[110] in Pthreads mailing list archive
Re: Patches for pthreads-1.60 beta3
daemon@ATHENA.MIT.EDU (NIIBE Yutaka)
Wed Sep 6 00:57:50 1995
Date: Wed, 6 Sep 1995 13:37:35 +0900
From: NIIBE Yutaka <gniibe@mri.co.jp>
To: Greg Hudson <ghudson@MIT.EDU>
Cc: pthreads@MIT.EDU
In-Reply-To: <199509060424.NAA00828@megatherium.mri.co.jp>
NIIBE Yutaka writes:
> What I think is:
> The pthread library always sets __FD_NONBLOCK on the file descripters,
> as it should not block on I/O requesto for pipe/socket/tty.
> However, it unsets __FD_NONBLOCK in fd_table[fd]->flags.
> I guess that the reason is:
> as it is implementation issue, it is
> better that it isn't seen from user's view.
> The users might request non-blocking I/O, this is redundant.
>
> For example, the file descripter is SETFL-ed with __FD_NONBLOCK, but
> corresponding fd_table[fd]->flags is not set in open(), pipe(),
> socket() in fd_kern.c.
>
> Am I wrong?
I understand now. Sorry, I was wrong.
The fd_table[fd]->flags _is_ user's view.
If __FD_NONBLOCK is set in fd_table[fd]->flags, it means user wants
non blocking I/O, so I/O operations can return EAGAIN.
The fix of (1) in my report is wrong.
--
NIIBE Yutaka
