On Fri, 4 Feb 2000, Steven Bonneville wrote:

> You sound fairly hostile about XFS.

That's an accurate interpretation.  I'm actually hostile about quite a few
things, of which I somehow feel XFS is a hazily-prototypical example.
Hopefully, I can make good arguments whether I'm hostile or not, and my
hostility will not interfere with judging and considering the points I
make.  Unless, of course, you happen to be an XFS developer. :)

The problem here is that the JFS release is, among other things, an
advertisement for developers.  IBM is trying to convince open-source
developers to work on their code, for free and in their c.f.t., by means
of an advertisement.  The advertisement does not fairly discuss the
technical merits of the filesystem, nor the practicality of making use of
the released code.  It typically does not even discuss the scope of what's
to be released.  Based on this (lack of) information, Slashdot articles
are posted, and developers (and the Hoardes of Advocates) get committed to
the project, usually without investigating alternative works-in-progress
or doing a single search of publicly available literature at the library.  
We all know advertising can interfere with good judgement.  And we could
spend a couple hours discussing the myriad ways these ads do their work
upon potential developers, prototypical ad-victim thought processes, and
so on.  Mainly, I want to point out that these so-called ``release
announcements'' are even worse because they often aren't perceived
consciously enough as the ads that they are.

It's not merely a matter of being dependent upon them.  Dependent or not,
they consume OSS development time, and we don't have a lot of that to
spare.  They make the already bad problem of agenda-focusing in free
software even worse.

> I just think we should be glad that we can learn from a closer look at
> what they did.

There is _so_ much out there to learn, and so many ways of learning it,
that I am somewhat less than convinced that the porting (and rewriting of
encumbered portions) of XFS to Linux is an efficent way for said Linux
developers to learn how to write good filesystems.  And, I don't think
that's _why_ said Linux developers are doing the work, either.

We already have a model for teaching filesystem technology: journal
articles.  If the SGI people actually do have something we can learn from,
why don't they write an article about it?  It's not a tremendous amount of
work (compared to, say, porting an encumbered Irix filesystem to Linux),
and it also gives them publicity.  It's a rather, ahem, proven technique,
in that LFS, softupdates, and probably most of the clever stuff inside XFS
are all based on published articles in technical journals.  If they can't
spare the resources to write an article, I don't see any problem with
leaving XFS's hypothetical pearls of wisdom among the many other things I
will not have time to learn in my lifetime.

AFAIK XFS was released around the same time Konrad started working on
resurrecting LFS.  If such corporate releases are indeed so useful and
informative, how come Konrad has almost single-handedly delivered a
working filesystem, while what is presumably a hoard of Linux developers
haven't even finished acquiring and de-encumbering the XFS code, much less
getting it working? While these people were tricked into working on
proprietary SGI code, NetBSD implemented softupdates and working LFS.  In
the same timeframe, Linux implemented ext3fs and Reiserfs (whichBTW still
don't work with RAID, do they?)?  Why should I believe JFS is going to be
any more useful than XFS?  And why shouldn't I expend some effort trying
to explain why people need to stop falling for the same transparent little
corporate trick every time?

-- 
Miles Nordin / v:+1 720 841-8308 fax:+1 530 579-8680
555 Bryant Street PMB 182 / Palo Alto, CA 94301-1700 / US