Andrew Clegg wrote in a message to Mike Bilow:

 MB> Rockridge is a strict superset of ISO 9660.  IBM should support it in
 MB> CDFS.

 AC> Do you mean, CDFS is supposed to support it already, or IBM
 AC> should change CDFS to support it?

Sorry about the ambiguous phrasing.  What I mean is that there is at
present no code in CDFS to support Rockridge, but that it would be a good
idea for IBM to change CDFS to put it in.

 MB> Linux Second Extended is the one thing that should be implemented for
 MB> OS/2.  It is a well-designed file system that is very similar to HPFS
 MB> in overall structure, but is intended to be driven by a 32-bit CPU. 

 AC> Also it has symbolic and hard links, which I have often
 AC> wished for in HPFS. I know I could use TVFS -- when I get a
 AC> larger hard drive and more memory I probably will -- but
 AC> it's too resource-hungry to make it worthwhile on a hard
 AC> drive as small(ish) as mine.

I have never really played with TVFS.  For whatever reason, the file system
is one of those components which makes me uneasy to play with too much.

 MB> I have suggested several times that it would be a good project for
 MB> someone to do, although I would not have the time for at least
 MB> several months

 AC> I would offer to help but I know about as much about IFS
 AC> programming as the Pope knows about errogenous zones, so I
                                         ^^^
                             (But the Pope is infallible!)

 AC> wouldn't be a lot of help. 

There are not too many people around who know anything about IFS
programming.  It is easier these days, but it took me a long time just to
get the IFS spec from IBM several years ago.  Peter has said that he is
interested in giving a talk on the subject at ColoradOS/2, and I would go
to the talk if I was going to that conference.  However, I wonder if the
two of us might not be the only people in attendance, as both of us have
done IFS work.

The OS/2 IFS uses a completely different architecture from that of a Unix
file system.  There are a lot of reasons for that, mostly historical, that
are not worth going into.  In any case, an IFS is technically a 16-bit Ring
0 DLL that exports a set of reserved names.  The kernel calls into the DLL
using these function names and expects certain behavior.  Numerous
enhancements have been piled onto this basic structure to allow direct
communication and callbacks between an IFS and a DMD, mostly for
performance or security reasons.
 
-- Mike


---