Hi,
Well... it was the order of the bundles within the auto.start.1... The
telnetd.jar or telnetconsole.jar bundle has to be the first entry (well, for
me), ahead of shell and tui... When I put telnetd.jar / telnetconsole.jar last,
well -- then I get a conflict...
For what it's worth, Craig
From: Craig Phillips
Sent: Thu 6/26/2008 1:10 PM
To: dev@felix.apache.org
Subject: RE: telnetd discussion
Hi,
I'm experiencing something weird in one newly created workspace/project, which
is pretty much a replica of others...
If I install any of the various telnet bundles, from the one below to the
original telnetd to the one-off I recreated from source, set up a sandbox to
run (meaning, bin and bundle and conf and cache directories and all items
thereof), and run felix, here's the odd behavior:
- The bundles all seem to load and resolve, from what I can tell from stdout,
almost... (note I also use SCR)
- Until I hit a keystroke / enter key (such as 'ps'), the services are not
promulgated
- At such time I hit a keystroke / enter key, then the services are propagated
and telnetd gets its shell service and SCR gets busy and any play-hello-world
bundle I have gets it's pax logger and away we go
- If I remove shell.tui from the mix ('er, auto.start.1), all the above is not
necessary... the services load, and telnetd (or telnet.console) and my
hello-world bundles are all up and running, services and the like
Now, I don't see this same behavior in other "sandboxes", just a newly created
project I just set up.. all I have is a hello world in it...
Obviously, I'm not asking anyone to diagnose my workspace or anything... All
I'm asking is whether someone has seen this before and if they remember the
circumstances and what was the fix...
I'm posting / replying to this thread because I think it is apropos to the
discussion -- there is obviously some sort of shell conflict with the shell.tui
bundle, seemingly on the inputStream/outputStream/printStream, but it's beyond
my comprehension ATT...
Trying to narrow down the difference between this stock hello world app I just
created could take me a while...
For what it's worth, Craig
From: Dieter Wimberger
Sent: Wed 6/25/2008 6:10 PM
To: dev@felix.apache.org
Subject: Re: telnetd discussion
Richard, all:
Thought I put the "simple access alternative" together so you can
decide upon facts not ideas.
http://www.karanet.at/~wimpi/felix/org.apache.felix.telnetconsole.jar
Size is close to 15kB, only dependency is the felix shell service
bundle. Port is 6666 by default and may be configured using the system
property "osgi.shell.telnet.port".
(No command history, but BS, DEL and Strg-U work).
Regards,
Dieter
On 25 Jun 2008, at 15:35, Richard S. Hall wrote:
Dieter Wimberger wrote:
Richard, all:
I'd suggest to first take a step back and ask yourselves a question.
As far as I understood from the discussion, you would be looking
for an occasional, simple telnet based remote access to the felix
shell service.
If this is correct, then I wonder whether it really requires a
telnet/SSH2 compliant server with connection management to achieve
this. Actually, taking equinox as an example, it's nothing more
than a simple single connection without any telnet protocol
negotiation happening at all.
So, this is the question I'd suggest to ask yourselves before we
proceed to find a solution:
Do you need something like the simple equinox "telnet" access, or
do you need a "real" telnet/SSH server?
If the answer is "we need a simple access", then I would actually
suggest to hack a simple listener implementation into the glue
bundle, make it "the telnet" bundle and go with it (from my point
of view, telnetd-osgi would be simply an overkill tool for that job).
Good question. Most of my use cases would be the simple variety, but
I wouldn't want to be constrained to it either. I think the point,
for me, is to create a useful tool that is flexible enough to be
used in simple cases as well as more sophisticated ones.
-> richard
Regards,
Dieter
The following is intended as a summary of the recent discussion on
telnetd (most of this analysis is from an email Felix Meschberger
sent me).
Felix, could you please drop me a copy of this analysis? :)
The following bundles are necessary for remote shell access to
Felix:
1. Felix' standard shell bundle (i.e., shell bundle).
2. Dieter's telnetd bundle (i.e., telnetd bundle).
3. Dieter shell-telnetd glue bundle (i.e., glue bundle).
Dieter also mentioned that the telnetd bundle depended on a
commons bundle, but we could easily package this into the telnetd
bundle so that it is self-contained (we can help make this happen
with the maven bundle plugin).
From Felix' analysis, we could simplify creating remote shell
access by having the glue bundle inject a dummy configuration into
telnetd's ManagedServiceFactory so that the Config Admin
dependency could be optional. This all sounds good. (We could even
consider embedding the telnetd stuff directly into the glue
bundle, but that is another discussion.)
Given this setup, we can ponder where should the telnetd and glue
bundle projects reside? The obvious choices are at the Source
Forge telnetd site or at Felix. I think that any combination can
be reasonably argued. Here is my personal take...
I definitely think it makes sense to create a subproject for the
glue bundle at Felix, I am less certain about the telnetd bundle
itself. While I definitely want to support the telnetd bundle, I
am not sure if it really falls into the scope of the Felix project.
I guess the question is, is telnetd a completely generic telnet
implementation that could easily be used outside of OSGi or not?
If so, then it seems like it should be separate from Felix. On the
other hand, if the implementation is somehow tied to OSGi, then it
might make sense to host it at Felix.
Another possibility is that telnetd is generic, but that it has
some sort of wrapper that integrates it into an OSGi environment,
then maybe it makes sense to host the wrapper at Felix, keeping
the generic library at SF.
I would definitely like to see this functionality available. My
mind is open as to how to achieve it, so what does everyone else
think?
-> richard
