when applied to files in /proc and /sys, even when the FIONREAD
ioctl produces nonsensical results. Before this change, cat would
produce no output (or truncated output), for some linux kernels.
* src/cat.c (write_pending): New function, factored out of cat.
(cat): Also interpret a negative ioctl/FIONREAD count as indicating
that there are bytes to read. Some versions of linux-2.6.16 do that.
Write any pending output before returning.
Reported by Dan Jacobson in <http://bugs.debian.org/370583>.
* NEWS: Mention this bug fix.
* tests/misc/cat-proc: New file. Test for the above.
* tests/misc/Makefile.am (TESTS): Add cat-proc.
link system call cannot create a hard link to a symbolic link.
(copy_internal) [LINK_FOLLOWS_SYMLINKS]: Don't use
the link syscall on a symlink when it would do the wrong thing.
Based on the patch by Aurelien Jarno: <http://bugs.debian.org/329451>
(copy_internal): Don't manually prohibit a move where
the destination is an existing directory. Sometimes doing that is
valid. Let the rename system call enforce the rules. That is
allowed only when the source is a directory and the destination
directory (to be replaced) is empty. Reported by Eric Blake.
choosing sort function; added new sort_numtypes member for
compile-time check.
(time_type): Add new time_numtypes member for compile-time check.
(directories_first): New global variable.
(GROUP_DIRECTORIES_FIRST_OPTION): New enum.
(long_options): Add --directories-first.
(main): Support new option.
(is_directory): New function.
(extract_dirs_from_files): Use it.
(DIRFIRST_CHECK, DEFINE_SORT_FUNCTIONS)
(LIST_SORTFUNCTION_VARIANTS): New macros.
(sort_functions): New global variable.
(sort_files): Use it.
(usage): Document new option.
what must be a typo. Strengthen it to its intended value.
(dopass): Don't use alloca; it's not worth the aggravation here,
since it's used only to get a page-aligned buffer, and page
alignment doesn't buy us much here. I'm suspicious that alloca
causes problems on some hosts, due to a recent bug report by Adam
Waltman.
(print_long_format): Use (new) filemodestring rather than
(old) mode_string, so that we get more file types right, at least
in theory. Adjust to filemode changes.
Include <sys/statvfs.h> and use statvfs only if USE_STATVFS.
(NAMEMAX_FORMAT): define a bit more clearly, now that the
statvfs-using code is a bit more regular.
what POSIX requires. Also, don't let anchors match
newline; this fixes an incompatibility with tradition and with POSIX.
Don't warn about leading ^. POSIX says it is unspecified whether
^ is a special character, which means that implementations can
either treat it as special or not, but either way a warning is not
allowed (unless the regexp is otherwise invalid). Instead, anchor
the expression but treat ^ as an anchor; this is the traditional
behavior (e.g., Solaris 10).
(eval4, eval3, eval2): Treat non-numeric args, division by zero,
and the like as invalid expressions (exit status 2), not as
failure of 'expr' (exit status 3). This is more consistent with
how Solaris behaves.
(context_regex, word_regex): New vars, replacing the above.
All uses changed.
(struct regex_data): New type.
(compile_regex): Renamed from alloc_and_compile_regex, since
we no longer allocate storage. Arg is now a struct regex_data *,
not a const char *. All uses changed. Don't allocate the fastmap;
instead, take it from the caller. Don't convert size_t to int,
to avoid arithmetic overflow problems. Don't bother freeing
storage afterwards; it's not worth the aggravation.
end, since it's large. Change regexpr member from char * to bool;
all uses changed. Add new member fastmap.
(extract_regexp): regexp arg is now char const *, not char *.
Don't bother duplicating the regular expression; it's not needed.
Set fastmap from new fastmap member. Don't bother allocating
a buffer, as the regexp code does a better job than we do.
