I have a project whose core domain is dramatically changing. It's possible to use 50% of the core functionality from this site and just add the 50% new functionality, but I am starting to consider that maybe it might be faster to simply redo the product in Rails. Development speed is very important.
There are some things I really like about java - the performance and scalability are very good. I am not a crappy Java developer, so my apps tend to run very well - better than the Rails sites I've seen. I've always accepted the idea that people probably just throw a little more money at the problem when it comes to using Rails, which probably works itself out in the end because of the insane productivity benefits.
I am actually quite agile with Java. I know it will still take me longer to add a basic entity to the system, but I am quick at it and I don't mind it that much. At least it's easy and straight-forward to do.
What I do mind is:
- having to start/stop the server just to fix a route, lazy load exception, controller is going to wrong view, etc.
- putting up with the fact that unit/integration tests sometimes have different results than the production environment (because annotations on controllers can't be tested, or lazy-loading exceptions occur during asynchronous service calls, or things like that). Knowing if your Jackson is marshaling your data properly is another Tomcat-only thing because it's handled by Spring. There are lots of things that go wrong after you have tested all that you can, and this frankly annoys the crap out of me.
- putting up with the occasional maven/classloader problem that doesn't rear its ugly head until you deploy into tomcat. It gives the false impression that everything is "a-okay" when you are in your IDE.
- having to put more effort to do database migrations than the ruby people ever have to.
- putting up with framework bugs in Spring that block (it's happened about 5 times on this project since 2009) or Hibernate. I also don't like upgrading Spring Security and having them constantly change the configuration, apis and tag libraries over and over again. This is annoying.
- wasting so much time uploading 58 MB war files to the server! These take me 12 minutes to upload whenever I need to deploy changes. If I forgot to do 'mvn clean' before I upload, Spring might complain that 2 beans exist with the same name because I moved one to a new package... and then I have to re-upload the whole stupid war file again. Why isn't "clean" run by default whenever you do 'mvn package' for?!?! Sometimes these frameworks and tools use the stupidest default settings. This is just so common in the Java world.
- Having to spend hour(s) to figure out where a framework wants to plug-in your own custom implementation for something. This is very annoying. You can spend 2 hours sifting through Google and crappy documentation trying to figure out how to override Spring Security's authentication mechanism for example... and then spend only 5 minutes writing the actual implementation. Of course, they wrote paragraphs upon paragraphs explaining the architecture and how awesome it is, but nobody cares. For something so common, why not just give example source code and be done with it?
- Waiting 10-15 seconds for Spring to start up whenever you want to run your integration tests. This is a drag.
There are a few things I like about Java though. Role-based access is very easy to do with Spring Security. Authentication is never that big of a gain, but I like the implementation inside of Spring.
I also like Spring's form-backing objects and @ModelAttribute. These are huge wins when it comes to controllers, and I don't know if Rails can do these things. I honestly never liked passing request parameters around in every action - Spring MVC is actually a lot easier to use when it comes to this common bloat.
Being able to cache really massive structures in memory and have them stay in memory when you start the application is also highly desirable, especially for this application actually. I have an in-memory thesaurus and grammar checker that needs to get called hundreds of times per request, so in memory is pretty much the fastest option for me.
Even still, I think I could rebuild what I have in 2-3 weeks, and then add all of the new features in a few weeks using rails.
I'd appreciate some advice on the subject before I continue.
PS: I could also go to Spring-ROO... but that would also be a considerable rework. I was never using JPA - I was using Hibernate directly. I am also not using JSP's - I am using Freemarker.
It takes more time to get good at Ruby, and Rails. I worked as an independent contractor as Spring and Hibernate expert myself, but I felt strangled by java and it's web frameworks so I decided to learn Ruby on Rails.
I would advice you to learn Ruby, from what I read you would probably master it, although get pretty frustrated with the very different way the use the ORM. I had issues with it, used to working on aggregate roots in Hibernate to the ActiveRecord one class one table kind of pattern. But hey, you could easily try out MongoDB to have some real fun.
- less code
- it's fast and scalable (slower than java on the specific tasks, but you get rid of stacks of layers.)
- the problems are more often; which gem should I use. Luxorious!
- a unique, big, sharing and caring open source community
- nice frameworks, as Rails and Sinatra
Would I advice you to do the project you describe in Ruby.
Not if speed of development matters. You will be slower, trust me. There's a lot to learn, it's conventions are not familiar to a java programmer and when you get stuck, lots of hours fly by.
The best option would be to hire a senior ruby developer to pair up with you and teach you. Be a good apprentice and you'll learn fast. Faster than me, I had to learn most by myself, which is really inefficient.