[buckminster-dev] Re: Headless (JUnit) launching

[Achim Demelt <a.demelt@xxxxxxxxxxxxx>]

Carsten,

comments below.

> Hi Achim,
> 
> Great work! We are using the Buckminster JUnit Command and refactored JDT
> and PDE from Google Code for a few weeks already and they are working
> great. I assume the JUnit version in the google svn is not quite
> up-to-date? I've made a few changes against it, mostly some logging in the
> TestListener and catching some configuration and launching errors in the
> command itself, should you be interested in a patch.
> 

You're correct. The code at Google's SVN repo is terribly outdated. I have a 
new version waiting to be checked into the Eclipse repository. In 
particular, the new version writes XML output that is very similar to what 
the regular JUnit Ant task produces. This output can be correctly parsed by  
Bamboo and hopefully Hudson, too.

I'd be interested in your changes so that I can compare them to the changes 
that I've made. Can you please file a bug in the Google Code project and 
attach a patch with your changes?

> Two questions of my own, though:
> 
> 1) Are there any plans/has there been any discussion about providing JUnit
> support as an actor instead of/in addition to a command?
> 

There are no discussions or plans that I am aware of. Personally, I think 
that launching is a workspace-related thing and as such, a command is the 
best way to handle this. I may be wrong, though...

> 2) Do you have a good way to deal with platform-specific settings in your
> launch configurations? We would've really liked to have launch
> configurations that we can use both on our development boxes and on the ci
> server, so we can be reasonably sure that the tests we execute locally are
> configured the same as on the server. However that's a bit problematic in
> light of platform-specific plugins like swt and filesystem. Right now we
> provide a template launch configuration that contains all
> platform-specific plugins from all platforms and copy that to create new
> configurations. Do you have a better idea to handle that? We don't want to
> use "All plug-ins and features" in the unit tests, because we want to
> isolate the tests and plug-ins under test as much as possible during
> unit-testing. And "Add required plug-ins" only adds requirements for my
> current platform.
> 

That's a good question. Honestly, I currently have no platform-dependent 
bundles in my test launch configs. Off the top of my head I'd say that you 
can safely run the configuration containing all platform-specific bundles. 
Equinox should filter out any bundles that are not allowed to run on the 
current platform. Have you tried that?

Cheers,
 Achim

> Best regards,
> Carsten
> 
> On 03.09.2009 11:42, Achim Demelt wrote:
>> It seems that with Helios M2 we'll have all prerequisites for headless
>> JUnit launching available. I'll be ready to check in the Buckminster
>> JUnit command shortly after.
>> 
>> I have two questions, though:
>> 
>> 1) If we build Galileo SR1 from trunk, I'd better not include the new
>> bundle in the o.e.buckminster.pde.headless feature, right? Otherwise I'll
>> break the build because of the 3.6 dependencies.
>> 
>> 2) With a little extra effort --- and I mean _really_ little extra
>> effort, < 4hrs --- I could provide not only JUnit launching, but _any_
>> kind of launching from the headless workspace. Plain Java and OSGi
>> launching will come out-of-the-box. Launching any other kind of launch
>> config will only require that the respective LaunchConfigurationDelegates
>> are installed in the Buckminster headless host. What do think about that?
>> 
>> Regards,
>> Achim