Fiasco, the OSKit, and C++ (was Re: Moving forward)

[Michael Hohmuth]

L4-Hurd people,

I just returned from SOSP, only to find 50+ l4-hurd-related messages
in my mailbox.  I am slowly catching up and I am going to comment in
two or three messages.  This message is related to the Fiasco
microkernel.

Let me begin by stating that I am perfectly happy with you guys using
L4KA's Hazelnut kernel.  It is the sensible thing to do if you are
more comfortable with that kernel.  However, as the main author of
Fiasco, I want to comment on problems people had with the Fiasco
kernel before these reports develop into some sort of myth.

OSKit dependencies:
-------------------

Ian Duggan <address@hidden> writes:
> 2) Having multiple oskits in the fiasco modules is annoying. How far is
> fiasco from being able to use a single oskit version?

Fiasco itself depends only on the OSKit 0.6.  Some other L4 packages
from Dresden also depend on the OSKit 0.6 (most notably, this applies
to Sigma0 and RMGR -- including L4KA's version of RMGR, which comes
with OSKit binaries for convenience), while most other packages depend
on newer OSKits (0.96 or later).  Converting the packages that depend
on the old OSKit to using the new one is not a technical problem.

However, there is one small political problem: We would like the
Fiasco microkernel (and only the kernel, not the packages) to remain
independent from the new OSKit.  That is, if Fiasco was to use the new
OSKit, it would need to do this through an interface layer that is
compatible with the OSKit 0.6.  Alternatively, Fiasco could become
independent of the OSKit.  Neither of these options is a big deal
technically.

The political reason is that we must maintain a version of Fiasco that
we can relicense (i.e., one that doesn't contain GPL'd code we don't
own), and the smaller the differences between the main, free version
and the relicensable version, the better we are off.  As the old OSKit
was licensed under the BSD license but the new OSKit is GPL'd, we
haven't switched yet.

However, Fiasco's user community is an equally important factor to us,
and should we receive important enhancements for the free version of
Fiasco, we will let the two versions diverge.

Build problems:
---------------

Farid Hajji <address@hidden> writes:
> [Ian Duggan <address@hidden> writes:]
> > 2) Having multiple oskits in the fiasco modules is annoying. How far is
> > fiasco from being able to use a single oskit version?
> 
> Hmmm... personally, I'd prefer that we start with Hazelnut. It's smaller
> and you can build it without oskit/oskit10 support at all. I've played
> both with fiasco and hazelnut, but I've found that using the environment
> of hazelnut on my {Open,Net,Free}BSD systems is somewhat easier.

I regularly build Fiasco on FreeBSD.  Most of the time, it works out
of the box.  I have to admit, though, that packaging has not been one
of Fiasco's (or in general, Dresden's) strengths.  Your criticism is
warranted, and we will try to improve the current status.

Again, please keep in mind that L4KA's RMGR also needs the OSKit, but
comes with a binary version of some OSKit libraries for convenience.

C++, Fiasco's implementation language:
--------------------------------------

Jeroen Dekkers <address@hidden> writes:
> I really dislike fiaco because it is written in C++, IMHO C is better for
> a microkernel and the GNU Coding Standards also recommends using C. Hazelnut
> uses a few bytes of C++ which are easy to remove AFAICS, so it's a
> better choice if you ask me. Writing an own L4 variant is also option.

In my experience, students have been able to read, understand and
contribute to Fiasco's source code.  Have you looked at a recent
version (downloaded from the CVS)?

Please tell us why you dislike anything written in C++ and why a
microkernel should be written in C in your opinion.  This is not
intended as flaim bait, but I am genuinely interested in your
thoughts.

Reporting bugs:
---------------

Ian Duggan <address@hidden> writes:
> [Using Dresden's modules unmodified] makes sense too. I was thinking
> it'd be easier if we could grab everything that we needed from a
> single CVS. I'm not expecing a lot of changes either. The only one I
> can think of offhand is a patch to allow fiasco to boot with more
> that 128MB of memory installed.

Please report bugs in Fiasco and in Dresden's L4 packages to our
bug-tracking system at <address@hidden>.  You can
visit the bug tracker at
<URL:http://os.inf.tu-dresden.de/~bugs/fiasco/>.

Regards,
Michael
-- 
address@hidden, address@hidden
http://home.pages.de/~hohmuth/