|
|
|
| Re: get the instance of the Main AcceleoGenerator class from query invoke [message #1841988 is a reply to message #1841984] |
Sun, 06 June 2021 06:49   |
 Ed Willink
Messages: 7220
Registered: July 2009 |
Senior Member |
|
|
Hi
If you are enthusiastic about Acceleo and want to make it available in a different context, I recommend that you establish a close relationship with the Acceleo team so that your work can enhance rather than fork Acceleo.
I fear that you may find that your users have different metamodel resolution requirements and so you may find that the current URI fudge, that only works for simple models, has to be revisited. You may find that other pain points also justify your time.
wrt an Acceleo tooling model, consider a similar Eclipse tooling model, which already exists as workbench.xmi which instantiates many metamodels such as http://www.eclipse.org/ui/2010/UIModel/application. You could write a Java invoke to get at the Workbench root and from there, an Acceleo M2T could locate many other things. But again I ask why bother? If a user really wants so much insight at the Java level, much better to program in Java or Xtend. I really doubt such users exist.
In a web application, providing an ability for users to browse the underlying application will of course aggravate security challenges,
Truly static in a multi-use application could well be a serious problem. The easiest solution is to ensure that each use runs a new copy of the Acceleo application. Else each use creates a new worker thread and you use ThreadLocals. New copies have the major advantage that it is really hard for one usage to crash the overall application and so mess up other usages. New threads allow test harnesses to simulate multiple concurrent usages. Maybe new runs and worker threads.
Regards
Ed Willink
[Updated on: Sun, 06 June 2021 07:03] Report message to a moderator |
|
|
| Re: get the instance of the Main AcceleoGenerator class from query invoke [message #1841991 is a reply to message #1841988] |
Sun, 06 June 2021 11:33 |
Thomas Barth
Messages: 11
Registered: March 2021 |
Junior Member |
|
|
Hi,
the metamodel would be fixed and the service would only offer generation for this one metamodel.
Let me summarize what I do and where I want to go with it:
As part of my PhD Thesis I defined a EMF metamodel describing a (safety)(multicore) deeply embedded system.
http://www.barth-dev.de/wp-content/uploads/2020/02/Session_4.2_Barth_paper.pdf
Signalflows can be drawn with Sirius and an model specific runtime environment as well as a safety report can be generated with Acceleo.
For now the Runtime Environment utilizes an underlying Realtime Operating System but I plan to get rid of that part after the PhD so that the runtime environment also handles scheduling, memory protections, stacks etc. and becomes a "all in one" solution but Im not sure if it really works outside of my head.
A prototype of the tool is already successfully used in academic and industrial projects.
Besides the use in an professional environment (for which I plan to offer commercial consulting) I am thinking about to make the tool freely available but with limited functionality (e.g. no Safety and requirements modeling). For that usecase Im thinking about a web application which basically has the same functionality as the Eclipse implementation. For that web application it would be nice if I could reuse as much code as possible from EMF so that maintenance of both implementations becomes easy and doable in a reasonable timeframe. Of course there also could be a "light" version of the eclipse application to do the same as the web app but here I would need to buy some consultancy because though its working, I am an electrical engineer and embedded developer and not a trained java developer so with my knowledge now (with knowledge being a function of time) it feels better to reuse the parts out of EMF I fully understand instead of releasing something with blind spots in it. Its a dynamic process on where Im heading with it right now.
[Updated on: Sun, 06 June 2021 22:23] Report message to a moderator |
|
|
| Re: get the instance of the Main AcceleoGenerator class from query invoke [message #1842021 is a reply to message #1841991] |
Mon, 07 June 2021 13:24 |
Ed Willink
Messages: 7220
Registered: July 2009 |
Senior Member |
|
|
Hi
Good luck making money from modeling. Few are succeeding. Most users expect it all to be free, particularly those of use who provide free software.
You will be be close to a trained Java developer by the time things work. You may find it helpful to have 100% (not 99,99%) Open Source plugins and distinct proprietary plugins. Stay well clear of fragment plugins.
I cannot really imagine what your textual output does, but if you want to tease your users adequately then provide the full safety and requirements modeling for free, restrict your charging to full functionality rather than stub functionality safety and requirements M2T.
Remember also that today, while working on your PhD, your work may seem like the most important thing in the world. Tomorrow when you are faced with the realities of earning money in the jobs market, your wonderful service may evolve into a millstone.
I write from bitter experience. OCL and QVT could be wonderful if they worked well, but with tiny support teams and almost zero funding, tool quality is very disappointing and really struggling to chase the mainstream.. The reality is that large parts of industry are seriously uninterested, and increasingly uninterested, in modeling. As for UML, for many it is just a joke.
Regards
Ed Willink
|
|
|
| Re: get the instance of the Main AcceleoGenerator class from query invoke [message #1842031 is a reply to message #1842021] |
Mon, 07 June 2021 20:36 |
Thomas Barth
Messages: 11
Registered: March 2021 |
Junior Member |
|
|
Hi Ed,
Thank you a lot for sharing your experience. I can imagine that it was/is bitter to see a brilliant idea not reaching its full potential.
I personally want to thank you for the effort you put into Eclipse, OCL etc. as it helped me to implement the concept of my thesis with reasonable effort while learning a lot about bigger scale open source software projects and Java.
I released some open source software projects and know how demanding and ungrateful that can be, so also thank you for your support in this forum.
You are definitely right with your warning!
I mainly focus on automotive and machinery where model based software development is widespread (e.g. AUTOSAR, Simulink etc.) and even becoming more popular due to the increasing complexity of hardware and functionality but also due to the fact that safety software development needs to be documented for certification and M2T partially brings that automatically.
In the first years of the thesis I felt like you said, that this is the most important thing in the world but over the years I realized that though there is a demand for what I do, its hard to get into the market and to compete with existing solutions.
There are many things that need to be right for a project to succeed. A brilliant idea and implementation even with a demand on the market might be not enough and I am aware of that.
The idea to get startet (for now) would be not to sell the tool (which would require ex(t/p)ensive documentation, qualification, certification) but to use it as part of general consultancy in the area of embedded systems (for safety critical applications).
Maybe that doesn't work out but its worth a try because I have seen that it adds value for the customer in the projects it is used in.
The project a nutshell:
It allows you to model dataflows and software behavior (with an own graphical notation based on UML and SySML) as well as underlying hardware, logical functionality and (safety)requirements.
Which piece of software is executed when and on which piece of hardware and which data is exchanged between all the pieces of software in your system.
The textual output is C or C++ code implementing an runtime environment (in accordance with safety standards and with tractability to all related elements in the model) where all the user has to do is to add logic within the blocks he defined in the model without the need to fiddle with the complexity of a state of the art (muliticore)microcontroller with RTOS or certain safety requirements stated by the standards, which can be pretty challenging if implemented by hand.
Moreover a a safety report shall be generated, on which Im currently working and which is the last practical part of the project which started around 5 years ago.
[Updated on: Mon, 07 June 2021 22:04] Report message to a moderator |
|
|
|
| Re: get the instance of the Main AcceleoGenerator class from query invoke [message #1842055 is a reply to message #1842039] |
Tue, 08 June 2021 09:10 |
Thomas Barth
Messages: 11
Registered: March 2021 |
Junior Member |
|
|
Hi
Thanks for this inspiration which sounds interesting but I don't think I need something like that.
The model isn't a 1 to 1 representation of resulting classes and objects but rather abstract at some points because its easier for the designer this way. I decide what to generate out of the abstract definition during transformation.
I could generate another model which represents the actual codebase as intermediate step but I don't see too much benefit from that. An interesting idea though.
The generated C code also follows an object oriented approach by creating structs as objects and designated functions for each type of struct. I can reuse a lot of code this way and it has certain advantages.
The C++ code is used by humans and the compilation process is kind of a first consistency check + since the system architecture is based on objects, the code is nicer, more compact and easier to understand.
I don't see a need to target other languages than C and C++ for now as they are still state of the art in this field. Maybe at some point rust but not in the near future.
[Updated on: Tue, 08 June 2021 12:01] Report message to a moderator |
|
|
|
Powered by
FUDForum. Page generated in 0.02384 seconds
Search
Help
Register
Login
Home
