So, how expensive is it to let NInject make my object?
Back here I made a quick and dirty performance comparison. In the comments I actually got some "flak" scrutinizing some of the "design decisions" behind the example shown. I would like to reinstate that the example was made purely for checking instantiation performance of IoC container, and that I could not be asked to use "Foo" and "Bar" or "Samurai" and "Sword" (no worries, Nate ;)
I was pointed to the injection container NInject, and how the benchmark would look like when using this framework. Now that said framework has been released with a v 1.0 tag (you've got to love it just for doing that, after all, there are customers out there that you simply cannot approach with a v0.9 or a release candidate 3) I took the time to play through the code.
Instead of using an XML-file you perform a programmatic binding within a special class that is treated by the kernel as a module:
class BankAccountModule : StandardModule
public override void Load()
Instantiation of the kernel looks like this:
IKernel c = new Ninject.Core.StandardKernel(new BankAccountModule());
And getting a BankAccount reference looks like that:
BankAccount acc = c.Get<BankAccount>();
The time we get (for reference the Spring.NET time run today as well)
- Container construction (Spring.NET): 0.0535 / 0.0459
- Container construction (NInject): 0.0298 / 0.0293
So, performance-wise NInject shows benefits due to its approach of generating dynamic methods. This is certainly not the last post on NInject as it makes a very good first impression and I need to delve a bit into some question marks.