I've been investigating OSGi for my company's software, but have recently been recommended to take a look at Impala. According to its web page, Impala is "a dynamic module framework for Java-based web applications, based on the Spring Framework."
At a glance, and looking at this blog post about the differences, the key differences I can see are that Impala is simpler than OSGi, does not manage versioning of third party components, and is far less widely used/known (I do not see a single question about it on Stack Overflow).
I wonder whether people who have direct experience with Impala and OSGi (i.e. those who have investigated it more deeply than reading blog posts and online docs), have any more insights into the practical differences between the two, and/or suggestions about what types of projects each one may be more or less suitable for.
Edit: It may also be interesting to include Springsource Slices into the comparison, although it is as yet an early prototype. At a glance, it appears to only work in DM Server.
Impala's approach to modularity is very weak when it comes to controlled sharing between modules. The problem is that Impala still follows the old J2EE-style hierarchical approach to classloading.
Anybody can write a module system that restricts visibility of classes across modules. The difficult part is how you reintroduce dependencies between modules such that specific classes and interfaces from one module can be seen by another module. In OSGi we do this by exporting and importing packages, so we have a non-hierarchical dependency graph.
In Impala, if you want to see the classes in another module, then your module must be a child or descendant of that module. That is, modules can only see their own classes and those of their ancestors. Now if you want to share some classes with your sibling module (e.g. a library that you both use) then you must move that library up into the classpath of your shared ancestor. In the worst case you have to move it right up to the root module. Now the library is visible to ALL other modules whether they want it or not! Indeed, if another module wanted to use a different version of the library they would be prevented from doing so.
If you simply have a copy of the library in each place where it is used, then you make it impossible for the modules using that library to communicate with each other. They will get ClassCastExceptions when they try to pass objects between each other.
A similar problem is inherent in J2EE if two web applications need to use the same library. Typically J2EE developers just copy the library, but this creates "silo" applications that cannot communicate with each other. It is simply not the way to build modular software.
Steven's points also seem pertinent. As far as I can tell, nobody is using Impala aside from its author.
In my eyes, there is no comparison. OSGi is a mature framework that's been around for 10 years and is the basis for the implementation of most of today's Java containers. OSGi has growing adoption, there are books available and, yes, people talk about it on Stack Overflow!
Impala hasn't even hit a stable release and appears to be a 1-man project, though he is asking for additional developers now.
So, it depends on your criteria. If you are investigating technology out of interest, then I don't see any issue writing stuff with Impala. If you are looking to base your company's future products on it, then I think that would be professionally negligent.
Related
background
I have designed many tools in the past year or so that is designed to help me program for XPages. These tools include primarily helper java classes, extended logging (making use of OpenLogger and my own stuff), and a few other things that I personally feel I cannot work without. It has been discussed with my employer, and we feel that it might be a good idea to start publishing these items to openNTF. Since these tools are made up of about 3 .nsfs, all designed to use the same java code, key javascript classes, css, and even a custom control or two, I would like to consolidate key items into a plug-in that can be installed at the server and client level. I want to do this consolidation before I even think about publishing any of the work I've done so far. It would just be far too much work to maintain, not just for me, but for potential users. I have not really found any information on how to do such a thing in google searches. I also have to make sure that I am able to make use of the ExtLib libraries, openNTF Domino API, and the Notes API.
my questions
How does one best go about designing such plug-ins? Must a designer
use eclipse, or is this it possible to do this directly in the Notes
Designer?
How does a designer best go about keeping a server and client up to date while designing and updating the plug-in code? Is this why GitHub is often used?
Where is the best place to get material to get started in this direction? I sort of feel lost in the woods, knowing I need to head north, but not having a compass for that first step.
Thank you very much for your input.
In my experience, I found that diving into plug-in development is a huge PITA until you get used to it, but it's definitely worth it overall.
As for whether you can use Designer for plugin development: yes, but you will likely eventually want to not do so. I started out by using Designer for this sort of thing for a while, presumably with the same sentiment as you: why bother installing another instance of Eclipse when I'm already sitting in one all day? However, between Designer's age (it's roughly equivalent to, I think, Eclipse 3.4), oddities when it comes to working sets between the "Applications" and "Project Explorer" views, and, in my case, my desire to use a Mac app, I ended up switching.
There are two major starting points: the XSP Starter Kit (http://www.openntf.org/internal/home.nsf/project.xsp?name=XSP%20Starter%20Kit) and Niklas Heidloff's video on setting up Eclipse for XPages development (http://www.openntf.org/main.nsf/blog.xsp?permaLink=NHEF-8RVB5H). The latter mentions the XPages SDK (http://www.openntf.org/internal/home.nsf/project.xsp?name=XPages%20SDK%20for%20Eclipse%20RCP), which is also useful. In my setup, I found the video largely useful, but some aspects either difficult to find (IBM's downloads are shifting sands) or optional (debugging, which will depend on whether or not you're using Eclipse on Windows).
Those resources should generally get you set up. The main thing to worry about when setting up your Eclipse environment will be making sure your Plug-In Execution Environment is properly done. If you're following the SDK setup instructions, that SHOULD get you where you need to be.
The next thing to know about is the way plugins are structured. Each plugin you want to install in Designer or Domino will also be paired with a feature project (a feature can house several plugins), and potentially an update site - the last one is optional if you just want to import the features into an Update Site NSF. That's how I often do my normal plugin development: export the paired feature to a directory and then import the feature into the server's Update Site NSF and then install in Designer from there using Application -> Install. You can also set things up so that you deploy into the server's plugin/feature directories instead of taking the step of installing into an update site if you'd prefer. GitHub doesn't really come into play for this aspect - it's more about sharing/collaborating with your code and also having a remote storage location for your git repositories (which I highly advise).
And as for the "lost in the woods" feeling: yep, you'll have that for a good while. There are lots of moving parts and esoteric concepts to get a hold of all at once. If you mostly follow the above links and then start with some basics from the XSP Starter Kit (which is itself a plugin project that you can pair with a feature) - say, printing text in the Activator class and making an implicit global variable just to make sure it works - that should help get your feet wet.
It's best done in Eclipse. You can debug your code running on the server from there, as well as run it directly from there. The editors are also more up-to-date. You want:
Eclipse for RCP and RAP developers
XPages SDK for Eclipse RCP (from OpenNTF)
XPages Debug Plugin (from OpenNTF - basically allows you to load the plugins to the Domino server dynamically, rather than exporting to an Update Site all the time)
XSP Starter Kit on OpenNTF is a good starting point for a plugin. There are various references to the library id, which has to be unique for your plugin. Basically, references to org.openntf.xsp.starter need changing to whatever you want to call your plugin. You're also best advised to remove what you don't need. I tend to work in a copy of the Starter, remove stuff, build and if there are errors with required classes (Activator.java obviously will be required and some others), then paste them back in from the Starter.
XPages OpenLog Logger is a good cross-reference, that was built from XPages Starter Kit. It's pretty much stripped down and you'll be able to see what had to be changed. A lot of the elements of the XSP Starter Kit correspond to Java classes you'll probably be familiar with from your XPages Java development.
GitHub etc tend to be used as source control, which is useful for working out what's changed from time to time.
I'm working on legacy code in a DotNetNuke module, trying to get classes and behaviors under a testing framework: I'm taking this opportunity to follow advice from the "Working effectively with legacy code" book, so what happens is that i'm trying to define areas which can be tested thoroughly and then converted to services. Then i'd like to use an IoC framework for it to work. For now, i've set eyes on Ninject.
However i'm hitting a design problem: as i'm in a DotNetNuke module, i can't really change application-wide structure: for example i can't derive the Application from NinjectHttpApplication. I can't use these suggestions from SO either.
I was thinking about having the Kernel in a static class that my module would set up and then use but from what i've read around it's a very bad idea.
So i'm starting to ask myself if it's possible to use an IoC in an application that hasn't been set up to support it from scratch. If i'm supposed to have a whole dependency tree loaded for each request, how can i rewrite legacy code locally and benefit from IoC? Is there a pattern where IoC use can grow out from very local rewrites?
Even though i'm working with DotNetNuke, any standalone component that can be installed into an independent framework begs the same question. Also i'm not targeting Ninject specifically, if another IoC framework can help in this case i'm willing to consider it.
From my experience, your best bet to get this type of abstraction within the context of DotNetNuke is by using the WebFormsMVP framework. This is really the only sane way I've found to do unit testing in a DNN module, and if memory serves I spent awhile trying to wire up Ninject a year or so ago.
But be warned, it is still WebForms and will never be drop dead simple. And without knowing your existing code base, I'd have a hard time knowing how easy it will be to make the migration.
I have a couple of resources on GitHub that you can check out for reference:
The first is a module template that should act as a solid starting point:
https://github.com/irobinson/WebFormsMvp-DNN-Module-Template
The second is a small example project:
https://github.com/irobinson/BeerCollectionMVP
Depending on the version of DNN you're using, it may or may not already ship with WebFormsMVP, but you should be able to either bundle the dependencies w/ your module or upgrade to the newer version of DNN if that's reasonable.
I was wondering if there is an off the shelf tool to help create a 'web application' directory structure with a basic application up and running? The source code would already be pre-written I'm guessing or based on the options some files may/may-not be generated.
I'm not sure what you'd call it but say we have a custom framework* which are to be used for web application development - rather than 'creating' a directory structure all the time we could just create it once and have a console like interface similar to the play framework to generate a basic application or an empty one as per the developer's choice.
We could just give various types of 'zip' files and ask folks to unzip it and import it in their IDE of choice and continue. However, we'd prefer to have an 'installable' to run from the command line (or GUI but no such preference) to have a basic application up and running without everyone wanting to do it all over again.
How does the Play framework do it? What do they use? (I'm guess similar things exist for RoR, Groovy/Grails.)
*It's not custom per se, but similar to having all the spring/hibernate/restlet/freemarker etc files pre-configured, up and running and a directory structure with packages for the various components by convention
I think one of the key points here regarding the Play framework is that it uses the concept of convention over configuration. Your applications are forced to follow the same pattern for the different parts of your application, or it will not work. I personally like this because it makes working on different projects easier, as the rules are always the same, rather the somewhat unwritten rules of best practice.
Java EE on the other hand takes the concept of configuration over convention. Therefore all your files and structures are defined in your relevant XML documents that specify your frameworks, classpaths, etc. There do exist some tools to try to bridge the gap. For example
IDE's will have project creation tools for your chosen framework, so will create a Struts or Spring MVC project structure with a few simple wizard steps. Eclipse does this for sure as one example.
Spring MVC also has Roo. This is a boilerplate code generation tool that creates large parts of your initial project for you.
From your description it seems you have a few different frameworks that you want to have auto-generated, but I don't think any tool currently will serve your purpose. Your concept of a zip file is your best bet here.
If you want a kind of scaffolding in the Java EE world, take a look at Appfuse which provides some archetype with several implementations on the views layer (JSF, Spring MVC, Struts 2...).
I'm building several sites that need similar "modules." For example, the sites may have the exact same login system, forum, etc.
Is there a way I could build these modules once and just "drop" them in these various sites? Some of the challenges I see:
Keeping the code consistent in the various sites. Any changes made to a module should propagate to all of the sites using that module. I guess I need a way to upgrade?
Database: these functionality need to work as part of a bigger application. Maybe the module needs to define relationships with other tables in its respective site.
I'm sure there are more problems. I think I should be looking at this: https://bitbucket.org/wiredesignz/codeigniter-modular-extensions-hmvc/wiki/Home, but I don't have any experience with it.
So, I'm looking for solutions, suggestions, or more problems to this idea.
You can create and use third party packages by adding them to the third party folder (which is new for CI 2). There is not much about it in the docs, but i found this.
http://codeigniter.com/user_guide/libraries/loader.html
You can autoload the third party packages in the autoload file. Packages can have their own controllers, models, views etc.
Interestingly, Phil Sturgeon wrote a bit (http://philsturgeon.co.uk/blog/2010/04/codeigniter-packages-modules) about packages not being modules (in the strict sense of the term), but you could probably use third party packages for what you need.
I would write them as libraries and use Git submodules to manage each module. Phil Sturgeon actually just wrote a great post about doing this in CodeIgniter.
If you're not using version control, I can't see an easy way to sync across all of your applications. Yes, HMVC will let you break apart your application into actual modules, but it won't help in syncing those modules across your applications.
Here is my question about 'Database communication in modular software'
that you may find useful.
I'm little bit familiar with Drupal, and as a modular application, I think it can be taken as good example of how relationships between modules should be defined.
Here is one good post about art-of-separation-of-concerns
I would like to hear if you have run into some concrete challenges, solutions and references concerning modular design in CI.
What are some of the best practices for OOD with Ruby? Mainly, how should files and code be organized?
I have a project that uses several classes and files and I am just wondering how it should all be organized, grouped and included.
It sounds like you're asking which pieces go in which files.
Is your project a Web application? In that case you would most likely use the system of organization imposed by your framework (Rails, Merb, Sinatra, etc.)
Other kinds of projects also have their own typical structure that you can just follow. E.g. gems are usually set up in a certain way.
If it's a console app, there's no strict rule. Usually people put no more than one class or module in a file. You could have one main file that requires all the others.
Standard OOD concepts apply to ruby. For specifics, maybe this guide will be helpful:
http://www.rubyist.net/~slagell/ruby/oothinking.html