plc4x-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Bjoern Hoeper <hoe...@ltsoft.de>
Subject AW: Integration of other Languages / Daffodil
Date Wed, 10 Apr 2019 08:13:55 GMT
Hey,
I am also glad to join the team and I am really happy that the discussion took off this well.

I would +1 Julians Proposal for a personal meeting at due time for discussing the code generation.
Something that came to my mind was to join this meeting with the Apache Edgent Meetup you
mentioned at the conference (6th May if I remember this correctly). Would this be an option?

Regarding the C# API I can work out a draft with just classes and interfaces.
I have a bit of experience using CMake with .NET so I could also work out a way for integrating
the C# parts into the build. 
My ultimate goal would be to have the .NET Libraries compatible to .NET Standard 2.0 so that
it is usable with .NET Core and .NET Full. One implication of this would be that we need to
think about the exact driver integration (Native code gets a little bit harder but is possible;
but I would prefer a full .NET based solution).

Best Regards
Björn


-----Ursprüngliche Nachricht-----
Von: Julian Feinauer <j.feinauer@pragmaticminds.de> 
Gesendet: Mittwoch, 10. April 2019 09:21
An: dev@plc4x.apache.org
Betreff: Re: Integration of other Languages / Daffodil

Hi,

I did not intend to hurry you in any way, chris : ) You are the one to set the next steps...
I just wanted to bring the topic back (after all that paperwork).
And after your statements here, I would even more like to have a meetup where you present
this a bit more detailed.

What do you think?
Perhaps in a month or so?

Julian

PS.: With the growing community, we should consider to do this perhaps more regularly, or?

Am 10.04.19, 09:15 schrieb "Christofer Dutz" <christofer.dutz@c-ware.de>:

    Hi Julian,
    
    I agree with bringing this topic more back to the core ... as you can imagine it feels
like I have not been working on this to 
    100% as about 90% of my time and energy was consumed with pre-graduation work. I think
this will be over about 08:45 CET tomorrow ;-)
    
    I guess I still will need a few more days to get back in the saddle and finish a first
fully working draft of the code-generator.
    Currently it generates empty java class files ... guess I'll still have to do some tweaking
and adjusting, but should not be too much work.
    Till that I doubt working on templates makes too much sense as I think it would take more
time to do the transfer needed to 
    join in, than I would need to finish it. Of course would I document everything, but I
didn't up to now as I was building, tearing apart, 
    rebuilding, refactoring, ... didn't want to keep the documentation up to date.
    
    Also that will currently generated the model classes as well as the serializers and de-serializers
... after that I need to get started on the
    Driver logic (State-Machine) itself. But I guess having the class generation for the messages
would be a great first start.
    
    But what we really need, is an API module (which is not generated) with a beautiful C#
API module. Then integrating that into 
    Our build will also be a manageable challenge (As I mentioned CMake does seem to support
C# and .NET builds) ... would be great 
    if someone could work on that. 
    
    I guess till I've finished a first working prototype of the model generation, we could
then start porting these templates to other languages.
    
    In parallel me and some people interested, could work on the logic generation (with java
templates) and as soon as that's done,
    Hopefully the language templates for the model are done, we could create the logic templates
for the other languages ...
    
    You think that's a good idea and approach?
    
    Chris
    
    
    
    Am 10.04.19, 08:53 schrieb "Julian Feinauer" <j.feinauer@pragmaticminds.de>:
    
        Hi Björn,
        
        I'm very happy to see you here on the list!
        
        Currently, this "generated driver" topic is mostly on chris shoulders.
        This is fine, as he really pushes things forward and has all the experience and tools
he needs to make this work.
        BUT I think it’s the time now, to make this more of a "Top Level Concern" to get
others integrated in these efforts.
        
        I think this is something absolutely NECESSSARY for the project, so we should make
it a central concern and join efforts on the next steps.
        Especially having everything central in a repo (perhaps simly in a different "develop"
branch or so).
        I guess, the next steps would be to 
        1. Add some documentation or Confluence on all that’s been done previously and the
workings of Chris approach
        2. Play around to make templates for another language
        3. Make thinks stable, add test suites, ...
        
        So I could imagine (Chris, of course, feel free to hit the brakes) that you could
start by prototyping a S7 driver which is strongly oriented at the (generated) Java one which
can then be used to extract the freemarker templates from it. This would be the optimal way
towards C# integration, I think. All the tooling and build automation would be a 2nd level
concern for me.
        Or perhaps you could join forces with Markus (Sommer) which already has a lot of Cpp
and a generated Cpp driver could also have a C# frontend as a first step before we do all
on the .NET Core bases.
        
        What do you think of a dedicated Workshop or Hackday for this topic?
        We learned a lot during our Hackday here in Nürtingen and for this topic I would
suggest to have a bit of a stronger Agenda.
        But this could be a good kickoff to
        1. get all (new) participants of the list on board
        2. get a live introduction by Chris
        3. Drink beer together
        
        As we are no longer South Germany centered we could to this somewhere more centrally
located.
        
        What are others thoughts on this?
        
        Julian
        
        Am 10.04.19, 04:23 schrieb "Christofer Dutz" <cdutz@apache.org>:
        
            Hi Björn and others,
            
            glad you made it here ... hopefully email services will be back to normal soon
...
            
            Let me please explain in a little more detail, what I had in mind with all this
DSDL and generated drivers.
            
            We are currently in the situation where all drivers are written in Java and all
drivers are manually written. As for example the ADS driver is written by someone else than
the S7 driver, you can see huge differences between the way they are built. Also, being the
main author of the S7 driver I wasn't quite happy with how I did it and wanted to clean it
up greatly as I knew, porting this would not be easy.
            
            On the other side we are currently working on C++ versions of the drivers. Just
thinking about the manual effort and what could go wrong by manually porting the drivers and
keeping them in sync, just gave me the shivers. So an alternate approach had to come. I guess
we all agree that we have to automate most of the heavy lifting. 
            
            I am not a fan of any cross-compilation (Write drivers in Java and have them automagically
ported to C++, C#, Go ...) so I favor specifying the protocols in a machine readable way and
have all code generated from that specification. 
            
            For this I evaluated about 7 different approaches and wasn't quite happy with
them, till I tried out DFDL which I first heard from at ApacheCon NA 2018 in Montreal. And
it turned out to be the ideal format for this. However with DFDL you can only specify the
message format and nothing else, so I was still missing the state-machine part ... DFDL being
XML, it would be ideal if this format too was in XML so I could embed DFDL into that ... so
a simple google query "state machine xml" directly guided me to SCXML and it turned out that
a SCXML engine is part of Apache Commons project.
            
            So now I had to prove that it is possible to define an industrial protocol completely
with DFDL and SCXML. The result of this is the "dynamic-driver". The performance is not acceptable
(50-100ms for reading one address), also is it not able to du multiple concurrent requests
at the same time. So we need generated code to bring the normal PLC4X performance to dynamic
drivers. DFDL and SCXML both don't have any code generation till now and I bet this will remain
that way for quite some time. So I started writing a Maven plugin to use DFDL and Freemarker
to generate source files. In the ideal case we would require two artifacts as input: protocol-specification,
language-templates to generate a new driver for a given language. 
            
            Implementing the full DFDL spec is totally out of scope for now, so I am currently
working on a set of conventions we can use to write our protocol-specs which the generator
will be able to process ... this will be a small sub-set of the entire DFDL spec, but it should
be enough to achieve our goal ... and it does make writing specs for new protocols easier
as the author has a reduced set of options to choose from.
            
            The one part we'll not be able to generate is the API for a new language. This
will always be hand-written and that's a good thing. Cause an API designed for Java doesn't
look the same way an API written for C# or C++ will look. For example in Java we use CompletableFutures
and Optionals ... in other languages this might look different. So ideally every API module
for a language will adopt the ideal patterns for that language it is designed for. So that's
where you come in - Björn or others - I don't know how such an API would look like in C#.
So it would be super-great if you could whip up a small API module for the Language of your
choosing.
            
            The next step then would be for us to somehow integrate that modules build into
the PLC4X build system.
            This is usually quite challenging ... but I have read that CMake does have options
for this ... but we'll work that our as soon as it comes up.
            
            The step after that would be to manually port a driver from one existing language
to C# (ideally the generated classes) and create templates from that code ... or directly
start writing templates, however I think that's more challenging as you don't have content
assist of your IDE in Freemarker (I think).
            
            I hope that explains things and I really hope this email goes through (better
save it first)
            
            Chris
            
            
            On 2019/04/08 13:44:35, Bjoern Hoeper <hoeper@ltsoft.de> wrote: 
            > Hi everyone,
            > I got in touch with Chris and Julian during the buildingIoT Conference in
Cologne and I would like to participate in the further development of PLC4X. Our main interest
is to extend PLC4X to make it usable with .NET / C#.
            > 
            > So after looking at the sources and the discussions in the mailing list I
was wondering what would be the best move forward regarding the integration of other languages
in general and the decoupling of the protocol definition from the actual API implementation
in a particular language. As far as I understand the current status Apache Daffodil is  a
candidate technology for generating the adapter code. But at the current moment Daffodil does
not support templating of classes. Furthermore there are existing APIs for Java and Scala
but no implementation for languages not executed on the JVM so we would need to have an implementation
of the Daffodil APIs in every language we want to support in the future.
            > Are there any Ideas existing already about an Architecture to support languages
besides Java? If not I would like to encourage a discussion about this topic because I think
it will get quite fundamental as soon as the adoption of further languages gains some traction.
            > Best Regards
            > Björn
            > 
            > =================================
            > Dipl.-Ing. B. Höper
            > Geschäftsführer
            > 
            > LTSoft - Agentur für Leittechnik-Software GmbH
            > Veilchenweg 37a
            > 51107 Köln
            > 
            > Telefon:   +49 (0) 221 - 79 00 35 31
            > Telefax:   +49 (0) 221 - 79 00 35 35
            > Mobil:      +49 (0) 173 - 28 36 904
            > Email:      hoeper@ltsoft.de<mailto:hoeper@ltsoft.de>
            > =================================
            > 
            > 
            
        
        
    
    

Mime
View raw message