Hi guys,
sorry for not having replied to this mail yesturday, I was busy with
personal stuff.
Thanks Julien for having done this summarize. I have to add that we
didn't discussed all of the items on the list with Julien, but this is a
very good thing that Julien added the important items. We badly need a
roadmap for 2.0, taht's for sure.
Comments in line
Julien Vermillard wrote:
Hi,
next step is to create a wiki page due to the size of the list :)
Certainly.
<snip/>
Code features :
- integrating proxy patch
Don't know what this is about.
https://issues.apache.org/jira/browse/DIRMINA-415
Edouard big patch for supporting proxy such as simple port forwarding
and SOCKS.
It's somewhat a big patch to shallow.
ok. We have to look it closely.
- composite buffers and streams based on it
I definitely want to get involved here but I need to catch up on the
conversations had in the past about it.
We need to discuss this item on the ML, as it will imply a hell lot of
modifications.
- killing IoBuffer
+1
Sure, +1.
- fixing logging filter
+1
I will create a JIRA plus submit at least two proposals (code)
- removing netty2 codec module (who want to use it for Mina 2.0 ?)
+1
+1
- test coverage (yeah painfull at this point)
Yeah test coverage is important but luckily MINA is not that large.
Test coverage makes it easier for people to get into the code to try
to add features or fix a bug while knowing their changes/fixes will
not cause regressions or breakage in the behavior of the API. Plus
it's the best example code to use for how to learn to use different
features of MINA: I always look at test code when learning how to use
a new piece of OSS since the docs may not be there or may not be
totally up to date.
Emmanuel said me exactly the things word for word ;)
We are sharing a lot with Alex, but more important, we have empowered
each other to work better : removing bad habits ("Human habits ..." :)
by learning from the other, and listening to criticisms. Here, we were
on the very same page. It's also something we have gone through as we
are not young bloods... call it experience !
- write traffic throttling
Yes this is a concern of mine as well. I've been experimenting in
the last few days specifically with ApacheDS trying to get her to
behave like a lady (u know not giving everything she's got and
overwhelming clients). I realized we need to support the MINA
community by making the move to MINA 2 ASAP. Was going to do this
last night but was wrestling with Maven issues. We'll probably make
the move shortly and can supply a lot of feedback with respect to
this aspect.
I wrote a little resume here of ADS problem :
http://cwiki.apache.org/MINA/traffic-throttling.html
I don't understand why it's not doable using WriteFuture.
Would be very kind if someone of ADS can take 5 minutes to explain me
where I'm wrong.
I will try to explain the problem we have. May be I'm also missing some
points.
Doc feature :
- testing and perhaps tuning APR based transport
+1
+1
- doc on reqres filter
Have not had the pleasure of looking at this filter at all - don't
know what it's for.
it's for request/response protocols, so when you send a request you can
listen for the response in a non blocking way.
The whole module is doc less.
ok. So +1
- finishing doc on transports (core/APR/serial)
+1
+1
- overall documentation (tutorial, concepts)
+1
We badly need tutorials. Also, a while back, we discussed about buidling
a kind of Ethereal on top of MINA. The biggest advantage of doing so
(having a protocol codec for many well know protocols) is that people
who want to learn about MINA will have real life samples on how to
implement a protocol on top of MINA. Good candidates for such codec are
obviously HTTP, FTP or LDAP (which is quite complex).
- new website, clearer, and with better integration/links to
subprojects like ftpserver and asyncweb
+1
+1 !
- state of the art release tarballs
+1
I think might we have some reusable infrastructure over at Directory
that might help with this and things like deb, pkg, rpm installers.
We're not starting completely from scratch in other words.
Well MINA is a lib, so perhaps releasing linux package is a bit extreme.
Some of the work was already contributed, we just need to find a
workaround for a maven assembly bug.
Yeah, may be not necessary to deliver a rmp/deb. Maybe interesting for
AsyncWeb or FTPServer.
The discussion is open, the idea is to gather wish and ideas, reach
some consensus and clear the list of issue in JIRA.
Personally I'm disappointed by MINA's internals and how complex it is.
Another committer and I were discussing this complexity on IM a while
back. He said, "I have not read a single project that was great that
did not have lucid crazy simple code. It's like what Feynman said
about complex subjects: you don't really understand it. Same goes
for code. If it's complex and obfuscated you don't understand the
problem domain." I whole heartedly agree and there might be a need
for some rearchitecting for the sake of simplification and clarity.
This can be approached in phases with subtle refactorings of the
internals without impacting the API. In general this will improve:
- our ability to respond to bugs,
- requests for new features,
- increase the developer adoption rate with lucid code,
- have better more discrete test cases for internals,
- and better manage and interpret issues when internals are
explicit in what they are designed to do
After realizing that I need to step up, I've started reaclimating
myself with the internals. I could not believe how disorderly it is:
I thought it was in much better shape.
What is amazing, it's the simplicity of the API and the complexity of
the internals, it's sure NIO is not helping us here, but the is a lot
of things to fix/document in core. Emmanuel raised a good example with
LoggingFilter.
The death of IoBuffer will probably be the first big simplification of
the code base.
Damn yes ! IoBuffer represent itself around 15% of the code.
There is another aspect which is cumbersome and complex : the filter
chain. I personnally don't think we need to have such a complex
implementation (the chaining concept is ok). For instance, we have a
HeadFilter and a TailFilter, I don't think we need them. Plus I also
think that we should distinguish between the incoming chain and the
outgoing chain. last, not least, we may need to have more than one
chain, as events may not have the same pathways. This is something we
have to think about.
Another reason why the current implementation sucks, is taht when it
comes to debug it, it's really a PITA.
In the present state, it's
very hard to see people with limited time frames looking at the code
and understanding it in time to actually affect some change. The
barrier is too high but we can fix that one step at a time.
Alex
Yep actually MINA contributors are all hobbyist in the number of hours
allocated, so we need reasonable target.
The current community is pretty wealthy, and I'm quite please to see
that ithere is a lot of positive energy, and new blood. Also we are not
in a hurry, as we don't have to deliver MINA 2.0 in 3 months. We can
spend a few extra months in order to get a good 2.0 MINA. And I don't
want to have a 2.0 out if this is just to start working on a 3.0 which
will be almost a complete rewrite.
Thanks Julien !
Julien
--
--
cordialement, regards,
Emmanuel Lécharny
www.iktek.com
directory.apache.org
