On Wed, 2006-02-22 at 11:11 -0800, Paul Eggert wrote:
> yyyyyy@xxxxxxxxxxxxx writes:
>
> > Are implementation added flags allowed to be added to the set of
> > flags specified by the oflag argument to open() ? For example
> > O_LARGEFILE
> >
> > No specific suggestion at this time. One approach could be to bound
> > the possibilities with specific allowed values.
>
> Can you give a few more details about this proposed approach? It's
> common practice for implementations to add values to the list, e.g.,
> O_DIRECT, O_DIRECTORY, O_NOFOLLOW, and O_FSYNC for GNU/Linux, or
> O_XATTR for Solaris, or O_LARGEFILE and O_NDELAY for both operating
> systems. Surely people aren't thinking of prohibiting extensions like
> these. But it's not clear what sort of restrictions are being
> contemplated.
The intent of the aardvark, as it came out of the AG plenary meeting in
Ottawa this morning, was to explicitly permit implementations to add
such flag values. The current wording in the standard could be read as
prohibiting them, and as you say, it is widespread practice to permit
them!
It is also clearly wrong for an implementation to turn on a bit that IS
specified by the standard (e.g. O_NONBLOCK) ... but what about those not
specified, such as O_LARGEFILE?
We hadn't quite got the right words at the time the aardvark was filed,
and we simply wanted to capture enough of the discussion to allow us to
move on. Sorry if this has caused confusion!
--
Nick Stoughton <yyyy@xxxxxxxxxx>
|