Re: [PATCH] binary-io: do not treat set_binary_mode() on stdin/out/err as an error on OS/2

[Bruno Haible]

KO Myung-Hun wrote:
> * lib/binary-io.c (__gl_setmode_check) [__EMX__]: Remove __EMX__ guard.
> * lib/binary-io.h (__gl_setmode_check) [__EMX__]: Remove __EMX__ guard.
> (set_binary_mode) [__EMX__]: Override mode with O_TEXT if tty.

According to the EMX documentation of the setmode function
------------------------------------------------------------------------------
#include <io.h>                                                           [PC]
#include <fcntl.h>

int setmode (int handle, int mode);

    Change the text/binary mode of a file handle.  MODE must be either
    O_BINARY or O_TEXT.

    Note: Use _fsetmode() to change the mode of a stream.

  Return value:

    If there's an error, setmode() returns -1 and sets errno to EBADF
    or EINVAL otherwise setmode() returns the previous mode, that is,
    O_BINARY or O_TEXT.

  See also: _fsetmode(), open()
------------------------------------------------------------------------------

your proposed patch has the following effect:


BEFORE:
                                             sets mode to   return value
set_binary_mode on tty                        ---           -1, errno=EINVAL
set_binary_mode on regular file fails         ---           -1, errno=EBADF or 
EINVAL
set_binary_mode on regular file succeeds      mode          previous mode

AFTER:

                                             sets mode to   return value
set_binary_mode on tty                        O_TEXT        previous mode
set_binary_mode on regular file fails         ---           -1, errno=EBADF or 
EINVAL
set_binary_mode on regular file succeeds      mode          previous mode


Changing a function to *silently* do a different thing than what it was
requested to do? This is not good. The previous behaviour, to set an error
indicator, is better.

> Setting stdin/out/err to binary mode is allowed on OS/2. But it's not
> useful, because it generates stair-output hard to read.

On a tty, the existing code will refrain from invoking setmode. So no
stair-output in this case.

On a regular file, it depends on your text viewer. Emacs surely doesn't
produce stair-cased rendering of a file.

> Instead, let's set them to text mode all the time.
> 
> This fixes that tee always crashes at startup if stdin or stdout
> are not piped.

1) I don't see why the proposed patch would fix a crash. It only
   changes the way output is rendered outside of the program.
2) The proper place to do such changes is the 'tee' program.

Bruno