(IF_READDIR_NEEDS_REWINDDIR): Define.
[! HAVE_WORKING_READDIR] (remove_cwd_entries): If readdir has just
returned NULL and there has been at least one successful unlink or
rmdir call since the opendir or previous rewinddir, then call
rewinddir and reiterate the loop.
There's a bug in Darwin6.5's readdir that shows up only with
338 or more files.
Fix a bug in this test: `cd $pwd' (not to `..'), now that $tmp
has two components.
(dump_remainder): Move two declarations `down' into the scope
where they are used.
(xlseek): Return the resulting offset.
(file_lines): Rename parameter, file_length, to end_pos.
(pipe_lines): Don't coerce safe_read return value to `int'.
Adapt tests accordingly.
(pipe_bytes) [struct charbuffer] (nbytes): Change type from `int'
to `unsigned int'.
Change type of `total_bytes' from `int' to `size_t',
since the former wouldn't always be wide enough.
Don't coerce safe_read return value to `int',
and adapt tests accordingly.
Now that testing for a read error no longer involves
using `tmp', handle that case *after* freeing `tmp'.
(start_bytes): Clean up.
(tail_bytes): Now that `n_bytes' may be larger than
OFF_T_MAX, test for that condition and, if it's true, don't
use lseek optimizations.
(parse_options): Don't fail just because N_UNITS is larger than
the maximum size of a file -- tail may be applied to an input
stream (e.g., a pipe) with more data than that.
is not defined, don't run the test, and don't use the wrapper.
Otherwise, on the Hurd, it would take a long time to create
and remove a hierarchy about 4000 levels deep.
Based on a patch from Robert Millan.
<http://mail.gnu.org/archive/html/bug-coreutils/2003-04/msg00070.html>.
* doc/coreutils.texi (printf invocation): It's \NNN in the format,
\0NNN in the %b operand.
* src/printf.c (usage): Likewise.
(print_esc): New arg OCTAL0 to specify whether \0NNN or \NNN
is desired. All uses changed. Behave like Bash printf if %b
operand uses \NNN where the initial N is not 0.
