Re: The next step of GNU in pure technical perspective

[Olaf Buddenhagen]

Hi,

On Tue, Dec 16, 2014 at 12:19:25AM -0500, Richard Stallman wrote:

> I never read the code of the Hurd, but from what the developers told
> me, it does not directly implement POSIX interfaces.  The C library
> does substantial work in order to provide POSIX semantics on top of
> Hurd interfaces.

The Hurd part of glibc implements the POSIX system calls; the Hurd
servers provide various underlying mechanisms, both for faciliating the
system calls, and for providing the base for a UNIX-like environment in
general. The Hurd servers do not expose POSIX interfaces directly, but
they need to implement a lot of the necessary semantics to allow the
client-side libc to provide POSIX compliance.

Both the libc code and the Hurd servers (especially libports, which is
the major underlying library for all Hurd servers) contain some pretty
intricate code to implement the necessary semantics, such as
cancellation of in-progress system calls during signal processing etc.,
on top of the existing Mach facilities. This code is responsible for
quite a lot of complexity and bugs.

A system constructed in the way the Mach creators envisioned (and not
aiming at POSIX compliance) -- as demonstrated in their Mach-UX
reference system -- is constructed in quite a different, much simpler
fashion. A microkernel explicitly created as the base for a POSIX system
on the other hand would provide different facilities than Mach. The
latter is among the goals of Richard Braun's X15 project.

-antrik-