yyy@xxxxxxxxxxxxxxxxxxx wrote, on Fri 23 Feb 2001:
>
> There can be no implementations on
> conforming POSIX implementations (or UNIX-branded implementations); any
> implementation that tried this currently would not conform to POSIX
> requirements.
This argument is bogus. There is nothing to prevent a conforming
POSIX or UNIX system from implementing this. They would just need
a way of turning it off, e.g. a mount option or a kernel
configuration option. Many systems can be configured to turn
off _POSIX_CHOWN_RESTRICTED, even though doing so makes the system
non-compliant with UNIX95/98. Indeed many branded systems ship
with a non-compliant default configuration.
Alternatively, just don't include the new file system type in the
conformance document, where it lists which file systems can be used
in a conforming environment. I.e. treat it the same as other
non-POSIX file systems such as VFAT.
> The C Standard reserves several classes of identifiers
> for implementation use; this is an attempt to reserve one class of
> names in the file system for implementation use.
The big difference between reserving namespace in a language and
reserving namespace in the file system, is that language namespaces
have no equivalent of things like readdir(), ls, ftw() and find.
Reserving namespace in a language does not cause any problems for
applications. They simply need to make sure they don't use names
that are reserved for the implementation. They have no way to
know what reserved names the implementation is using and there
is no danger of unwanted interaction with such names.
The situation for a file system is completely different. If files
with reserved names are allowed to be visible, applications will
encounter these files during normal use via things like readddir().
My guess is that if anyone ever tries to implement this, they will
discover that it causes so many problems that it is completely
unworkable.
Standards bodies in general try to avoid invention and only
standardize existing practise, and with good reason. There is a
very real danger that invented parts of standards could have
unforeseen difficulties. I see this invention of reserved names
in the file system as particularly dangerous in that regard.
It must not be allowed into the standard unless or until there
are existing implementations.
--
Geoff Clare yyy@xxxxxxxxxxx
UniSoft Limited, London, England. yyy@xxxxxxxxxx
|