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)
585 /* Make a copy of dir, to be able to write to it. Otherwise, the
586 function is unsafe if called with a read-only char *argument. */
587 STRDUP_ALLOCA (dir, directory);
589 /* If the first character of dir is '/', skip it (and thus enable
590 creation of absolute-pathname directories. */
591 for (i = (*dir == '/'); 1; ++i)
593 for (; dir[i] && dir[i] != '/'; i++)
598 /* Check whether the directory already exists. Allow creation of
599 of intermediate directories to fail, as the initial path components
600 are not necessarily directories! */
601 if (!file_exists_p (dir))
602 ret = mkdir (dir, 0777);
613 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
614 should be a file name.
616 file_merge("/foo/bar", "baz") => "/foo/baz"
617 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
618 file_merge("foo", "bar") => "bar"
620 In other words, it's a simpler and gentler version of uri_merge_1. */
623 file_merge (const char *base, const char *file)
626 const char *cut = (const char *)strrchr (base, '/');
629 return xstrdup (file);
631 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
632 memcpy (result, base, cut - base);
633 result[cut - base] = '/';
634 strcpy (result + (cut - base) + 1, file);
639 static int in_acclist PARAMS ((const char *const *, const char *, int));
641 /* Determine whether a file is acceptable to be followed, according to
642 lists of patterns to accept/reject. */
644 acceptable (const char *s)
648 while (l && s[l] != '/')
655 return (in_acclist ((const char *const *)opt.accepts, s, 1)
656 && !in_acclist ((const char *const *)opt.rejects, s, 1));
658 return in_acclist ((const char *const *)opt.accepts, s, 1);
660 else if (opt.rejects)
661 return !in_acclist ((const char *const *)opt.rejects, s, 1);
665 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
666 `/something', frontcmp() will return 1 only if S2 begins with
667 `/something'. Otherwise, 0 is returned. */
669 frontcmp (const char *s1, const char *s2)
671 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
675 /* Iterate through STRLIST, and return the first element that matches
676 S, through wildcards or front comparison (as appropriate). */
678 proclist (char **strlist, const char *s, enum accd flags)
682 for (x = strlist; *x; x++)
683 if (has_wildcards_p (*x))
685 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
690 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
697 /* Returns whether DIRECTORY is acceptable for download, wrt the
698 include/exclude lists.
700 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
701 and absolute paths may be freely intermixed. */
703 accdir (const char *directory, enum accd flags)
705 /* Remove starting '/'. */
706 if (flags & ALLABS && *directory == '/')
710 if (!proclist (opt.includes, directory, flags))
715 if (proclist (opt.excludes, directory, flags))
721 /* Match the end of STRING against PATTERN. For instance:
723 match_backwards ("abc", "bc") -> 1
724 match_backwards ("abc", "ab") -> 0
725 match_backwards ("abc", "abc") -> 1 */
727 match_tail (const char *string, const char *pattern)
731 for (i = strlen (string), j = strlen (pattern); i >= 0 && j >= 0; i--, j--)
732 if (string[i] != pattern[j])
734 /* If the pattern was exhausted, the match was succesful. */
741 /* Checks whether string S matches each element of ACCEPTS. A list
742 element are matched either with fnmatch() or match_tail(),
743 according to whether the element contains wildcards or not.
745 If the BACKWARD is 0, don't do backward comparison -- just compare
748 in_acclist (const char *const *accepts, const char *s, int backward)
750 for (; *accepts; accepts++)
752 if (has_wildcards_p (*accepts))
754 /* fnmatch returns 0 if the pattern *does* match the
756 if (fnmatch (*accepts, s, 0) == 0)
763 if (match_tail (s, *accepts))
768 if (!strcmp (s, *accepts))
776 /* Return the location of STR's suffix (file extension). Examples:
777 suffix ("foo.bar") -> "bar"
778 suffix ("foo.bar.baz") -> "baz"
779 suffix ("/foo/bar") -> NULL
780 suffix ("/foo.bar/baz") -> NULL */
782 suffix (const char *str)
786 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
790 return (char *)str + i;
795 /* Return non-zero if FNAME ends with a typical HTML suffix. The
796 following (case-insensitive) suffixes are presumed to be HTML files:
800 ?html (`?' matches one character)
802 #### CAVEAT. This is not necessarily a good indication that FNAME
803 refers to a file that contains HTML! */
805 has_html_suffix_p (const char *fname)
809 if ((suf = suffix (fname)) == NULL)
811 if (!strcasecmp (suf, "html"))
813 if (!strcasecmp (suf, "htm"))
815 if (suf[0] && !strcasecmp (suf + 1, "html"))
820 /* Read a line from FP and return the pointer to freshly allocated
821 storage. The stoarage space is obtained through malloc() and
822 should be freed with free() when it is no longer needed.
824 The length of the line is not limited, except by available memory.
825 The newline character at the end of line is retained. The line is
826 terminated with a zero character.
828 After end-of-file is encountered without anything being read, NULL
829 is returned. NULL is also returned on error. To distinguish
830 between these two cases, use the stdio function ferror(). */
833 read_whole_line (FILE *fp)
837 char *line = (char *)xmalloc (bufsize);
839 while (fgets (line + length, bufsize - length, fp))
841 length += strlen (line + length);
843 /* Possible for example when reading from a binary file where
844 a line begins with \0. */
847 if (line[length - 1] == '\n')
850 /* fgets() guarantees to read the whole line, or to use up the
851 space we've given it. We can double the buffer
854 line = xrealloc (line, bufsize);
856 if (length == 0 || ferror (fp))
861 if (length + 1 < bufsize)
862 /* Relieve the memory from our exponential greediness. We say
863 `length + 1' because the terminating \0 is not included in
864 LENGTH. We don't need to zero-terminate the string ourselves,
865 though, because fgets() does that. */
866 line = xrealloc (line, length + 1);
870 /* Read FILE into memory. A pointer to `struct file_memory' are
871 returned; use struct element `content' to access file contents, and
872 the element `length' to know the file length. `content' is *not*
873 zero-terminated, and you should *not* read or write beyond the [0,
874 length) range of characters.
876 After you are done with the file contents, call read_file_free to
879 Depending on the operating system and the type of file that is
880 being read, read_file() either mmap's the file into memory, or
881 reads the file into the core using read().
883 If file is named "-", fileno(stdin) is used for reading instead.
884 If you want to read from a real file named "-", use "./-" instead. */
887 read_file (const char *file)
890 struct file_memory *fm;
892 int inhibit_close = 0;
894 /* Some magic in the finest tradition of Perl and its kin: if FILE
895 is "-", just use stdin. */
900 /* Note that we don't inhibit mmap() in this case. If stdin is
901 redirected from a regular file, mmap() will still work. */
904 fd = open (file, O_RDONLY);
907 fm = xmalloc (sizeof (struct file_memory));
912 if (fstat (fd, &buf) < 0)
914 fm->length = buf.st_size;
915 /* NOTE: As far as I know, the callers of this function never
916 modify the file text. Relying on this would enable us to
917 specify PROT_READ and MAP_SHARED for a marginal gain in
918 efficiency, but at some cost to generality. */
919 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
921 if (fm->content == (char *)MAP_FAILED)
931 /* The most common reason why mmap() fails is that FD does not point
932 to a plain file. However, it's also possible that mmap() doesn't
933 work for a particular type of file. Therefore, whenever mmap()
934 fails, we just fall back to the regular method. */
935 #endif /* HAVE_MMAP */
938 size = 512; /* number of bytes fm->contents can
939 hold at any given time. */
940 fm->content = xmalloc (size);
944 if (fm->length > size / 2)
946 /* #### I'm not sure whether the whole exponential-growth
947 thing makes sense with kernel read. On Linux at least,
948 read() refuses to read more than 4K from a file at a
949 single chunk anyway. But other Unixes might optimize it
950 better, and it doesn't *hurt* anything, so I'm leaving
953 /* Normally, we grow SIZE exponentially to make the number
954 of calls to read() and realloc() logarithmic in relation
955 to file size. However, read() can read an amount of data
956 smaller than requested, and it would be unreasonably to
957 double SIZE every time *something* was read. Therefore,
958 we double SIZE only when the length exceeds half of the
959 entire allocated size. */
961 fm->content = xrealloc (fm->content, size);
963 nread = read (fd, fm->content + fm->length, size - fm->length);
965 /* Successful read. */
976 if (size > fm->length && fm->length != 0)
977 /* Due to exponential growth of fm->content, the allocated region
978 might be much larger than what is actually needed. */
979 fm->content = xrealloc (fm->content, fm->length);
991 /* Release the resources held by FM. Specifically, this calls
992 munmap() or xfree() on fm->content, depending whether mmap or
993 malloc/read were used to read in the file. It also frees the
994 memory needed to hold the FM structure itself. */
997 read_file_free (struct file_memory *fm)
1002 munmap (fm->content, fm->length);
1007 xfree (fm->content);
1012 /* Free the pointers in a NULL-terminated vector of pointers, then
1013 free the pointer itself. */
1015 free_vec (char **vec)
1026 /* Append vector V2 to vector V1. The function frees V2 and
1027 reallocates V1 (thus you may not use the contents of neither
1028 pointer after the call). If V1 is NULL, V2 is returned. */
1030 merge_vecs (char **v1, char **v2)
1040 /* To avoid j == 0 */
1045 for (i = 0; v1[i]; i++);
1047 for (j = 0; v2[j]; j++);
1048 /* Reallocate v1. */
1049 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1050 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1055 /* A set of simple-minded routines to store strings in a linked list.
1056 This used to also be used for searching, but now we have hash
1059 /* It's a shame that these simple things like linked lists and hash
1060 tables (see hash.c) need to be implemented over and over again. It
1061 would be nice to be able to use the routines from glib -- see
1062 www.gtk.org for details. However, that would make Wget depend on
1063 glib, and I want to avoid dependencies to external libraries for
1064 reasons of convenience and portability (I suspect Wget is more
1065 portable than anything ever written for Gnome). */
1067 /* Append an element to the list. If the list has a huge number of
1068 elements, this can get slow because it has to find the list's
1069 ending. If you think you have to call slist_append in a loop,
1070 think about calling slist_prepend() followed by slist_nreverse(). */
1073 slist_append (slist *l, const char *s)
1075 slist *newel = (slist *)xmalloc (sizeof (slist));
1078 newel->string = xstrdup (s);
1083 /* Find the last element. */
1090 /* Prepend S to the list. Unlike slist_append(), this is O(1). */
1093 slist_prepend (slist *l, const char *s)
1095 slist *newel = (slist *)xmalloc (sizeof (slist));
1096 newel->string = xstrdup (s);
1101 /* Destructively reverse L. */
1104 slist_nreverse (slist *l)
1109 slist *next = l->next;
1117 /* Is there a specific entry in the list? */
1119 slist_contains (slist *l, const char *s)
1121 for (; l; l = l->next)
1122 if (!strcmp (l->string, s))
1127 /* Free the whole slist. */
1129 slist_free (slist *l)
1140 /* Sometimes it's useful to create "sets" of strings, i.e. special
1141 hash tables where you want to store strings as keys and merely
1142 query for their existence. Here is a set of utility routines that
1143 makes that transparent. */
1146 string_set_add (struct hash_table *ht, const char *s)
1148 /* First check whether the set element already exists. If it does,
1149 do nothing so that we don't have to free() the old element and
1150 then strdup() a new one. */
1151 if (hash_table_contains (ht, s))
1154 /* We use "1" as value. It provides us a useful and clear arbitrary
1155 value, and it consumes no memory -- the pointers to the same
1156 string "1" will be shared by all the key-value pairs in all `set'
1158 hash_table_put (ht, xstrdup (s), "1");
1161 /* Synonym for hash_table_contains... */
1164 string_set_contains (struct hash_table *ht, const char *s)
1166 return hash_table_contains (ht, s);
1170 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1177 string_set_free (struct hash_table *ht)
1179 hash_table_map (ht, string_set_free_mapper, NULL);
1180 hash_table_destroy (ht);
1184 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1191 /* Another utility function: call free() on all keys and values of HT. */
1194 free_keys_and_values (struct hash_table *ht)
1196 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1200 /* Engine for legible and legible_very_long; this function works on
1204 legible_1 (const char *repr)
1206 static char outbuf[128];
1211 /* Reset the pointers. */
1214 /* If the number is negative, shift the pointers. */
1220 /* How many digits before the first separator? */
1221 mod = strlen (inptr) % 3;
1223 for (i = 0; i < mod; i++)
1224 *outptr++ = inptr[i];
1225 /* Now insert the rest of them, putting separator before every
1227 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1229 if (i % 3 == 0 && i1 != 0)
1231 *outptr++ = inptr[i1];
1233 /* Zero-terminate the string. */
1238 /* Legible -- return a static pointer to the legibly printed long. */
1243 /* Print the number into the buffer. */
1244 number_to_string (inbuf, l);
1245 return legible_1 (inbuf);
1248 /* Write a string representation of NUMBER into the provided buffer.
1249 We cannot use sprintf() because we cannot be sure whether the
1250 platform supports printing of what we chose for VERY_LONG_TYPE.
1252 Example: Gcc supports `long long' under many platforms, but on many
1253 of those the native libc knows nothing of it and therefore cannot
1256 How long BUFFER needs to be depends on the platform and the content
1257 of NUMBER. For 64-bit VERY_LONG_TYPE (the most common case), 24
1258 bytes are sufficient. Using more might be a good idea.
1260 This function does not go through the hoops that long_to_string
1261 goes to because it doesn't aspire to be fast. (It's called perhaps
1262 once in a Wget run.) */
1265 very_long_to_string (char *buffer, VERY_LONG_TYPE number)
1270 /* Print the number backwards... */
1273 buffer[i++] = '0' + number % 10;
1278 /* ...and reverse the order of the digits. */
1279 for (j = 0; j < i / 2; j++)
1282 buffer[j] = buffer[i - 1 - j];
1283 buffer[i - 1 - j] = c;
1288 /* The same as legible(), but works on VERY_LONG_TYPE. See sysdep.h. */
1290 legible_very_long (VERY_LONG_TYPE l)
1293 /* Print the number into the buffer. */
1294 very_long_to_string (inbuf, l);
1295 return legible_1 (inbuf);
1298 /* Count the digits in a (long) integer. */
1300 numdigit (long 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 longs. */
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 /* Print NUMBER to BUFFER in base 10. This should be completely
1340 equivalent to `sprintf(buffer, "%ld", number)', only much faster.
1342 The speedup may make a difference in programs that frequently
1343 convert numbers to strings. Some implementations of sprintf,
1344 particularly the one in GNU libc, have been known to be extremely
1345 slow compared to this function.
1347 Return the pointer to the location where the terminating zero was
1348 printed. (Equivalent to calling buffer+strlen(buffer) after the
1351 BUFFER should be big enough to accept as many bytes as you expect
1352 the number to take up. On machines with 64-bit longs the maximum
1353 needed size is 24 bytes. That includes the digits needed for the
1354 largest 64-bit number, the `-' sign in case it's negative, and the
1355 terminating '\0'. */
1358 number_to_string (char *buffer, long number)
1363 #if (SIZEOF_LONG != 4) && (SIZEOF_LONG != 8)
1364 /* We are running in a strange or misconfigured environment. Let
1365 sprintf cope with it. */
1366 sprintf (buffer, "%ld", n);
1367 p += strlen (buffer);
1368 #else /* (SIZEOF_LONG == 4) || (SIZEOF_LONG == 8) */
1376 if (n < 10) { DIGITS_1 (1); }
1377 else if (n < 100) { DIGITS_2 (10); }
1378 else if (n < 1000) { DIGITS_3 (100); }
1379 else if (n < 10000) { DIGITS_4 (1000); }
1380 else if (n < 100000) { DIGITS_5 (10000); }
1381 else if (n < 1000000) { DIGITS_6 (100000); }
1382 else if (n < 10000000) { DIGITS_7 (1000000); }
1383 else if (n < 100000000) { DIGITS_8 (10000000); }
1384 else if (n < 1000000000) { DIGITS_9 (100000000); }
1385 #if SIZEOF_LONG == 4
1386 /* ``if (1)'' serves only to preserve editor indentation. */
1387 else if (1) { DIGITS_10 (1000000000); }
1388 #else /* SIZEOF_LONG != 4 */
1389 else if (n < 10000000000L) { DIGITS_10 (1000000000L); }
1390 else if (n < 100000000000L) { DIGITS_11 (10000000000L); }
1391 else if (n < 1000000000000L) { DIGITS_12 (100000000000L); }
1392 else if (n < 10000000000000L) { DIGITS_13 (1000000000000L); }
1393 else if (n < 100000000000000L) { DIGITS_14 (10000000000000L); }
1394 else if (n < 1000000000000000L) { DIGITS_15 (100000000000000L); }
1395 else if (n < 10000000000000000L) { DIGITS_16 (1000000000000000L); }
1396 else if (n < 100000000000000000L) { DIGITS_17 (10000000000000000L); }
1397 else if (n < 1000000000000000000L) { DIGITS_18 (100000000000000000L); }
1398 else { DIGITS_19 (1000000000000000000L); }
1399 #endif /* SIZEOF_LONG != 4 */
1402 #endif /* (SIZEOF_LONG == 4) || (SIZEOF_LONG == 8) */
1408 #undef ONE_DIGIT_ADVANCE
1430 /* Support for timers. */
1432 #undef TIMER_WINDOWS
1433 #undef TIMER_GETTIMEOFDAY
1436 /* Depending on the OS and availability of gettimeofday(), one and
1437 only one of the above constants will be defined. Virtually all
1438 modern Unix systems will define TIMER_GETTIMEOFDAY; Windows will
1439 use TIMER_WINDOWS. TIMER_TIME is a catch-all method for
1440 non-Windows systems without gettimeofday.
1442 #### Perhaps we should also support ftime(), which exists on old
1443 BSD 4.2-influenced systems? (It also existed under MS DOS Borland
1444 C, if memory serves me.) */
1447 # define TIMER_WINDOWS
1448 #else /* not WINDOWS */
1449 # ifdef HAVE_GETTIMEOFDAY
1450 # define TIMER_GETTIMEOFDAY
1454 #endif /* not WINDOWS */
1457 #ifdef TIMER_GETTIMEOFDAY
1466 #ifdef TIMER_WINDOWS
1467 ULARGE_INTEGER wintime;
1471 /* Allocate a timer. It is not legal to do anything with a freshly
1472 allocated timer, except call wtimer_reset() or wtimer_delete(). */
1475 wtimer_allocate (void)
1477 struct wget_timer *wt =
1478 (struct wget_timer *)xmalloc (sizeof (struct wget_timer));
1482 /* Allocate a new timer and reset it. Return the new timer. */
1487 struct wget_timer *wt = wtimer_allocate ();
1492 /* Free the resources associated with the timer. Its further use is
1496 wtimer_delete (struct wget_timer *wt)
1501 /* Reset timer WT. This establishes the starting point from which
1502 wtimer_elapsed() will return the number of elapsed
1503 milliseconds. It is allowed to reset a previously used timer. */
1506 wtimer_reset (struct wget_timer *wt)
1508 #ifdef TIMER_GETTIMEOFDAY
1510 gettimeofday (&t, NULL);
1511 wt->secs = t.tv_sec;
1512 wt->usecs = t.tv_usec;
1516 wt->secs = time (NULL);
1519 #ifdef TIMER_WINDOWS
1522 GetSystemTime (&st);
1523 SystemTimeToFileTime (&st, &ft);
1524 wt->wintime.HighPart = ft.dwHighDateTime;
1525 wt->wintime.LowPart = ft.dwLowDateTime;
1529 /* Return the number of milliseconds elapsed since the timer was last
1530 reset. It is allowed to call this function more than once to get
1531 increasingly higher elapsed values. */
1534 wtimer_elapsed (struct wget_timer *wt)
1536 #ifdef TIMER_GETTIMEOFDAY
1538 gettimeofday (&t, NULL);
1539 return (t.tv_sec - wt->secs) * 1000 + (t.tv_usec - wt->usecs) / 1000;
1543 time_t now = time (NULL);
1544 return 1000 * (now - wt->secs);
1551 GetSystemTime (&st);
1552 SystemTimeToFileTime (&st, &ft);
1553 uli.HighPart = ft.dwHighDateTime;
1554 uli.LowPart = ft.dwLowDateTime;
1555 return (long)((uli.QuadPart - wt->wintime.QuadPart) / 10000);
1559 /* Return the assessed granularity of the timer implementation. This
1560 is important for certain code that tries to deal with "zero" time
1564 wtimer_granularity (void)
1566 #ifdef TIMER_GETTIMEOFDAY
1567 /* Granularity of gettimeofday is hugely architecture-dependent.
1568 However, it appears that on modern machines it is better than
1574 /* This is clear. */
1578 #ifdef TIMER_WINDOWS
1584 /* This should probably be at a better place, but it doesn't really
1585 fit into html-parse.c. */
1587 /* The function returns the pointer to the malloc-ed quoted version of
1588 string s. It will recognize and quote numeric and special graphic
1589 entities, as per RFC1866:
1597 No other entities are recognized or replaced. */
1599 html_quote_string (const char *s)
1605 /* Pass through the string, and count the new size. */
1606 for (i = 0; *s; s++, i++)
1609 i += 4; /* `amp;' */
1610 else if (*s == '<' || *s == '>')
1611 i += 3; /* `lt;' and `gt;' */
1612 else if (*s == '\"')
1613 i += 5; /* `quot;' */
1617 res = (char *)xmalloc (i + 1);
1619 for (p = res; *s; s++)
1632 *p++ = (*s == '<' ? 'l' : 'g');
1659 /* Determine the width of the terminal we're running on. If that's
1660 not possible, return 0. */
1663 determine_screen_width (void)
1665 /* If there's a way to get the terminal size using POSIX
1666 tcgetattr(), somebody please tell me. */
1669 #else /* TIOCGWINSZ */
1673 if (opt.lfilename != NULL)
1676 fd = fileno (stderr);
1677 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1678 return 0; /* most likely ENOTTY */
1681 #endif /* TIOCGWINSZ */
1684 /* Return a random number between 0 and MAX-1, inclusive.
1686 If MAX is greater than the value of RAND_MAX+1 on the system, the
1687 returned value will be in the range [0, RAND_MAX]. This may be
1688 fixed in a future release.
1690 The random number generator is seeded automatically the first time
1693 This uses rand() for portability. It has been suggested that
1694 random() offers better randomness, but this is not required for
1695 Wget, so I chose to go for simplicity and use rand
1699 random_number (int max)
1707 srand (time (NULL));
1712 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1713 and enforce that assumption by masking other bits. */
1715 # define RAND_MAX 32767
1719 /* This is equivalent to rand() % max, but uses the high-order bits
1720 for better randomness on architecture where rand() is implemented
1721 using a simple congruential generator. */
1723 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1724 return (int)bounded;
1728 /* A debugging function for checking whether an MD5 library works. */
1730 #include "gen-md5.h"
1733 debug_test_md5 (char *buf)
1735 unsigned char raw[16];
1736 static char res[33];
1740 ALLOCA_MD5_CONTEXT (ctx);
1743 gen_md5_update ((unsigned char *)buf, strlen (buf), ctx);
1744 gen_md5_finish (ctx, raw);
1751 *p2++ = XDIGIT_TO_xchar (*p1 >> 4);
1752 *p2++ = XDIGIT_TO_xchar (*p1 & 0xf);