Hi Stefan.

Excuse me for wading in halfway through. I just want to add a couple of observations or suggestions. Firstly, as I have been doing more and more bdd-style projects, I have evolved into the following code structure:

- app [the application]
   - src
   - resources
   ...
- behaviour
   - stories [all the story/scenario/given/when/then classes]
      - src
      - resources
      ...
   - examples [all the object behaviour classes]
      - src
      - resources
      ...
      - integration [examples that use real services, like databases]
         - src
         - resources
         ...

I used to think there was a continuum between describing object interactions (traditional TDD plus mocking) and describing application behaviour (stories and scenarios) but I've since concluded that there are (at least) two discrete levels, hence the directory structure. (There is probably another one above this for interactions across multiple system boundaries, but enterprise BDD is still a little way off - I'm working on it!). I'm trying to come round to Brian Marick's term of "exemplars" rather than "examples" because it is more correct, but it just sounds odd to me.

My build starts by compiling up the app classes (sometimes into a jar), then compiles and runs the example classes /with the app classes on the classpath/, then compiles and runs the story classes with /both the examples and app/ classes on the classpath.

This three-stage approach means I know my app doesn't have any dependencies on any supporting code, and that my story behaviours can reuse code from my example behaviours (say, Builders or other helper classes). It also means my app classes are ready to be bundled up and deployed.

I have found that this is easy using ant (sorry, I'm not a maven user) with the added benefit that it doesn't require any classpath mojo from jbehave itself.

I would prefer to keep the classpath magic as simple as possible in jbehave, so I'd need to see a compelling reason to change Mauro's existing solution (other than tidying it up - have you /seen/ the code Mauro writes?)*

Cheers,
Dan

* For the record: Mauro writes some pretty good code :)


Stefan Hübner wrote:
Hi,

Mauro wrote:
Bringing discussion to the list:

[SNIP]

I personally see behaviours as more along the lines of acceptance testing than unit testing. So rather than a replacement for JUnit, I would see it a replacement for Fit.

That said, some people might want to use BDD as a replacement for TDD.

I would aim to find a configurable way to support both paradigms.

So this would boil down to placing behaviours/stories either in
"src/main/java" or "src/test/java" and scoping jbehave either in
compile or test scope.

The switch could be easily supported by a simple boolean configuration
option, I guess. The plugin then would create a classloader upon
either compile scope or test scope classpath elements.

> I took the approach the surefire-plugin is based on. It doesn't import *any* JUnit-specific classes, but instead loads them into a completely isolated classloader. Of course this brings some nasty reflection mechanics as a side effect, though.
>
> The difference to your implementation is, that the patched plugin doesn't leave the classloading hassles to the jbehave-library. Rather it puts the jbehave-classes into the same classloader, that serves as the classloader for the behaviours to be verified. So jbehave is on the same classpath as the behaviours and nowhere else.
>
> I did some experiments with the hellbound example, put it's sources into different modules and tried to run the behaviours. The patch worked fine. So I was happy with it and thought, you'd find it handy too.

I still need to convince myself of the best approach in this regard. I don't see "hyper-isolation" in classloading as necessarily a significant advantage - especially if offset against other cons or complications. The current approach is quite simple and configurable - it takes the libs configured in a given maven scope and builds a classloader with them. I'll give it more thought.

Fine. I'll those two classloader approaches a try in different scenarios.

> But either way, I'd just like to see a maven plugin that works, since I appreciate your effort very much. I do hope seeing the project gaining momentum in the near future. To me, a maven plugin is a must though, before I can spread the word.

Sure - and I appreciate your help greatly. I've been snowed under of late, but I'm aiming to get some work done (have some other patches to apply) and get out a 1.1 release soon.

Since we're talking about Maven2 support, may I also suggest to update
jbehave's parent- and core-poms? The dependencies should be mentioned
properly. This would help developers great deals.


Cheers

---------------------------------------------------------------------
To unsubscribe from this list please visit:

   http://xircles.codehaus.org/manage_email


Reply via email to