeout_env, -1);
1810 # endif /* not HAVE_SIGSETJMP */
1812 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1813 setitimer where available, alarm otherwise.
1815 TIMEOUT should be non-zero. If the timeout value is so small that
1816 it would be rounded to zero, it is rounded to the least legal value
1817 instead (1us for setitimer, 1s for alarm). That ensures that
1818 SIGALRM will be delivered in all cases. */
1821 alarm_set (double timeout)
1824 /* Use the modern itimer interface. */
1825 struct itimerval itv;
1827 itv.it_value.tv_sec = (long) timeout;
1828 itv.it_value.tv_usec = 1000000L * (timeout - (long)timeout);
1829 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1830 /* Ensure that we wait for at least the minimum interval.
1831 Specifying zero would mean "wait forever". */
1832 itv.it_value.tv_usec = 1;
1833 setitimer (ITIMER_REAL, &itv, NULL);
1834 #else /* not ITIMER_REAL */
1835 /* Use the old alarm() interface. */
1836 int secs = (int) timeout;
1838 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1839 because alarm(0) means "never deliver the alarm", i.e. "wait
1840 forever", which is not what someone who specifies a 0.5s
1841 timeout would expect. */
1844 #endif /* not ITIMER_REAL */
1847 /* Cancel the alarm set with alarm_set. */
1853 struct itimerval disable;
1855 setitimer (ITIMER_REAL, &disable, NULL);
1856 #else /* not ITIMER_REAL */
1858 #endif /* not ITIMER_REAL */
1861 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1862 seconds. Returns non-zero if the function was interrupted with a
1863 timeout, zero otherwise.
1865 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1866 using setitimer() or alarm(). The timeout is enforced by
1867 longjumping out of the SIGALRM handler. This has several
1868 advantages compared to the traditional approach of relying on
1869 signals causing system calls to exit with EINTR:
1871 * The callback function is *forcibly* interrupted after the
1872 timeout expires, (almost) regardless of what it was doing and
1873 whether it was in a syscall. For example, a calculation that
1874 takes a long time is interrupted as reliably as an IO
1877 * It works with both SYSV and BSD signals because it doesn't
1878 depend on the default setting of SA_RESTART.
1880 * It doesn't special handler setup beyond a simple call to
1881 signal(). (It does use sigsetjmp/siglongjmp, but they're
1884 The only downside is that, if FUN allocates internal resources that
1885 are normally freed prior to exit from the functions, they will be
1886 lost in case of timeout. */
1889 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1899 signal (SIGALRM, abort_run_with_timeout);
1900 if (SETJMP (run_with_timeout_env) != 0)
1902 /* Longjumped out of FUN with a timeout. */
1903 signal (SIGALRM, SIG_DFL);
1906 alarm_set (timeout);
1909 /* Preserve errno in case alarm() or signal() modifies it. */
1910 saved_errno = errno;
1912 signal (SIGALRM, SIG_DFL);
1913 errno = saved_errno;
1918 #else /* not USE_SIGNAL_TIMEOUT */
1921 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
1922 define it under Windows, because Windows has its own version of
1923 run_with_timeout that uses threads. */
1926 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
1931 #endif /* not WINDOWS */
1932 #endif /* not USE_SIGNAL_TIMEOUT */
1936 /* Sleep the specified amount of seconds. On machines without
1937 nanosleep(), this may sleep shorter if interrupted by signals. */
1940 xsleep (double seconds)
1942 #ifdef HAVE_NANOSLEEP
1943 /* nanosleep is the preferred interface because it offers high
1944 accuracy and, more importantly, because it allows us to reliably
1945 restart after having been interrupted by a signal such as
1947 struct timespec sleep, remaining;
1948 sleep.tv_sec = (long) seconds;
1949 sleep.tv_nsec = 1000000000L * (seconds - (long) seconds);
1950 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
1951 /* If nanosleep has been interrupted by a signal, adjust the
1952 sleeping period and return to sleep. */
1954 #else /* not HAVE_NANOSLEEP */
1956 /* If usleep is available, use it in preference to select. */
1959 /* usleep apparently accepts unsigned long, which means it can't
1960 sleep longer than ~70 min (35min if signed). If the period
1961 is larger than what usleep can safely handle, use sleep
1962 first, then add usleep for subsecond accuracy. */
1964 seconds -= (long) seconds;
1966 usleep (seconds * 1000000L);
1967 #else /* not HAVE_USLEEP */
1969 struct timeval sleep;
1970 sleep.tv_sec = (long) seconds;
1971 sleep.tv_usec = 1000000L * (seconds - (long) seconds);
1972 select (0, NULL, NULL, NULL, &sleep);
1973 /* If select returns -1 and errno is EINTR, it means we were
1974 interrupted by a signal. But without knowing how long we've
1975 actually slept, we can't return to sleep. Using gettimeofday to
1976 track sleeps is slow and unreliable due to clock skew. */
1977 #else /* not HAVE_SELECT */
1979 #endif /* not HAVE_SELECT */
1980 #endif /* not HAVE_USLEEP */
1981 #endif /* not HAVE_NANOSLEEP */
1984 #endif /* not WINDOWS */