1 /* Various functions of utilitarian nature.
2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001
3 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
27 #else /* not HAVE_STRING_H */
29 #endif /* not HAVE_STRING_H */
30 #include <sys/types.h>
35 # include <sys/mman.h>
44 #ifdef HAVE_SYS_UTIME_H
45 # include <sys/utime.h>
49 # include <libc.h> /* for access() */
54 /* For TIOCGWINSZ and friends: */
55 #ifdef HAVE_SYS_IOCTL_H
56 # include <sys/ioctl.h>
71 /* This section implements several wrappers around the basic
72 allocation routines. This is done for two reasons: first, so that
73 the callers of these functions need not consistently check for
74 errors. If there is not enough virtual memory for running Wget,
75 something is seriously wrong, and Wget exits with an appropriate
78 The second reason why these are useful is that, if DEBUG_MALLOC is
79 defined, they also provide a handy (if crude) malloc debugging
80 interface that checks memory leaks. */
82 /* Croak the fatal memory error and bail out with non-zero exit
85 memfatal (const char *what)
87 /* Make sure we don't try to store part of the log line, and thus
89 log_set_save_context (0);
90 logprintf (LOG_ALWAYS, _("%s: %s: Not enough memory.\n"), exec_name, what);
94 /* These functions end with _real because they need to be
95 distinguished from the debugging functions, and from the macros.
98 If memory debugging is not turned on, wget.h defines these:
100 #define xmalloc xmalloc_real
101 #define xrealloc xrealloc_real
102 #define xstrdup xstrdup_real
105 In case of memory debugging, the definitions are a bit more
106 complex, because we want to provide more information, *and* we want
107 to call the debugging code. (The former is the reason why xmalloc
108 and friends need to be macros in the first place.) Then it looks
111 #define xmalloc(a) xmalloc_debug (a, __FILE__, __LINE__)
112 #define xfree(a) xfree_debug (a, __FILE__, __LINE__)
113 #define xrealloc(a, b) xrealloc_debug (a, b, __FILE__, __LINE__)
114 #define xstrdup(a) xstrdup_debug (a, __FILE__, __LINE__)
116 Each of the *_debug function does its magic and calls the real one. */
119 # define STATIC_IF_DEBUG static
121 # define STATIC_IF_DEBUG
124 STATIC_IF_DEBUG void *
125 xmalloc_real (size_t size)
127 void *ptr = malloc (size);
133 STATIC_IF_DEBUG void *
134 xrealloc_real (void *ptr, size_t newsize)
138 /* Not all Un*xes have the feature of realloc() that calling it with
139 a NULL-pointer is the same as malloc(), but it is easy to
142 newptr = realloc (ptr, newsize);
144 newptr = malloc (newsize);
146 memfatal ("realloc");
150 STATIC_IF_DEBUG char *
151 xstrdup_real (const char *s)
157 copy = malloc (l + 1);
160 memcpy (copy, s, l + 1);
161 #else /* HAVE_STRDUP */
165 #endif /* HAVE_STRDUP */
172 /* Crude home-grown routines for debugging some malloc-related
175 * Counting the number of malloc and free invocations, and reporting
176 the "balance", i.e. how many times more malloc was called than it
177 was the case with free.
179 * Making malloc store its entry into a simple array and free remove
180 stuff from that array. At the end, print the pointers which have
181 not been freed, along with the source file and the line number.
182 This also has the side-effect of detecting freeing memory that
185 Note that this kind of memory leak checking strongly depends on
186 every malloc() being followed by a free(), even if the program is
187 about to finish. Wget is careful to free the data structure it
188 allocated in init.c. */
190 static int malloc_count, free_count;
196 } malloc_debug[100000];
198 /* Both register_ptr and unregister_ptr take O(n) operations to run,
199 which can be a real problem. It would be nice to use a hash table
200 for malloc_debug, but the functions in hash.c are not suitable
201 because they can call malloc() themselves. Maybe it would work if
202 the hash table were preallocated to a huge size, and if we set the
203 rehash threshold to 1.0. */
205 /* Register PTR in malloc_debug. Abort if this is not possible
206 (presumably due to the number of current allocations exceeding the
207 size of malloc_debug.) */
210 register_ptr (void *ptr, const char *file, int line)
213 for (i = 0; i < ARRAY_SIZE (malloc_debug); i++)
214 if (malloc_debug[i].ptr == NULL)
216 malloc_debug[i].ptr = ptr;
217 malloc_debug[i].file = file;
218 malloc_debug[i].line = line;
224 /* Unregister PTR from malloc_debug. Abort if PTR is not present in
225 malloc_debug. (This catches calling free() with a bogus pointer.) */
228 unregister_ptr (void *ptr)
231 for (i = 0; i < ARRAY_SIZE (malloc_debug); i++)
232 if (malloc_debug[i].ptr == ptr)
234 malloc_debug[i].ptr = NULL;
240 /* Print the malloc debug stats that can be gathered from the above
241 information. Currently this is the count of mallocs, frees, the
242 difference between the two, and the dump of the contents of
243 malloc_debug. The last part are the memory leaks. */
246 print_malloc_debug_stats (void)
249 printf ("\nMalloc: %d\nFree: %d\nBalance: %d\n\n",
250 malloc_count, free_count, malloc_count - free_count);
251 for (i = 0; i < ARRAY_SIZE (malloc_debug); i++)
252 if (malloc_debug[i].ptr != NULL)
253 printf ("0x%08ld: %s:%d\n", (long)malloc_debug[i].ptr,
254 malloc_debug[i].file, malloc_debug[i].line);
258 xmalloc_debug (size_t size, const char *source_file, int source_line)
260 void *ptr = xmalloc_real (size);
262 register_ptr (ptr, source_file, source_line);
267 xfree_debug (void *ptr, const char *source_file, int source_line)
269 assert (ptr != NULL);
271 unregister_ptr (ptr);
276 xrealloc_debug (void *ptr, size_t newsize, const char *source_file, int source_line)
278 void *newptr = xrealloc_real (ptr, newsize);
282 register_ptr (newptr, source_file, source_line);
284 else if (newptr != ptr)
286 unregister_ptr (ptr);
287 register_ptr (newptr, source_file, source_line);
293 xstrdup_debug (const char *s, const char *source_file, int source_line)
295 char *copy = xstrdup_real (s);
297 register_ptr (copy, source_file, source_line);
301 #endif /* DEBUG_MALLOC */
303 /* Utility function: like xstrdup(), but also lowercases S. */
306 xstrdup_lower (const char *s)
308 char *copy = xstrdup (s);
315 /* Return a count of how many times CHR occurs in STRING. */
318 count_char (const char *string, char chr)
322 for (p = string; *p; p++)
328 /* Copy the string formed by two pointers (one on the beginning, other
329 on the char after the last char) to a new, malloc-ed location.
332 strdupdelim (const char *beg, const char *end)
334 char *res = (char *)xmalloc (end - beg + 1);
335 memcpy (res, beg, end - beg);
336 res[end - beg] = '\0';
340 /* Parse a string containing comma-separated elements, and return a
341 vector of char pointers with the elements. Spaces following the
342 commas are ignored. */
344 sepstring (const char *s)
358 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
359 res[i] = strdupdelim (p, s);
362 /* Skip the blanks following the ','. */
370 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
371 res[i] = strdupdelim (p, s);
376 /* Return pointer to a static char[] buffer in which zero-terminated
377 string-representation of TM (in form hh:mm:ss) is printed.
379 If TM is non-NULL, the current time-in-seconds will be stored
382 (#### This is misleading: one would expect TM would be used instead
383 of the current time in that case. This design was probably
384 influenced by the design time(2), and should be changed at some
385 points. No callers use non-NULL TM anyway.) */
388 time_str (time_t *tm)
390 static char output[15];
392 time_t secs = time (tm);
396 /* In case of error, return the empty string. Maybe we should
397 just abort if this happens? */
401 ptm = localtime (&secs);
402 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
406 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
409 datetime_str (time_t *tm)
411 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
413 time_t secs = time (tm);
417 /* In case of error, return the empty string. Maybe we should
418 just abort if this happens? */
422 ptm = localtime (&secs);
423 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
424 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
425 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
429 /* The Windows versions of the following two functions are defined in
434 fork_to_background (void)
437 /* Whether we arrange our own version of opt.lfilename here. */
442 opt.lfilename = unique_name (DEFAULT_LOGFILE);
454 /* parent, no error */
455 printf (_("Continuing in background, pid %d.\n"), (int)pid);
457 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
458 exit (0); /* #### should we use _exit()? */
461 /* child: give up the privileges and keep running. */
463 freopen ("/dev/null", "r", stdin);
464 freopen ("/dev/null", "w", stdout);
465 freopen ("/dev/null", "w", stderr);
467 #endif /* not WINDOWS */
469 /* "Touch" FILE, i.e. make its atime and mtime equal to the time
470 specified with TM. */
472 touch (const char *file, time_t tm)
474 #ifdef HAVE_STRUCT_UTIMBUF
475 struct utimbuf times;
476 times.actime = times.modtime = tm;
479 times[0] = times[1] = tm;
482 if (utime (file, ×) == -1)
483 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
486 /* Checks if FILE is a symbolic link, and removes it if it is. Does
487 nothing under MS-Windows. */
489 remove_link (const char *file)
494 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
496 DEBUGP (("Unlinking %s (symlink).\n", file));
499 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
500 file, strerror (errno));
505 /* Does FILENAME exist? This is quite a lousy implementation, since
506 it supplies no error codes -- only a yes-or-no answer. Thus it
507 will return that a file does not exist if, e.g., the directory is
508 unreadable. I don't mind it too much currently, though. The
509 proper way should, of course, be to have a third, error state,
510 other than true/false, but that would introduce uncalled-for
511 additional complexity to the callers. */
513 file_exists_p (const char *filename)
516 return access (filename, F_OK) >= 0;
519 return stat (filename, &buf) >= 0;
523 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
524 Returns 0 on error. */
526 file_non_directory_p (const char *path)
529 /* Use lstat() rather than stat() so that symbolic links pointing to
530 directories can be identified correctly. */
531 if (lstat (path, &buf) != 0)
533 return S_ISDIR (buf.st_mode) ? 0 : 1;
536 /* Return a unique filename, given a prefix and count */
538 unique_name_1 (const char *fileprefix, int count)
544 filename = (char *)xmalloc (strlen (fileprefix) + numdigit (count) + 2);
545 sprintf (filename, "%s.%d", fileprefix, count);
548 filename = xstrdup (fileprefix);
550 if (!file_exists_p (filename))
559 /* Return a unique file name, based on PREFIX. */
561 unique_name (const char *prefix)
567 file = unique_name_1 (prefix, count++);
571 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
572 are missing, create them first. In case any mkdir() call fails,
573 return its error status. Returns 0 on successful completion.
575 The behaviour of this function should be identical to the behaviour
576 of `mkdir -p' on systems where mkdir supports the `-p' option. */
578 make_directory (const char *directory)
584 /* Make a copy of dir, to be able to write to it. Otherwise, the
585 function is unsafe if called with a read-only char *argument. */
586 STRDUP_ALLOCA (dir, directory);
588 /* If the first character of dir is '/', skip it (and thus enable
589 creation of absolute-pathname directories. */
590 for (i = (*dir == '/'); 1; ++i)
592 for (; dir[i] && dir[i] != '/'; i++)
597 /* Check whether the directory already exists. */
598 if (!file_exists_p (dir))
600 if (mkdir (dir, 0777) < 0)
611 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
612 should be a file name.
614 file_merge("/foo/bar", "baz") => "/foo/baz"
615 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
616 file_merge("foo", "bar") => "bar"
618 In other words, it's a simpler and gentler version of uri_merge_1. */
621 file_merge (const char *base, const char *file)
624 const char *cut = (const char *)strrchr (base, '/');
627 return xstrdup (file);
629 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
630 memcpy (result, base, cut - base);
631 result[cut - base] = '/';
632 strcpy (result + (cut - base) + 1, file);
637 static int in_acclist PARAMS ((const char *const *, const char *, int));
639 /* Determine whether a file is acceptable to be followed, according to
640 lists of patterns to accept/reject. */
642 acceptable (const char *s)
646 while (l && s[l] != '/')
653 return (in_acclist ((const char *const *)opt.accepts, s, 1)
654 && !in_acclist ((const char *const *)opt.rejects, s, 1));
656 return in_acclist ((const char *const *)opt.accepts, s, 1);
658 else if (opt.rejects)
659 return !in_acclist ((const char *const *)opt.rejects, s, 1);
663 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
664 `/something', frontcmp() will return 1 only if S2 begins with
665 `/something'. Otherwise, 0 is returned. */
667 frontcmp (const char *s1, const char *s2)
669 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
673 /* Iterate through STRLIST, and return the first element that matches
674 S, through wildcards or front comparison (as appropriate). */
676 proclist (char **strlist, const char *s, enum accd flags)
680 for (x = strlist; *x; x++)
681 if (has_wildcards_p (*x))
683 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
688 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
695 /* Returns whether DIRECTORY is acceptable for download, wrt the
696 include/exclude lists.
698 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
699 and absolute paths may be freely intermixed. */
701 accdir (const char *directory, enum accd flags)
703 /* Remove starting '/'. */
704 if (flags & ALLABS && *directory == '/')
708 if (!proclist (opt.includes, directory, flags))
713 if (proclist (opt.excludes, directory, flags))
719 /* Match the end of STRING against PATTERN. For instance:
721 match_backwards ("abc", "bc") -> 1
722 match_backwards ("abc", "ab") -> 0
723 match_backwards ("abc", "abc") -> 1 */
725 match_tail (const char *string, const char *pattern)
729 for (i = strlen (string), j = strlen (pattern); i >= 0 && j >= 0; i--, j--)
730 if (string[i] != pattern[j])
732 /* If the pattern was exhausted, the match was succesful. */
739 /* Checks whether string S matches each element of ACCEPTS. A list
740 element are matched either with fnmatch() or match_tail(),
741 according to whether the element contains wildcards or not.
743 If the BACKWARD is 0, don't do backward comparison -- just compare
746 in_acclist (const char *const *accepts, const char *s, int backward)
748 for (; *accepts; accepts++)
750 if (has_wildcards_p (*accepts))
752 /* fnmatch returns 0 if the pattern *does* match the
754 if (fnmatch (*accepts, s, 0) == 0)
761 if (match_tail (s, *accepts))
766 if (!strcmp (s, *accepts))
774 /* Return the location of STR's suffix (file extension). Examples:
775 suffix ("foo.bar") -> "bar"
776 suffix ("foo.bar.baz") -> "baz"
777 suffix ("/foo/bar") -> NULL
778 suffix ("/foo.bar/baz") -> NULL */
780 suffix (const char *str)
784 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
788 return (char *)str + i;
793 /* Read a line from FP and return the pointer to freshly allocated
794 storage. The stoarage space is obtained through malloc() and
795 should be freed with free() when it is no longer needed.
797 The length of the line is not limited, except by available memory.
798 The newline character at the end of line is retained. The line is
799 terminated with a zero character.
801 After end-of-file is encountered without anything being read, NULL
802 is returned. NULL is also returned on error. To distinguish
803 between these two cases, use the stdio function ferror(). */
806 read_whole_line (FILE *fp)
810 char *line = (char *)xmalloc (bufsize);
812 while (fgets (line + length, bufsize - length, fp))
814 length += strlen (line + length);
816 /* Possible for example when reading from a binary file where
817 a line begins with \0. */
820 if (line[length - 1] == '\n')
823 /* fgets() guarantees to read the whole line, or to use up the
824 space we've given it. We can double the buffer
827 line = xrealloc (line, bufsize);
829 if (length == 0 || ferror (fp))
834 if (length + 1 < bufsize)
835 /* Relieve the memory from our exponential greediness. We say
836 `length + 1' because the terminating \0 is not included in
837 LENGTH. We don't need to zero-terminate the string ourselves,
838 though, because fgets() does that. */
839 line = xrealloc (line, length + 1);
843 /* Read FILE into memory. A pointer to `struct file_memory' are
844 returned; use struct element `content' to access file contents, and
845 the element `length' to know the file length. `content' is *not*
846 zero-terminated, and you should *not* read or write beyond the [0,
847 length) range of characters.
849 After you are done with the file contents, call read_file_free to
852 Depending on the operating system and the type of file that is
853 being read, read_file() either mmap's the file into memory, or
854 reads the file into the core using read().
856 If file is named "-", fileno(stdin) is used for reading instead.
857 If you want to read from a real file named "-", use "./-" instead. */
860 read_file (const char *file)
863 struct file_memory *fm;
865 int inhibit_close = 0;
867 /* Some magic in the finest tradition of Perl and its kin: if FILE
868 is "-", just use stdin. */
873 /* Note that we don't inhibit mmap() in this case. If stdin is
874 redirected from a regular file, mmap() will still work. */
877 fd = open (file, O_RDONLY);
880 fm = xmalloc (sizeof (struct file_memory));
885 if (fstat (fd, &buf) < 0)
887 fm->length = buf.st_size;
888 /* NOTE: As far as I know, the callers of this function never
889 modify the file text. Relying on this would enable us to
890 specify PROT_READ and MAP_SHARED for a marginal gain in
891 efficiency, but at some cost to generality. */
892 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
894 if (fm->content == (char *)MAP_FAILED)
904 /* The most common reason why mmap() fails is that FD does not point
905 to a plain file. However, it's also possible that mmap() doesn't
906 work for a particular type of file. Therefore, whenever mmap()
907 fails, we just fall back to the regular method. */
908 #endif /* HAVE_MMAP */
911 size = 512; /* number of bytes fm->contents can
912 hold at any given time. */
913 fm->content = xmalloc (size);
917 if (fm->length > size / 2)
919 /* #### I'm not sure whether the whole exponential-growth
920 thing makes sense with kernel read. On Linux at least,
921 read() refuses to read more than 4K from a file at a
922 single chunk anyway. But other Unixes might optimize it
923 better, and it doesn't *hurt* anything, so I'm leaving
926 /* Normally, we grow SIZE exponentially to make the number
927 of calls to read() and realloc() logarithmic in relation
928 to file size. However, read() can read an amount of data
929 smaller than requested, and it would be unreasonably to
930 double SIZE every time *something* was read. Therefore,
931 we double SIZE only when the length exceeds half of the
932 entire allocated size. */
934 fm->content = xrealloc (fm->content, size);
936 nread = read (fd, fm->content + fm->length, size - fm->length);
938 /* Successful read. */
949 if (size > fm->length && fm->length != 0)
950 /* Due to exponential growth of fm->content, the allocated region
951 might be much larger than what is actually needed. */
952 fm->content = xrealloc (fm->content, fm->length);
964 /* Release the resources held by FM. Specifically, this calls
965 munmap() or xfree() on fm->content, depending whether mmap or
966 malloc/read were used to read in the file. It also frees the
967 memory needed to hold the FM structure itself. */
970 read_file_free (struct file_memory *fm)
975 munmap (fm->content, fm->length);
985 /* Free the pointers in a NULL-terminated vector of pointers, then
986 free the pointer itself. */
988 free_vec (char **vec)
999 /* Append vector V2 to vector V1. The function frees V2 and
1000 reallocates V1 (thus you may not use the contents of neither
1001 pointer after the call). If V1 is NULL, V2 is returned. */
1003 merge_vecs (char **v1, char **v2)
1013 /* To avoid j == 0 */
1018 for (i = 0; v1[i]; i++);
1020 for (j = 0; v2[j]; j++);
1021 /* Reallocate v1. */
1022 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1023 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1028 /* A set of simple-minded routines to store strings in a linked list.
1029 This used to also be used for searching, but now we have hash
1032 /* It's a shame that these simple things like linked lists and hash
1033 tables (see hash.c) need to be implemented over and over again. It
1034 would be nice to be able to use the routines from glib -- see
1035 www.gtk.org for details. However, that would make Wget depend on
1036 glib, and I want to avoid dependencies to external libraries for
1037 reasons of convenience and portability (I suspect Wget is more
1038 portable than anything ever written for Gnome). */
1040 /* Append an element to the list. If the list has a huge number of
1041 elements, this can get slow because it has to find the list's
1042 ending. If you think you have to call slist_append in a loop,
1043 think about calling slist_prepend() followed by slist_nreverse(). */
1046 slist_append (slist *l, const char *s)
1048 slist *newel = (slist *)xmalloc (sizeof (slist));
1051 newel->string = xstrdup (s);
1056 /* Find the last element. */
1063 /* Prepend S to the list. Unlike slist_append(), this is O(1). */
1066 slist_prepend (slist *l, const char *s)
1068 slist *newel = (slist *)xmalloc (sizeof (slist));
1069 newel->string = xstrdup (s);
1074 /* Destructively reverse L. */
1077 slist_nreverse (slist *l)
1082 slist *next = l->next;
1090 /* Is there a specific entry in the list? */
1092 slist_contains (slist *l, const char *s)
1094 for (; l; l = l->next)
1095 if (!strcmp (l->string, s))
1100 /* Free the whole slist. */
1102 slist_free (slist *l)
1113 /* Sometimes it's useful to create "sets" of strings, i.e. special
1114 hash tables where you want to store strings as keys and merely
1115 query for their existence. Here is a set of utility routines that
1116 makes that transparent. */
1119 string_set_add (struct hash_table *ht, const char *s)
1121 /* First check whether the set element already exists. If it does,
1122 do nothing so that we don't have to free() the old element and
1123 then strdup() a new one. */
1124 if (hash_table_contains (ht, s))
1127 /* We use "1" as value. It provides us a useful and clear arbitrary
1128 value, and it consumes no memory -- the pointers to the same
1129 string "1" will be shared by all the key-value pairs in all `set'
1131 hash_table_put (ht, xstrdup (s), "1");
1134 /* Synonym for hash_table_contains... */
1137 string_set_contains (struct hash_table *ht, const char *s)
1139 return hash_table_contains (ht, s);
1143 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1150 string_set_free (struct hash_table *ht)
1152 hash_table_map (ht, string_set_free_mapper, NULL);
1153 hash_table_destroy (ht);
1157 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1164 /* Another utility function: call free() on all keys and values of HT. */
1167 free_keys_and_values (struct hash_table *ht)
1169 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1173 /* Engine for legible and legible_very_long; this function works on
1177 legible_1 (const char *repr)
1179 static char outbuf[128];
1184 /* Reset the pointers. */
1187 /* If the number is negative, shift the pointers. */
1193 /* How many digits before the first separator? */
1194 mod = strlen (inptr) % 3;
1196 for (i = 0; i < mod; i++)
1197 *outptr++ = inptr[i];
1198 /* Now insert the rest of them, putting separator before every
1200 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1202 if (i % 3 == 0 && i1 != 0)
1204 *outptr++ = inptr[i1];
1206 /* Zero-terminate the string. */
1211 /* Legible -- return a static pointer to the legibly printed long. */
1216 /* Print the number into the buffer. */
1217 number_to_string (inbuf, l);
1218 return legible_1 (inbuf);
1221 /* Write a string representation of NUMBER into the provided buffer.
1222 We cannot use sprintf() because we cannot be sure whether the
1223 platform supports printing of what we chose for VERY_LONG_TYPE.
1225 Example: Gcc supports `long long' under many platforms, but on many
1226 of those the native libc knows nothing of it and therefore cannot
1229 How long BUFFER needs to be depends on the platform and the content
1230 of NUMBER. For 64-bit VERY_LONG_TYPE (the most common case), 24
1231 bytes are sufficient. Using more might be a good idea.
1233 This function does not go through the hoops that long_to_string
1234 goes to because it doesn't aspire to be fast. (It's called perhaps
1235 once in a Wget run.) */
1238 very_long_to_string (char *buffer, VERY_LONG_TYPE number)
1243 /* Print the number backwards... */
1246 buffer[i++] = '0' + number % 10;
1251 /* ...and reverse the order of the digits. */
1252 for (j = 0; j < i / 2; j++)
1255 buffer[j] = buffer[i - 1 - j];
1256 buffer[i - 1 - j] = c;
1261 /* The same as legible(), but works on VERY_LONG_TYPE. See sysdep.h. */
1263 legible_very_long (VERY_LONG_TYPE l)
1266 /* Print the number into the buffer. */
1267 very_long_to_string (inbuf, l);
1268 return legible_1 (inbuf);
1271 /* Count the digits in a (long) integer. */
1273 numdigit (long number)
1281 while ((number /= 10) > 0)
1286 #define ONE_DIGIT(figure) *p++ = n / (figure) + '0'
1287 #define ONE_DIGIT_ADVANCE(figure) (ONE_DIGIT (figure), n %= (figure))
1289 #define DIGITS_1(figure) ONE_DIGIT (figure)
1290 #define DIGITS_2(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_1 ((figure) / 10)
1291 #define DIGITS_3(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_2 ((figure) / 10)
1292 #define DIGITS_4(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_3 ((figure) / 10)
1293 #define DIGITS_5(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_4 ((figure) / 10)
1294 #define DIGITS_6(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_5 ((figure) / 10)
1295 #define DIGITS_7(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_6 ((figure) / 10)
1296 #define DIGITS_8(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_7 ((figure) / 10)
1297 #define DIGITS_9(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_8 ((figure) / 10)
1298 #define DIGITS_10(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_9 ((figure) / 10)
1300 /* DIGITS_<11-20> are only used on machines with 64-bit longs. */
1302 #define DIGITS_11(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_10 ((figure) / 10)
1303 #define DIGITS_12(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_11 ((figure) / 10)
1304 #define DIGITS_13(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_12 ((figure) / 10)
1305 #define DIGITS_14(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_13 ((figure) / 10)
1306 #define DIGITS_15(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_14 ((figure) / 10)
1307 #define DIGITS_16(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_15 ((figure) / 10)
1308 #define DIGITS_17(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_16 ((figure) / 10)
1309 #define DIGITS_18(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_17 ((figure) / 10)
1310 #define DIGITS_19(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_18 ((figure) / 10)
1312 /* Print NUMBER to BUFFER in base 10. This should be completely
1313 equivalent to `sprintf(buffer, "%ld", number)', only much faster.
1315 The speedup may make a difference in programs that frequently
1316 convert numbers to strings. Some implementations of sprintf,
1317 particularly the one in GNU libc, have been known to be extremely
1318 slow compared to this function.
1320 Return the pointer to the location where the terminating zero was
1321 printed. (Equivalent to calling buffer+strlen(buffer) after the
1324 BUFFER should be big enough to accept as many bytes as you expect
1325 the number to take up. On machines with 64-bit longs the maximum
1326 needed size is 24 bytes. That includes the digits needed for the
1327 largest 64-bit number, the `-' sign in case it's negative, and the
1328 terminating '\0'. */
1331 number_to_string (char *buffer, long number)
1336 #if (SIZEOF_LONG != 4) && (SIZEOF_LONG != 8)
1337 /* We are running in a strange or misconfigured environment. Let
1338 sprintf cope with it. */
1339 sprintf (buffer, "%ld", n);
1340 p += strlen (buffer);
1341 #else /* (SIZEOF_LONG == 4) || (SIZEOF_LONG == 8) */
1349 if (n < 10) { DIGITS_1 (1); }
1350 else if (n < 100) { DIGITS_2 (10); }
1351 else if (n < 1000) { DIGITS_3 (100); }
1352 else if (n < 10000) { DIGITS_4 (1000); }
1353 else if (n < 100000) { DIGITS_5 (10000); }
1354 else if (n < 1000000) { DIGITS_6 (100000); }
1355 else if (n < 10000000) { DIGITS_7 (1000000); }
1356 else if (n < 100000000) { DIGITS_8 (10000000); }
1357 else if (n < 1000000000) { DIGITS_9 (100000000); }
1358 #if SIZEOF_LONG == 4
1359 /* ``if (1)'' serves only to preserve editor indentation. */
1360 else if (1) { DIGITS_10 (1000000000); }
1361 #else /* SIZEOF_LONG != 4 */
1362 else if (n < 10000000000L) { DIGITS_10 (1000000000L); }
1363 else if (n < 100000000000L) { DIGITS_11 (10000000000L); }
1364 else if (n < 1000000000000L) { DIGITS_12 (100000000000L); }
1365 else if (n < 10000000000000L) { DIGITS_13 (1000000000000L); }
1366 else if (n < 100000000000000L) { DIGITS_14 (10000000000000L); }
1367 else if (n < 1000000000000000L) { DIGITS_15 (100000000000000L); }
1368 else if (n < 10000000000000000L) { DIGITS_16 (1000000000000000L); }
1369 else if (n < 100000000000000000L) { DIGITS_17 (10000000000000000L); }
1370 else if (n < 1000000000000000000L) { DIGITS_18 (100000000000000000L); }
1371 else { DIGITS_19 (1000000000000000000L); }
1372 #endif /* SIZEOF_LONG != 4 */
1375 #endif /* (SIZEOF_LONG == 4) || (SIZEOF_LONG == 8) */
1381 #undef ONE_DIGIT_ADVANCE
1403 /* Support for timers. */
1405 #undef TIMER_WINDOWS
1406 #undef TIMER_GETTIMEOFDAY
1409 /* Depending on the OS and availability of gettimeofday(), one and
1410 only one of the above constants will be defined. Virtually all
1411 modern Unix systems will define TIMER_GETTIMEOFDAY; Windows will
1412 use TIMER_WINDOWS. TIMER_TIME is a catch-all method for
1413 non-Windows systems without gettimeofday.
1415 #### Perhaps we should also support ftime(), which exists on old
1416 BSD 4.2-influenced systems? (It also existed under MS DOS Borland
1417 C, if memory serves me.) */
1420 # define TIMER_WINDOWS
1421 #else /* not WINDOWS */
1422 # ifdef HAVE_GETTIMEOFDAY
1423 # define TIMER_GETTIMEOFDAY
1427 #endif /* not WINDOWS */
1430 #ifdef TIMER_GETTIMEOFDAY
1439 #ifdef TIMER_WINDOWS
1440 ULARGE_INTEGER wintime;
1444 /* Allocate a timer. It is not legal to do anything with a freshly
1445 allocated timer, except call wtimer_reset() or wtimer_delete(). */
1448 wtimer_allocate (void)
1450 struct wget_timer *wt =
1451 (struct wget_timer *)xmalloc (sizeof (struct wget_timer));
1455 /* Allocate a new timer and reset it. Return the new timer. */
1460 struct wget_timer *wt = wtimer_allocate ();
1465 /* Free the resources associated with the timer. Its further use is
1469 wtimer_delete (struct wget_timer *wt)
1474 /* Reset timer WT. This establishes the starting point from which
1475 wtimer_elapsed() will return the number of elapsed
1476 milliseconds. It is allowed to reset a previously used timer. */
1479 wtimer_reset (struct wget_timer *wt)
1481 #ifdef TIMER_GETTIMEOFDAY
1483 gettimeofday (&t, NULL);
1484 wt->secs = t.tv_sec;
1485 wt->usecs = t.tv_usec;
1489 wt->secs = time (NULL);
1492 #ifdef TIMER_WINDOWS
1495 GetSystemTime (&st);
1496 SystemTimeToFileTime (&st, &ft);
1497 wt->wintime.HighPart = ft.dwHighDateTime;
1498 wt->wintime.LowPart = ft.dwLowDateTime;
1502 /* Return the number of milliseconds elapsed since the timer was last
1503 reset. It is allowed to call this function more than once to get
1504 increasingly higher elapsed values. */
1507 wtimer_elapsed (struct wget_timer *wt)
1509 #ifdef TIMER_GETTIMEOFDAY
1511 gettimeofday (&t, NULL);
1512 return (t.tv_sec - wt->secs) * 1000 + (t.tv_usec - wt->usecs) / 1000;
1516 time_t now = time (NULL);
1517 return 1000 * (now - wt->secs);
1524 GetSystemTime (&st);
1525 SystemTimeToFileTime (&st, &ft);
1526 uli.HighPart = ft.dwHighDateTime;
1527 uli.LowPart = ft.dwLowDateTime;
1528 return (long)((uli.QuadPart - wt->wintime.QuadPart) / 10000);
1532 /* Return the assessed granularity of the timer implementation. This
1533 is important for certain code that tries to deal with "zero" time
1537 wtimer_granularity (void)
1539 #ifdef TIMER_GETTIMEOFDAY
1540 /* Granularity of gettimeofday is hugely architecture-dependent.
1541 However, it appears that on modern machines it is better than
1547 /* This is clear. */
1551 #ifdef TIMER_WINDOWS
1557 /* This should probably be at a better place, but it doesn't really
1558 fit into html-parse.c. */
1560 /* The function returns the pointer to the malloc-ed quoted version of
1561 string s. It will recognize and quote numeric and special graphic
1562 entities, as per RFC1866:
1570 No other entities are recognized or replaced. */
1572 html_quote_string (const char *s)
1578 /* Pass through the string, and count the new size. */
1579 for (i = 0; *s; s++, i++)
1582 i += 4; /* `amp;' */
1583 else if (*s == '<' || *s == '>')
1584 i += 3; /* `lt;' and `gt;' */
1585 else if (*s == '\"')
1586 i += 5; /* `quot;' */
1590 res = (char *)xmalloc (i + 1);
1592 for (p = res; *s; s++)
1605 *p++ = (*s == '<' ? 'l' : 'g');
1632 /* Determine the width of the terminal we're running on. If that's
1633 not possible, return 0. */
1636 determine_screen_width (void)
1638 /* If there's a way to get the terminal size using POSIX
1639 tcgetattr(), somebody please tell me. */
1642 #else /* TIOCGWINSZ */
1646 if (opt.lfilename != NULL)
1649 fd = fileno (stderr);
1650 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1651 return 0; /* most likely ENOTTY */
1654 #endif /* TIOCGWINSZ */
1657 /* Return a random number between 0 and MAX-1, inclusive.
1659 If MAX is greater than the value of RAND_MAX+1 on the system, the
1660 returned value will be in the range [0, RAND_MAX]. This may be
1661 fixed in a future release.
1663 The random number generator is seeded automatically the first time
1666 This uses rand() for portability. It has been suggested that
1667 random() offers better randomness, but this is not required for
1668 Wget, so I chose to go for simplicity and use rand
1672 random_number (int max)
1680 srand (time (NULL));
1685 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1686 and enforce that assumption by masking other bits. */
1688 # define RAND_MAX 32767
1692 /* This is equivalent to rand() % max, but uses the high-order bits
1693 for better randomness on architecture where rand() is implemented
1694 using a simple congruential generator. */
1696 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1697 return (int)bounded;
1701 /* A debugging function for checking whether an MD5 library works. */
1703 #include "gen-md5.h"
1706 debug_test_md5 (char *buf)
1708 unsigned char raw[16];
1709 static char res[33];
1713 ALLOCA_MD5_CONTEXT (ctx);
1716 gen_md5_update ((unsigned char *)buf, strlen (buf), ctx);
1717 gen_md5_finish (ctx, raw);
1724 *p2++ = XDIGIT_TO_xchar (*p1 >> 4);
1725 *p2++ = XDIGIT_TO_xchar (*p1 & 0xf);