New API specification document

[Nick Garnett]

"Michael A. Olson" <mao@sleepycat.com> writes:

Hi Mike, 

> I have a question on the revised spec.
> 
> Several of the interfaces are marked "o" (optional).  Are these optional
> based on the inclusion of a package, like a file system?

The intention here is to indicate interfaces whose inclusion is under
the direct control of the user, rather than those that as a
consequence of some other configuration option. For example, mkfifo()
is only possible when there is a certain amount of file system
infrastructure present, but the user can also opt not to have FIFOs at
all. Does that make sense?

> 
> Also, some of the interfaces are marked "m" (modified semantics possible).
> Generally, it looks like these are intended to provide some basic support
> but maybe not the full POSIX semantics of the corresponding calls (for
> example, printf and friends).  Is that right?  Is there a minimum
> conformance level for these?
> 

That is exactly right. At present we don't really have a clear idea
what the minimum level for these will be (apart from the null option
of returning ENOSYS, but that is not really very useful). To some
extent this will be configurable; for example, the semantics of the
signal functions will vary depending on what signal handling options
are configured. However, at level one it all depends on the
capabilities of the underlying system.

-- 
Nick Garnett
Red Hat, Cambridge, UK