[Fledeskum]

The company I work for (which would qualify as an enterprise by any standards) decided to switch over to Django almost 3 years ago, rewriting our existing Java infrastructure on a project-to-project basis. The reasoning was primarily two-fold, we'd begun to have trouble recruiting good new talent interested in working with Spring/Java and we felt that for what we were doing, Java was not the strongest technology choice anymore (not like it was 10 years ago).

Last year, we switched a 55k loc Spring MVC (with JAX-RS for the json api) over to Django (using tastypie for the api side and celery for offline tasks). The result ended up being little over 8k loc of Python. Closing a ticket has gone from taking on average around 5 days to a little over 4 hours. That was the third project we ported over, the others having had similar success. YMMV.

[edwinnathaniel]

I find your story to be very interesting. It used to be the case that it is easier to find talents in Java/.NET but that doesn't seem the case these days.

Why Django not Rails since Rails seem to be more popular?

I found it hard to believe that you can switch 55k LoC of Spring MVC + JAX-RS to Django with resulted of 8k. Because from what I've learned over the year, the amount of code requires to write Spring MVC + JAX-RS is very minimum and very close to Rails (my experience is more with Rails) at the very least from the Controller point of view. My personal like of the paradigm is to re-use the code between Spring MVC and JAX-RS. Love it so much.

That is of course if you discount everything else: imports, comments, parentheses, configurations. What about the actual logic?

We typically use JS heavy on the front end and less on JSF/JSP.

But you're right, YMMV. Projects have different requirements and skillset/experience. We typically don't have technical bugs but more of requirement bugs: someone forgot to handle the case of X,Y,Z and as such, it requires about a day or less to plug the issues.

[Fledeskum]

Most of the gain was from actual logic, if there had been only models & controllers we wouldn't have seen any major benefits. But there was a lot of IoC with the customary mountains of rather pointless interfaces, an "admin" of sorts & a "messaging" queue running tasks. Most of the POJO data classes ended up being replaced by Python's built in dictionary. Basically the mixture of Python's stdlib, the language ecosystem & Django's got rid of most of our code. In the cases where we couldn't reuse anything from anywhere, something like a business logic class would normally shrink by 2/3. And the frontend was not JS heavy so that factors in.

I hadn't started when the decision to go Django instead of Rails was made (Rails & .NET MVC were also considered) but I've been told it came down to a number of things:

* This is in Europe, generally Ruby is not used very much over here, Python has been used in some shape or form in every company I've have experience of but I've only seen Ruby recently and then mostly for Chef or Puppet.

* Lack of explicit imports did not work in Ruby's favour.

* If needed numpy & scipy don't have a analogue in the Ruby ecosystem.

* General negativity towards the religious zeal permeating through the Rails community.

[edwinnathaniel]

From what I know, JavaEE 6 IoC doesn't require interface so that may cut the code a little bit further. The thing about Java ecosystem is that they rely heavily on IoC/Interface for testing and I can't really blame them because that set-up guarantee super fast unit-tests unlike Rails ActiveRecord (in fact there's a little bit of movement recently in the Rails community that picked up from Java).

ASP.NET MVC won't have much gain over Spring MVC/JAX-RS. In fact, there's nothing similar to JAX-RS in ASP.NET MVC (you'd have to write your own stuff here and there and wire them together, not as straightforward as JAX-RS).

This is an excellent story to share both from technical perspective (Java+Frameworks => Django+Python+Pythonic mindset) and from recruiting perspective. I hope one day you could share (perhaps in slides/presentation style) the actual code and technique in details.

Mind if I ask you where in Europe? (Country/City?) For a while I thought Europe is Rails heavy.

[irahul]

> Last year, we switched a 55k loc Spring MVC (with JAX-RS for the json api) over to Django (using tastypie for the api side and celery for offline tasks). The result ended up being little over 8k loc of Python.

While comparing loc, take into account the fact that there were no unknowns - you are simply porting functionality. You generally don't have to solve a problem when you are re-writing a Java codebase in Python, just pythonizing the existing java solution. I would wager if the same system was re-written in Java, it would shave some bloat(provided you are re-implementing and not adding features).

With that said, 55k to 8k is still very impressive.