Re: Hurd translators on BSD

[Sergiu Ivanov]

Hello,

On Tue, Oct 27, 2009 at 01:55:39AM +0100, address@hidden wrote:
> On Thu, Sep 24, 2009 at 08:30:29PM +0300, Sergiu Ivanov wrote:
> 
> > I wonder whether it is reasonable to try FUSE to implement
> > translators.
> 
> I mentioned this as one possibility in my original HELL blog post:
> 
>    http://tri-ceps.blogspot.com/2005/09/welcome-to-hell.html

Ah, sure!  I couldn't remember where I took this idea from.
 
> However, I was a bit sceptical about that from the beginning; and with
> the greater understanding I have now, I'm pretty sure that implementing
> translators on top of FUSE isn't realistic.
> 
> While offering a FUSE interface might be useful for interaction with
> non-Hurd programs, the communication among the Hurdish components
> requires a different interface I believe, that can directly map generic
> Hurd RPCs and the distributed VFS concept.

Yes, I do agree with this.  The idea of translators on FUSE initially
came on my mind as an attempt to attract people to the hurdish way of
thinking, but I realize now that having a single simple translator
ported to FUSE is insufficient.  An important feature to the hurdish
way of thinking as I see it is modularity, and talking about
modularity in the context of one translator is, of course,
meaningless, while trying to port a translator *infrastructure* will
require implementing this ``different interface'' you are talking
about.
 
> > I've never been much into FUSE, but the Wikipedia article says that
> > ``FUSE is particularly useful for writing virtual file systems'',
> > which, AIUI, is precisely the matter of interest for most translators.
> > (At least FUSE filesystems like GmailFS, WikipediaFS, archivemount
> > sound a lot like actual translator names.)
> 
> What surprises me is that although FUSE has been around for a couple of
> years now, and many many Hurd-style FUSE modules are available by now
> for all kinds of things, the idea of such virtual filesystems still
> doesn't seem to have arrived in most people's minds...
> 
> At a PostgreSQL presentation I saw earlier this year, one of the things
> shown was a FUSE module giving access to the database objects. For me as
> a Hurd developer this was the most normal thing in the world; but the
> rest of the audience was completely amazed -- it seems the idea that
> filesystems can be used like that never appeared to them before...

Very sad to hear such things indeed :-(
 
> Well, at least it's no wonder we were never able to sell the idea of
> virtual filesystems in the Hurd -- considering that apparently the
> concept is still not sellable now, although it's widely available in
> Linux, including countless useful examples... :-(

I think the problem is that the concept of a virtual filesystem does
not pass the ``What do I need this for?'' barrier in the minds of the
majority.  Unfourtunately, many people would just consider an xmlfs
translator useless once they have a fancy (graphical, of course) XML
editor.  When having problems, they would ask on forums and, although
the solution they be suggested may employ the concept of a virtual
filesystem, they won't normally tire themselves with understanding
what they did.

Regards,
scolobb