Re: Standard libraries (Was: Environment variables)

[kilgallen@eisner.decus.org (Larry Kilgallen)]

In article <56n7p5$7co@fozzie.sun3.iaf.nl>, geert@fozzie.sun3.iaf.nl (Geert Bosch) writes:

> I wasn't saying that it is not possible to have a POSIX-layer
> (I called it POSIX-emulation) on different platforms, just that the

No, it really is POSIX -- a set of interfaces defined to be capable
of implementation on Unix and non-Unix systems alike.

> abstraction used is not very usable. It is just backwards to use 
> very specific Unix calls which do not represent what you really want
> to do and then have an emulation layer to translate the very specific
> call to the more simple call that the underlying system provides.

If you look at the work various non-Unix vendors had to do to provide
POSIX support, you get an idea of the scope of effort required to get
"common" interfaces across operating systems.  I am not personally a
fan of the POSIX interfaces, but at least it has a chance to be common.

> For Ada it would be preferable to have simple extensions that use
> the current standard library as a basis. Named pipes between processes
> would be established by implementing a special File_Type that
> can be be used as Ada.Streams.Root_Stream_Type'Class. There is a
> reason that this is suggested in RM95 13.13(1). Certainly using the
> Unix API standardized by POSIX is not EXACTLY the same.

For an Ada vendor to implement named pipes without support from the
OS vendor seems to me highly unlikely to be efficient, secure and
correct.  I would prefer Ada implementors spent their time on the
quality of debugger support and interfaces to OS constructs which
are already well supported.

At some point, activities are just not portable.  I have seen many
programs perform horribly after porting because someone neglected
the _fact_ that an operation which is quite efficient on one
operating system may well be horribly inefficient on another.

Larry Kilgallen