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 /* HACK: expose save_log_p from log.c, so we can turn it off in
88 order to prevent saving the log. Saving the log is dangerous
89 because logprintf() and logputs() can call malloc(), so this
90 could infloop. When logging is turned off, infloop can no longer
93 #### This is no longer really necessary because the new routines
94 in log.c cons only if the line exceeds eighty characters. But
95 this can come at the end of a line, so it's OK to be careful.
97 On a more serious note, it would be good to have a
98 log_forced_shutdown() routine that exposes this cleanly. */
99 extern int save_log_p;
102 logprintf (LOG_ALWAYS, _("%s: %s: Not enough memory.\n"), exec_name, what);
106 /* These functions end with _real because they need to be
107 distinguished from the debugging functions, and from the macros.
110 If memory debugging is not turned on, wget.h defines these:
112 #define xmalloc xmalloc_real
113 #define xrealloc xrealloc_real
114 #define xstrdup xstrdup_real
117 In case of memory debugging, the definitions are a bit more
118 complex, because we want to provide more information, *and* we want
119 to call the debugging code. (The former is the reason why xmalloc
120 and friends need to be macros in the first place.) Then it looks
123 #define xmalloc(a) xmalloc_debug (a, __FILE__, __LINE__)
124 #define xfree(a) xfree_debug (a, __FILE__, __LINE__)
125 #define xrealloc(a, b) xrealloc_debug (a, b, __FILE__, __LINE__)
126 #define xstrdup(a) xstrdup_debug (a, __FILE__, __LINE__)
128 Each of the *_debug function does its magic and calls the real one. */
131 # define STATIC_IF_DEBUG static
133 # define STATIC_IF_DEBUG
136 STATIC_IF_DEBUG void *
137 xmalloc_real (size_t size)
139 void *ptr = malloc (size);
145 STATIC_IF_DEBUG void *
146 xrealloc_real (void *ptr, size_t newsize)
150 /* Not all Un*xes have the feature of realloc() that calling it with
151 a NULL-pointer is the same as malloc(), but it is easy to
154 newptr = realloc (ptr, newsize);
156 newptr = malloc (newsize);
158 memfatal ("realloc");
162 STATIC_IF_DEBUG char *
163 xstrdup_real (const char *s)
169 copy = malloc (l + 1);
172 memcpy (copy, s, l + 1);
173 #else /* HAVE_STRDUP */
177 #endif /* HAVE_STRDUP */
184 /* Crude home-grown routines for debugging some malloc-related
187 * Counting the number of malloc and free invocations, and reporting
188 the "balance", i.e. how many times more malloc was called than it
189 was the case with free.
191 * Making malloc store its entry into a simple array and free remove
192 stuff from that array. At the end, print the pointers which have
193 not been freed, along with the source file and the line number.
194 This also has the side-effect of detecting freeing memory that
197 Note that this kind of memory leak checking strongly depends on
198 every malloc() being followed by a free(), even if the program is
199 about to finish. Wget is careful to free the data structure it
200 allocated in init.c. */
202 static int malloc_count, free_count;
208 } malloc_debug[100000];
210 /* Both register_ptr and unregister_ptr take O(n) operations to run,
211 which can be a real problem. It would be nice to use a hash table
212 for malloc_debug, but the functions in hash.c are not suitable
213 because they can call malloc() themselves. Maybe it would work if
214 the hash table were preallocated to a huge size, and if we set the
215 rehash threshold to 1.0. */
217 /* Register PTR in malloc_debug. Abort if this is not possible
218 (presumably due to the number of current allocations exceeding the
219 size of malloc_debug.) */
222 register_ptr (void *ptr, const char *file, int line)
225 for (i = 0; i < ARRAY_SIZE (malloc_debug); i++)
226 if (malloc_debug[i].ptr == NULL)
228 malloc_debug[i].ptr = ptr;
229 malloc_debug[i].file = file;
230 malloc_debug[i].line = line;
236 /* Unregister PTR from malloc_debug. Abort if PTR is not present in
237 malloc_debug. (This catches calling free() with a bogus pointer.) */
240 unregister_ptr (void *ptr)
243 for (i = 0; i < ARRAY_SIZE (malloc_debug); i++)
244 if (malloc_debug[i].ptr == ptr)
246 malloc_debug[i].ptr = NULL;
252 /* Print the malloc debug stats that can be gathered from the above
253 information. Currently this is the count of mallocs, frees, the
254 difference between the two, and the dump of the contents of
255 malloc_debug. The last part are the memory leaks. */
258 print_malloc_debug_stats (void)
261 printf ("\nMalloc: %d\nFree: %d\nBalance: %d\n\n",
262 malloc_count, free_count, malloc_count - free_count);
263 for (i = 0; i < ARRAY_SIZE (malloc_debug); i++)
264 if (malloc_debug[i].ptr != NULL)
265 printf ("0x%08ld: %s:%d\n", (long)malloc_debug[i].ptr,
266 malloc_debug[i].file, malloc_debug[i].line);
270 xmalloc_debug (size_t size, const char *source_file, int source_line)
272 void *ptr = xmalloc_real (size);
274 register_ptr (ptr, source_file, source_line);
279 xfree_debug (void *ptr, const char *source_file, int source_line)
281 assert (ptr != NULL);
283 unregister_ptr (ptr);
288 xrealloc_debug (void *ptr, size_t newsize, const char *source_file, int source_line)
290 void *newptr = xrealloc_real (ptr, newsize);
294 register_ptr (newptr, source_file, source_line);
296 else if (newptr != ptr)
298 unregister_ptr (ptr);
299 register_ptr (newptr, source_file, source_line);
305 xstrdup_debug (const char *s, const char *source_file, int source_line)
307 char *copy = xstrdup_real (s);
309 register_ptr (copy, source_file, source_line);
313 #endif /* DEBUG_MALLOC */
315 /* Utility function: like xstrdup(), but also lowercases S. */
318 xstrdup_lower (const char *s)
320 char *copy = xstrdup (s);
327 /* Return a count of how many times CHR occurs in STRING. */
330 count_char (const char *string, char chr)
334 for (p = string; *p; p++)
340 /* Copy the string formed by two pointers (one on the beginning, other
341 on the char after the last char) to a new, malloc-ed location.
344 strdupdelim (const char *beg, const char *end)
346 char *res = (char *)xmalloc (end - beg + 1);
347 memcpy (res, beg, end - beg);
348 res[end - beg] = '\0';
352 /* Parse a string containing comma-separated elements, and return a
353 vector of char pointers with the elements. Spaces following the
354 commas are ignored. */
356 sepstring (const char *s)
370 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
371 res[i] = strdupdelim (p, s);
374 /* Skip the blanks following the ','. */
382 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
383 res[i] = strdupdelim (p, s);
388 /* Return pointer to a static char[] buffer in which zero-terminated
389 string-representation of TM (in form hh:mm:ss) is printed.
391 If TM is non-NULL, the current time-in-seconds will be stored
394 (#### This is misleading: one would expect TM would be used instead
395 of the current time in that case. This design was probably
396 influenced by the design time(2), and should be changed at some
397 points. No callers use non-NULL TM anyway.) */
400 time_str (time_t *tm)
402 static char output[15];
404 time_t secs = time (tm);
408 /* In case of error, return the empty string. Maybe we should
409 just abort if this happens? */
413 ptm = localtime (&secs);
414 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
418 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
421 datetime_str (time_t *tm)
423 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
425 time_t secs = time (tm);
429 /* In case of error, return the empty string. Maybe we should
430 just abort if this happens? */
434 ptm = localtime (&secs);
435 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
436 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
437 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
441 /* The Windows versions of the following two functions are defined in
446 fork_to_background (void)
449 /* Whether we arrange our own version of opt.lfilename here. */
454 opt.lfilename = unique_name (DEFAULT_LOGFILE);
466 /* parent, no error */
467 printf (_("Continuing in background.\n"));
469 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
472 /* child: keep running */
474 #endif /* not WINDOWS */
476 /* Resolve "." and ".." elements of PATH by destructively modifying
477 PATH. "." is resolved by removing that path element, and ".." is
478 resolved by removing the preceding path element. Leading and
479 trailing slashes are preserved.
481 Return non-zero if any changes have been made.
483 For example, "a/b/c/./../d/.." will yield "a/b/". More exhaustive
484 test examples are provided below. If you change anything in this
485 function, run test_path_simplify to make sure you haven't broken a
488 A previous version of this function was based on path_simplify()
489 from GNU Bash, but it has been rewritten for Wget 1.8.1. */
492 path_simplify (char *path)
498 ++path; /* preserve the leading '/'. */
501 end = p + strlen (p) + 1; /* position past the terminating zero. */
506 /* P should point to the beginning of a path element. */
508 if (*p == '.' && (*(p + 1) == '/' || *(p + 1) == '\0'))
510 /* Handle "./foo" by moving "foo" two characters to the
515 memmove (p, p + 2, end - p);
526 else if (*p == '.' && *(p + 1) == '.'
527 && (*(p + 2) == '/' || *(p + 2) == '\0'))
529 /* Handle "../foo" by moving "foo" one path element to the
531 char *b = p; /* not p-1 because P can equal PATH */
533 /* Backtrack by one path element, but not past the beginning
542 /* Move backwards until B hits the beginning of the
543 previous path element or the beginning of path. */
544 for (--b; b > path && *(b - 1) != '/'; b--)
551 memmove (b, p + 3, end - (p + 3));
565 /* Remove empty path elements. Not mandated by rfc1808 et
566 al, but empty path elements are not all that useful, and
567 the rest of Wget might not deal with them well. */
577 memmove (p, q, end - q);
582 /* Skip to the next path element. */
583 while (*p && *p != '/')
588 /* Make sure P points to the beginning of the next path element,
589 which is location after the slash. */
596 /* "Touch" FILE, i.e. make its atime and mtime equal to the time
597 specified with TM. */
599 touch (const char *file, time_t tm)
601 #ifdef HAVE_STRUCT_UTIMBUF
602 struct utimbuf times;
603 times.actime = times.modtime = tm;
606 times[0] = times[1] = tm;
609 if (utime (file, ×) == -1)
610 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
613 /* Checks if FILE is a symbolic link, and removes it if it is. Does
614 nothing under MS-Windows. */
616 remove_link (const char *file)
621 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
623 DEBUGP (("Unlinking %s (symlink).\n", file));
626 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
627 file, strerror (errno));
632 /* Does FILENAME exist? This is quite a lousy implementation, since
633 it supplies no error codes -- only a yes-or-no answer. Thus it
634 will return that a file does not exist if, e.g., the directory is
635 unreadable. I don't mind it too much currently, though. The
636 proper way should, of course, be to have a third, error state,
637 other than true/false, but that would introduce uncalled-for
638 additional complexity to the callers. */
640 file_exists_p (const char *filename)
643 return access (filename, F_OK) >= 0;
646 return stat (filename, &buf) >= 0;
650 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
651 Returns 0 on error. */
653 file_non_directory_p (const char *path)
656 /* Use lstat() rather than stat() so that symbolic links pointing to
657 directories can be identified correctly. */
658 if (lstat (path, &buf) != 0)
660 return S_ISDIR (buf.st_mode) ? 0 : 1;
663 /* Return a unique filename, given a prefix and count */
665 unique_name_1 (const char *fileprefix, int count)
671 filename = (char *)xmalloc (strlen (fileprefix) + numdigit (count) + 2);
672 sprintf (filename, "%s.%d", fileprefix, count);
675 filename = xstrdup (fileprefix);
677 if (!file_exists_p (filename))
686 /* Return a unique file name, based on PREFIX. */
688 unique_name (const char *prefix)
694 file = unique_name_1 (prefix, count++);
698 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
699 are missing, create them first. In case any mkdir() call fails,
700 return its error status. Returns 0 on successful completion.
702 The behaviour of this function should be identical to the behaviour
703 of `mkdir -p' on systems where mkdir supports the `-p' option. */
705 make_directory (const char *directory)
711 /* Make a copy of dir, to be able to write to it. Otherwise, the
712 function is unsafe if called with a read-only char *argument. */
713 STRDUP_ALLOCA (dir, directory);
715 /* If the first character of dir is '/', skip it (and thus enable
716 creation of absolute-pathname directories. */
717 for (i = (*dir == '/'); 1; ++i)
719 for (; dir[i] && dir[i] != '/'; i++)
724 /* Check whether the directory already exists. */
725 if (!file_exists_p (dir))
727 if (mkdir (dir, 0777) < 0)
738 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
739 should be a file name.
741 file_merge("/foo/bar", "baz") => "/foo/baz"
742 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
743 file_merge("foo", "bar") => "bar"
745 In other words, it's a simpler and gentler version of uri_merge_1. */
748 file_merge (const char *base, const char *file)
751 const char *cut = (const char *)strrchr (base, '/');
754 return xstrdup (file);
756 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
757 memcpy (result, base, cut - base);
758 result[cut - base] = '/';
759 strcpy (result + (cut - base) + 1, file);
764 static int in_acclist PARAMS ((const char *const *, const char *, int));
766 /* Determine whether a file is acceptable to be followed, according to
767 lists of patterns to accept/reject. */
769 acceptable (const char *s)
773 while (l && s[l] != '/')
780 return (in_acclist ((const char *const *)opt.accepts, s, 1)
781 && !in_acclist ((const char *const *)opt.rejects, s, 1));
783 return in_acclist ((const char *const *)opt.accepts, s, 1);
785 else if (opt.rejects)
786 return !in_acclist ((const char *const *)opt.rejects, s, 1);
790 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
791 `/something', frontcmp() will return 1 only if S2 begins with
792 `/something'. Otherwise, 0 is returned. */
794 frontcmp (const char *s1, const char *s2)
796 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
800 /* Iterate through STRLIST, and return the first element that matches
801 S, through wildcards or front comparison (as appropriate). */
803 proclist (char **strlist, const char *s, enum accd flags)
807 for (x = strlist; *x; x++)
808 if (has_wildcards_p (*x))
810 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
815 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
822 /* Returns whether DIRECTORY is acceptable for download, wrt the
823 include/exclude lists.
825 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
826 and absolute paths may be freely intermixed. */
828 accdir (const char *directory, enum accd flags)
830 /* Remove starting '/'. */
831 if (flags & ALLABS && *directory == '/')
835 if (!proclist (opt.includes, directory, flags))
840 if (proclist (opt.excludes, directory, flags))
846 /* Match the end of STRING against PATTERN. For instance:
848 match_backwards ("abc", "bc") -> 1
849 match_backwards ("abc", "ab") -> 0
850 match_backwards ("abc", "abc") -> 1 */
852 match_tail (const char *string, const char *pattern)
856 for (i = strlen (string), j = strlen (pattern); i >= 0 && j >= 0; i--, j--)
857 if (string[i] != pattern[j])
859 /* If the pattern was exhausted, the match was succesful. */
866 /* Checks whether string S matches each element of ACCEPTS. A list
867 element are matched either with fnmatch() or match_tail(),
868 according to whether the element contains wildcards or not.
870 If the BACKWARD is 0, don't do backward comparison -- just compare
873 in_acclist (const char *const *accepts, const char *s, int backward)
875 for (; *accepts; accepts++)
877 if (has_wildcards_p (*accepts))
879 /* fnmatch returns 0 if the pattern *does* match the
881 if (fnmatch (*accepts, s, 0) == 0)
888 if (match_tail (s, *accepts))
893 if (!strcmp (s, *accepts))
901 /* Return the location of STR's suffix (file extension). Examples:
902 suffix ("foo.bar") -> "bar"
903 suffix ("foo.bar.baz") -> "baz"
904 suffix ("/foo/bar") -> NULL
905 suffix ("/foo.bar/baz") -> NULL */
907 suffix (const char *str)
911 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
915 return (char *)str + i;
920 /* Read a line from FP and return the pointer to freshly allocated
921 storage. The stoarage space is obtained through malloc() and
922 should be freed with free() when it is no longer needed.
924 The length of the line is not limited, except by available memory.
925 The newline character at the end of line is retained. The line is
926 terminated with a zero character.
928 After end-of-file is encountered without anything being read, NULL
929 is returned. NULL is also returned on error. To distinguish
930 between these two cases, use the stdio function ferror(). */
933 read_whole_line (FILE *fp)
937 char *line = (char *)xmalloc (bufsize);
939 while (fgets (line + length, bufsize - length, fp))
941 length += strlen (line + length);
943 /* Possible for example when reading from a binary file where
944 a line begins with \0. */
947 if (line[length - 1] == '\n')
950 /* fgets() guarantees to read the whole line, or to use up the
951 space we've given it. We can double the buffer
954 line = xrealloc (line, bufsize);
956 if (length == 0 || ferror (fp))
961 if (length + 1 < bufsize)
962 /* Relieve the memory from our exponential greediness. We say
963 `length + 1' because the terminating \0 is not included in
964 LENGTH. We don't need to zero-terminate the string ourselves,
965 though, because fgets() does that. */
966 line = xrealloc (line, length + 1);
970 /* Read FILE into memory. A pointer to `struct file_memory' are
971 returned; use struct element `content' to access file contents, and
972 the element `length' to know the file length. `content' is *not*
973 zero-terminated, and you should *not* read or write beyond the [0,
974 length) range of characters.
976 After you are done with the file contents, call read_file_free to
979 Depending on the operating system and the type of file that is
980 being read, read_file() either mmap's the file into memory, or
981 reads the file into the core using read().
983 If file is named "-", fileno(stdin) is used for reading instead.
984 If you want to read from a real file named "-", use "./-" instead. */
987 read_file (const char *file)
990 struct file_memory *fm;
992 int inhibit_close = 0;
994 /* Some magic in the finest tradition of Perl and its kin: if FILE
995 is "-", just use stdin. */
1000 /* Note that we don't inhibit mmap() in this case. If stdin is
1001 redirected from a regular file, mmap() will still work. */
1004 fd = open (file, O_RDONLY);
1007 fm = xmalloc (sizeof (struct file_memory));
1012 if (fstat (fd, &buf) < 0)
1014 fm->length = buf.st_size;
1015 /* NOTE: As far as I know, the callers of this function never
1016 modify the file text. Relying on this would enable us to
1017 specify PROT_READ and MAP_SHARED for a marginal gain in
1018 efficiency, but at some cost to generality. */
1019 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1020 MAP_PRIVATE, fd, 0);
1021 if (fm->content == (char *)MAP_FAILED)
1031 /* The most common reason why mmap() fails is that FD does not point
1032 to a plain file. However, it's also possible that mmap() doesn't
1033 work for a particular type of file. Therefore, whenever mmap()
1034 fails, we just fall back to the regular method. */
1035 #endif /* HAVE_MMAP */
1038 size = 512; /* number of bytes fm->contents can
1039 hold at any given time. */
1040 fm->content = xmalloc (size);
1044 if (fm->length > size / 2)
1046 /* #### I'm not sure whether the whole exponential-growth
1047 thing makes sense with kernel read. On Linux at least,
1048 read() refuses to read more than 4K from a file at a
1049 single chunk anyway. But other Unixes might optimize it
1050 better, and it doesn't *hurt* anything, so I'm leaving
1053 /* Normally, we grow SIZE exponentially to make the number
1054 of calls to read() and realloc() logarithmic in relation
1055 to file size. However, read() can read an amount of data
1056 smaller than requested, and it would be unreasonably to
1057 double SIZE every time *something* was read. Therefore,
1058 we double SIZE only when the length exceeds half of the
1059 entire allocated size. */
1061 fm->content = xrealloc (fm->content, size);
1063 nread = read (fd, fm->content + fm->length, size - fm->length);
1065 /* Successful read. */
1066 fm->length += nread;
1076 if (size > fm->length && fm->length != 0)
1077 /* Due to exponential growth of fm->content, the allocated region
1078 might be much larger than what is actually needed. */
1079 fm->content = xrealloc (fm->content, fm->length);
1086 xfree (fm->content);
1091 /* Release the resources held by FM. Specifically, this calls
1092 munmap() or xfree() on fm->content, depending whether mmap or
1093 malloc/read were used to read in the file. It also frees the
1094 memory needed to hold the FM structure itself. */
1097 read_file_free (struct file_memory *fm)
1102 munmap (fm->content, fm->length);
1107 xfree (fm->content);
1112 /* Free the pointers in a NULL-terminated vector of pointers, then
1113 free the pointer itself. */
1115 free_vec (char **vec)
1126 /* Append vector V2 to vector V1. The function frees V2 and
1127 reallocates V1 (thus you may not use the contents of neither
1128 pointer after the call). If V1 is NULL, V2 is returned. */
1130 merge_vecs (char **v1, char **v2)
1140 /* To avoid j == 0 */
1145 for (i = 0; v1[i]; i++);
1147 for (j = 0; v2[j]; j++);
1148 /* Reallocate v1. */
1149 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1150 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1155 /* A set of simple-minded routines to store strings in a linked list.
1156 This used to also be used for searching, but now we have hash
1159 /* It's a shame that these simple things like linked lists and hash
1160 tables (see hash.c) need to be implemented over and over again. It
1161 would be nice to be able to use the routines from glib -- see
1162 www.gtk.org for details. However, that would make Wget depend on
1163 glib, and I want to avoid dependencies to external libraries for
1164 reasons of convenience and portability (I suspect Wget is more
1165 portable than anything ever written for Gnome). */
1167 /* Append an element to the list. If the list has a huge number of
1168 elements, this can get slow because it has to find the list's
1169 ending. If you think you have to call slist_append in a loop,
1170 think about calling slist_prepend() followed by slist_nreverse(). */
1173 slist_append (slist *l, const char *s)
1175 slist *newel = (slist *)xmalloc (sizeof (slist));
1178 newel->string = xstrdup (s);
1183 /* Find the last element. */
1190 /* Prepend S to the list. Unlike slist_append(), this is O(1). */
1193 slist_prepend (slist *l, const char *s)
1195 slist *newel = (slist *)xmalloc (sizeof (slist));
1196 newel->string = xstrdup (s);
1201 /* Destructively reverse L. */
1204 slist_nreverse (slist *l)
1209 slist *next = l->next;
1217 /* Is there a specific entry in the list? */
1219 slist_contains (slist *l, const char *s)
1221 for (; l; l = l->next)
1222 if (!strcmp (l->string, s))
1227 /* Free the whole slist. */
1229 slist_free (slist *l)
1240 /* Sometimes it's useful to create "sets" of strings, i.e. special
1241 hash tables where you want to store strings as keys and merely
1242 query for their existence. Here is a set of utility routines that
1243 makes that transparent. */
1246 string_set_add (struct hash_table *ht, const char *s)
1248 /* First check whether the set element already exists. If it does,
1249 do nothing so that we don't have to free() the old element and
1250 then strdup() a new one. */
1251 if (hash_table_contains (ht, s))
1254 /* We use "1" as value. It provides us a useful and clear arbitrary
1255 value, and it consumes no memory -- the pointers to the same
1256 string "1" will be shared by all the key-value pairs in all `set'
1258 hash_table_put (ht, xstrdup (s), "1");
1261 /* Synonym for hash_table_contains... */
1264 string_set_contains (struct hash_table *ht, const char *s)
1266 return hash_table_contains (ht, s);
1270 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1277 string_set_free (struct hash_table *ht)
1279 hash_table_map (ht, string_set_free_mapper, NULL);
1280 hash_table_destroy (ht);
1284 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1291 /* Another utility function: call free() on all keys and values of HT. */
1294 free_keys_and_values (struct hash_table *ht)
1296 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1300 /* Engine for legible and legible_very_long; this function works on
1304 legible_1 (const char *repr)
1306 static char outbuf[128];
1311 /* Reset the pointers. */
1314 /* If the number is negative, shift the pointers. */
1320 /* How many digits before the first separator? */
1321 mod = strlen (inptr) % 3;
1323 for (i = 0; i < mod; i++)
1324 *outptr++ = inptr[i];
1325 /* Now insert the rest of them, putting separator before every
1327 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1329 if (i % 3 == 0 && i1 != 0)
1331 *outptr++ = inptr[i1];
1333 /* Zero-terminate the string. */
1338 /* Legible -- return a static pointer to the legibly printed long. */
1343 /* Print the number into the buffer. */
1344 number_to_string (inbuf, l);
1345 return legible_1 (inbuf);
1348 /* Write a string representation of NUMBER into the provided buffer.
1349 We cannot use sprintf() because we cannot be sure whether the
1350 platform supports printing of what we chose for VERY_LONG_TYPE.
1352 Example: Gcc supports `long long' under many platforms, but on many
1353 of those the native libc knows nothing of it and therefore cannot
1356 How long BUFFER needs to be depends on the platform and the content
1357 of NUMBER. For 64-bit VERY_LONG_TYPE (the most common case), 24
1358 bytes are sufficient. Using more might be a good idea.
1360 This function does not go through the hoops that long_to_string
1361 goes to because it doesn't aspire to be fast. (It's called perhaps
1362 once in a Wget run.) */
1365 very_long_to_string (char *buffer, VERY_LONG_TYPE number)
1370 /* Print the number backwards... */
1373 buffer[i++] = '0' + number % 10;
1378 /* ...and reverse the order of the digits. */
1379 for (j = 0; j < i / 2; j++)
1382 buffer[j] = buffer[i - 1 - j];
1383 buffer[i - 1 - j] = c;
1388 /* The same as legible(), but works on VERY_LONG_TYPE. See sysdep.h. */
1390 legible_very_long (VERY_LONG_TYPE l)
1393 /* Print the number into the buffer. */
1394 very_long_to_string (inbuf, l);
1395 return legible_1 (inbuf);
1398 /* Count the digits in a (long) integer. */
1400 numdigit (long number)
1408 while ((number /= 10) > 0)
1413 #define ONE_DIGIT(figure) *p++ = n / (figure) + '0'
1414 #define ONE_DIGIT_ADVANCE(figure) (ONE_DIGIT (figure), n %= (figure))
1416 #define DIGITS_1(figure) ONE_DIGIT (figure)
1417 #define DIGITS_2(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_1 ((figure) / 10)
1418 #define DIGITS_3(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_2 ((figure) / 10)
1419 #define DIGITS_4(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_3 ((figure) / 10)
1420 #define DIGITS_5(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_4 ((figure) / 10)
1421 #define DIGITS_6(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_5 ((figure) / 10)
1422 #define DIGITS_7(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_6 ((figure) / 10)
1423 #define DIGITS_8(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_7 ((figure) / 10)
1424 #define DIGITS_9(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_8 ((figure) / 10)
1425 #define DIGITS_10(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_9 ((figure) / 10)
1427 /* DIGITS_<11-20> are only used on machines with 64-bit longs. */
1429 #define DIGITS_11(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_10 ((figure) / 10)
1430 #define DIGITS_12(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_11 ((figure) / 10)
1431 #define DIGITS_13(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_12 ((figure) / 10)
1432 #define DIGITS_14(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_13 ((figure) / 10)
1433 #define DIGITS_15(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_14 ((figure) / 10)
1434 #define DIGITS_16(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_15 ((figure) / 10)
1435 #define DIGITS_17(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_16 ((figure) / 10)
1436 #define DIGITS_18(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_17 ((figure) / 10)
1437 #define DIGITS_19(figure) ONE_DIGIT_ADVANCE (figure); DIGITS_18 ((figure) / 10)
1439 /* Print NUMBER to BUFFER in base 10. This should be completely
1440 equivalent to `sprintf(buffer, "%ld", number)', only much faster.
1442 The speedup may make a difference in programs that frequently
1443 convert numbers to strings. Some implementations of sprintf,
1444 particularly the one in GNU libc, have been known to be extremely
1445 slow compared to this function.
1447 Return the pointer to the location where the terminating zero was
1448 printed. (Equivalent to calling buffer+strlen(buffer) after the
1451 BUFFER should be big enough to accept as many bytes as you expect
1452 the number to take up. On machines with 64-bit longs the maximum
1453 needed size is 24 bytes. That includes the digits needed for the
1454 largest 64-bit number, the `-' sign in case it's negative, and the
1455 terminating '\0'. */
1458 number_to_string (char *buffer, long number)
1463 #if (SIZEOF_LONG != 4) && (SIZEOF_LONG != 8)
1464 /* We are running in a strange or misconfigured environment. Let
1465 sprintf cope with it. */
1466 sprintf (buffer, "%ld", n);
1467 p += strlen (buffer);
1468 #else /* (SIZEOF_LONG == 4) || (SIZEOF_LONG == 8) */
1476 if (n < 10) { DIGITS_1 (1); }
1477 else if (n < 100) { DIGITS_2 (10); }
1478 else if (n < 1000) { DIGITS_3 (100); }
1479 else if (n < 10000) { DIGITS_4 (1000); }
1480 else if (n < 100000) { DIGITS_5 (10000); }
1481 else if (n < 1000000) { DIGITS_6 (100000); }
1482 else if (n < 10000000) { DIGITS_7 (1000000); }
1483 else if (n < 100000000) { DIGITS_8 (10000000); }
1484 else if (n < 1000000000) { DIGITS_9 (100000000); }
1485 #if SIZEOF_LONG == 4
1486 /* ``if (1)'' serves only to preserve editor indentation. */
1487 else if (1) { DIGITS_10 (1000000000); }
1488 #else /* SIZEOF_LONG != 4 */
1489 else if (n < 10000000000L) { DIGITS_10 (1000000000L); }
1490 else if (n < 100000000000L) { DIGITS_11 (10000000000L); }
1491 else if (n < 1000000000000L) { DIGITS_12 (100000000000L); }
1492 else if (n < 10000000000000L) { DIGITS_13 (1000000000000L); }
1493 else if (n < 100000000000000L) { DIGITS_14 (10000000000000L); }
1494 else if (n < 1000000000000000L) { DIGITS_15 (100000000000000L); }
1495 else if (n < 10000000000000000L) { DIGITS_16 (1000000000000000L); }
1496 else if (n < 100000000000000000L) { DIGITS_17 (10000000000000000L); }
1497 else if (n < 1000000000000000000L) { DIGITS_18 (100000000000000000L); }
1498 else { DIGITS_19 (1000000000000000000L); }
1499 #endif /* SIZEOF_LONG != 4 */
1502 #endif /* (SIZEOF_LONG == 4) || (SIZEOF_LONG == 8) */
1508 #undef ONE_DIGIT_ADVANCE
1530 /* Support for timers. */
1532 #undef TIMER_WINDOWS
1533 #undef TIMER_GETTIMEOFDAY
1536 /* Depending on the OS and availability of gettimeofday(), one and
1537 only one of the above constants will be defined. Virtually all
1538 modern Unix systems will define TIMER_GETTIMEOFDAY; Windows will
1539 use TIMER_WINDOWS. TIMER_TIME is a catch-all method for
1540 non-Windows systems without gettimeofday.
1542 #### Perhaps we should also support ftime(), which exists on old
1543 BSD 4.2-influenced systems? (It also existed under MS DOS Borland
1544 C, if memory serves me.) */
1547 # define TIMER_WINDOWS
1548 #else /* not WINDOWS */
1549 # ifdef HAVE_GETTIMEOFDAY
1550 # define TIMER_GETTIMEOFDAY
1554 #endif /* not WINDOWS */
1557 #ifdef TIMER_GETTIMEOFDAY
1566 #ifdef TIMER_WINDOWS
1567 ULARGE_INTEGER wintime;
1571 /* Allocate a timer. It is not legal to do anything with a freshly
1572 allocated timer, except call wtimer_reset() or wtimer_delete(). */
1575 wtimer_allocate (void)
1577 struct wget_timer *wt =
1578 (struct wget_timer *)xmalloc (sizeof (struct wget_timer));
1582 /* Allocate a new timer and reset it. Return the new timer. */
1587 struct wget_timer *wt = wtimer_allocate ();
1592 /* Free the resources associated with the timer. Its further use is
1596 wtimer_delete (struct wget_timer *wt)
1601 /* Reset timer WT. This establishes the starting point from which
1602 wtimer_elapsed() will return the number of elapsed
1603 milliseconds. It is allowed to reset a previously used timer. */
1606 wtimer_reset (struct wget_timer *wt)
1608 #ifdef TIMER_GETTIMEOFDAY
1610 gettimeofday (&t, NULL);
1611 wt->secs = t.tv_sec;
1612 wt->usecs = t.tv_usec;
1616 wt->secs = time (NULL);
1619 #ifdef TIMER_WINDOWS
1622 GetSystemTime (&st);
1623 SystemTimeToFileTime (&st, &ft);
1624 wt->wintime.HighPart = ft.dwHighDateTime;
1625 wt->wintime.LowPart = ft.dwLowDateTime;
1629 /* Return the number of milliseconds elapsed since the timer was last
1630 reset. It is allowed to call this function more than once to get
1631 increasingly higher elapsed values. */
1634 wtimer_elapsed (struct wget_timer *wt)
1636 #ifdef TIMER_GETTIMEOFDAY
1638 gettimeofday (&t, NULL);
1639 return (t.tv_sec - wt->secs) * 1000 + (t.tv_usec - wt->usecs) / 1000;
1643 time_t now = time (NULL);
1644 return 1000 * (now - wt->secs);
1651 GetSystemTime (&st);
1652 SystemTimeToFileTime (&st, &ft);
1653 uli.HighPart = ft.dwHighDateTime;
1654 uli.LowPart = ft.dwLowDateTime;
1655 return (long)((uli.QuadPart - wt->wintime.QuadPart) / 10000);
1659 /* Return the assessed granularity of the timer implementation. This
1660 is important for certain code that tries to deal with "zero" time
1664 wtimer_granularity (void)
1666 #ifdef TIMER_GETTIMEOFDAY
1667 /* Granularity of gettimeofday is hugely architecture-dependent.
1668 However, it appears that on modern machines it is better than
1674 /* This is clear. */
1678 #ifdef TIMER_WINDOWS
1684 /* This should probably be at a better place, but it doesn't really
1685 fit into html-parse.c. */
1687 /* The function returns the pointer to the malloc-ed quoted version of
1688 string s. It will recognize and quote numeric and special graphic
1689 entities, as per RFC1866:
1697 No other entities are recognized or replaced. */
1699 html_quote_string (const char *s)
1705 /* Pass through the string, and count the new size. */
1706 for (i = 0; *s; s++, i++)
1709 i += 4; /* `amp;' */
1710 else if (*s == '<' || *s == '>')
1711 i += 3; /* `lt;' and `gt;' */
1712 else if (*s == '\"')
1713 i += 5; /* `quot;' */
1717 res = (char *)xmalloc (i + 1);
1719 for (p = res; *s; s++)
1732 *p++ = (*s == '<' ? 'l' : 'g');
1759 /* Determine the width of the terminal we're running on. If that's
1760 not possible, return 0. */
1763 determine_screen_width (void)
1765 /* If there's a way to get the terminal size using POSIX
1766 tcgetattr(), somebody please tell me. */
1769 #else /* TIOCGWINSZ */
1773 if (opt.lfilename != NULL)
1776 fd = fileno (stderr);
1777 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1778 return 0; /* most likely ENOTTY */
1781 #endif /* TIOCGWINSZ */
1785 /* A debugging function for checking whether an MD5 library works. */
1787 #include "gen-md5.h"
1790 debug_test_md5 (char *buf)
1792 unsigned char raw[16];
1793 static char res[33];
1797 ALLOCA_MD5_CONTEXT (ctx);
1800 gen_md5_update ((unsigned char *)buf, strlen (buf), ctx);
1801 gen_md5_finish (ctx, raw);
1808 *p2++ = XDIGIT_TO_xchar (*p1 >> 4);
1809 *p2++ = XDIGIT_TO_xchar (*p1 & 0xf);
1819 /* Debugging and testing support for path_simplify. */
1821 /* Debug: run path_simplify on PATH and return the result in a new
1822 string. Useful for calling from the debugger. */
1826 char *copy = xstrdup (path);
1827 path_simplify (copy);
1832 run_test (char *test, char *expected_result, int expected_change)
1834 char *test_copy = xstrdup (test);
1835 int modified = path_simplify (test_copy);
1837 if (0 != strcmp (test_copy, expected_result))
1839 printf ("Failed path_simplify(\"%s\"): expected \"%s\", got \"%s\".\n",
1840 test, expected_result, test_copy);
1842 if (modified != expected_change)
1844 if (expected_change == 1)
1845 printf ("Expected no modification with path_simplify(\"%s\").\n",
1848 printf ("Expected modification with path_simplify(\"%s\").\n",
1855 test_path_simplify (void)
1858 char *test, *result;
1864 { "foo", "foo", 0 },
1865 { "foo/bar", "foo/bar", 0 },
1866 { "foo///bar", "foo/bar", 1 },
1867 { "foo/.", "foo/", 1 },
1868 { "foo/./", "foo/", 1 },
1869 { "foo./", "foo./", 0 },
1870 { "foo/../bar", "bar", 1 },
1871 { "foo/../bar/", "bar/", 1 },
1872 { "foo/bar/..", "foo/", 1 },
1873 { "foo/bar/../x", "foo/x", 1 },
1874 { "foo/bar/../x/", "foo/x/", 1 },
1875 { "foo/..", "", 1 },
1876 { "foo/../..", "", 1 },
1877 { "a/b/../../c", "c", 1 },
1878 { "./a/../b", "b", 1 }
1882 for (i = 0; i < ARRAY_SIZE (tests); i++)
1884 char *test = tests[i].test;
1885 char *expected_result = tests[i].result;
1886 int expected_change = tests[i].should_modify;
1887 run_test (test, expected_result, expected_change);
1890 /* Now run all the tests with a leading slash before the test case,
1891 to prove that the slash is being preserved. */
1892 for (i = 0; i < ARRAY_SIZE (tests); i++)
1894 char *test, *expected_result;
1895 int expected_change = tests[i].should_modify;
1897 test = xmalloc (1 + strlen (tests[i].test) + 1);
1898 sprintf (test, "/%s", tests[i].test);
1900 expected_result = xmalloc (1 + strlen (tests[i].result) + 1);
1901 sprintf (expected_result, "/%s", tests[i].result);
1903 run_test (test, expected_result, expected_change);
1906 xfree (expected_result);