Re: So... is this dead?

[EBo]

On Tue, 28 Aug 2012 07:50:22 -0700 (PDT), Josh Leverette wrote:
> Is this dead?
> Just wondering... I've found traces of life on the IRC channel -- I
> think,
> but the website hasn't been updated in 3 or so years and this page
> hasn't
> had activity in a couple of years either. It sounds like an
> interesting
> project, so I'm curious.

speaking for myself I moved on. I have not kept up with anyone else
that might be playing with it.

I did port 9vx to the TinyCoreLinux distro, but stopped banging my head
against that wall because someone kept making changes to the control
files which broke my maintenance tool chain. I never did figure out who
was trying to "help" maintain the system without ever telling me. I
could have showed them an easier way that would not have wasted weeks of
my time -- till I just gave up.

EBo --

[Anant Narayanan]

I haven't touched the project in years, but am more than happy to help
anyone who might be interested in working on it!

Cheers,
-Anant

> --
> To post to this group, send email to gle...@googlegroups.com
> To unsubscribe from this group, send email to
> glendix+u...@googlegroups.com
> For more options, visit this group at
> http://groups.google.com/group/glendix?hl=en

[Ericson2314]

Hello,

I don't have the time just yet, but I am interested in working on Glendix in the near future. Exactly how much has yet to be done?

Part of my interest is because I am wondering whether rio could be implemented on top of Wayland, he upcoming "replacement" for X. Wayland isn't particularly like rio, but neither is X, and building off wayland should bring in less POSIX or other annoying dependencies. Also Wayland biggest drawback is the lack of networking transparency. The devs have decided to push that problem down the road. Implementing rio on Wayland should at the very least inspire ideas for remote graphics and Wayland, and possibly interest other developers in Plan9 and Glendix.

Thanks,

[Anant Narayanan]

Hey,

We've only implemented the most important system calls for now (read, write, open, clone and couple others), in order to get Rio working, I'm assuming you will have to implement about a dozen other system calls, and then implement /dev/draw with a Wayland backend. We have an early implementation of /net that you can use as reference while implementing /dev/draw.

Cheers,

-Anant

[Ericson2314]

Wow, thanks for the quick reply. I saw the branch with /net in the repository, that certainly will help.

Anyways, I am curious what sort of broad issues to look out for. Is there any sort of hitch or new viewpoint that made lead you to stop working on the project when you did? Glendix seems like it could bring a huge amount of energy into Plan9, but If I am going to put some serious effort into it I want to try to make sure I know what's coming and avoid any nasty surprises.

I guess as similar questions, if all 51 syscalls were implemented right now, what would the next challenge be? In native plan9, how to what extent are the special filesystems like /net implemented on those system calls? does taking advantage of Linux's existing drivers mean making higher abstractions beyond the syscalls and/or filesystems?

Thanks again,
Ericson2314

[Anant Narayanan]

As the old adage goes, "it's software, anything is possible"! There
are certainly no technical restrictions as to why one can't make rio
work with a Wayland backend. It might be hard, but not impossible.

Once we got a few basic Plan 9 programs working, it was a combination
of the lack of time and motivation that lead to no progress. Also,
vx32/9vx was released around the same time which gives you most of
what Glendix does. If you haven't looked at 9vx yet, you certainly
should: http://swtch.com/9vx/.

Rio doesn't have any dependencies other than (a subset of) system
calls, /dev/draw, and possibly a few other file systems. The most
efficient way to implement those file systems would be for them to
directly talk to the backend (in this particular case, that would be
Wayland). /net went a step further and implemented the system in
kernel space, but it can also be done from user space, the choice is
ultimately yours to make. One thing that tripped us up was the
per-process namespace that Plan 9 has, which isn't trivial to map to
the Linux world (which has namespaces but they're designed
differently).

Don't hesitate to post to the list with specific issues, the 9fans
list can also be very helpful at times.

All the best,
-Anant

[Ericson2314]

Thanks for the details. I have only done userspace C so there's probably tons of stuff lets less obvious to me. It looks like user namespaces were the your biggest hurdle, and even if I were just to try to do rio AFIAK /dev/draw and the other special filesystems heavily depend on user namespaces. Fortunately I think some progress might be made:

The most recent article I could find is this (April)
https://lwn.net/Articles/491310/

and the git repository (last updated in May)
https://git.kernel.org/?p=linux/kernel/git/ebiederm/user-namespace.git

I'm not quite sure, but It's possible the patches might be pretty complete, and they are being withheld from mainline because of other reasons.

[Jens Staal]

2012/10/20 Ericson2314 <erics...@yahoo.com>:

> --
> To post to this group, send email to gle...@googlegroups.com
> To unsubscribe from this group, send email to
> glendix+u...@googlegroups.com
> For more options, visit this group at
> http://groups.google.com/group/glendix?hl=en

One thing that might interest you is the devwsys project
http://summerofdevdraw.blogspot.be/
https://bitbucket.org/yiyus/devwsys-prev

It would definitely be interesting to have a Wayland port of this,
since it would offer an upgrade path for plan9port, vx32, Inferno etc
to Wayland.

A proper file system interface for network and GUI in Linux would be
neat however :) There is a paper on remote desktop solutions using the
9P protocol on Linux and I think a lot of resource sharing betwen
virtualized OSes and host happens via 9P2000 - so there would
definitely be a use for it.

Might be interesting for the Wayland people now when they are moving
away from the X11 network transparency.