|[tigerstripe-dev] RE: Plugin failures|
FYI, I’ve been looking at running Tigerstripe headless “properly”, i.e. using the Eclipse Headless framework rather than running some code in a separate JVM.
The key thing is that prior to that move we could have no dependency on Eclipse in the base plugin code, which forced unnatural constructs to go around that.
Jim had explored how to integrate the headless run with Maven. I’ve also looked at how other projects deal with that. Buckminster seems to actually build their own “Product” which simplifies the install. They also provide some bootstrapping code allowing to call the Eclipse Headless stuff *from* a JVM.
So long story short:
- there is a bugzilla for this: https://bugs.eclipse.org/bugs/show_bug.cgi?id=215833
- the first iteration should only provide an “Application” entry point for Tigertripe “generation”, but I can see other applications (like packaging modules)
- The return code should definitely be passed back: it is a matter of conventions here.
- All the “PostInstallation” procedure currently happening needs to be refactored (and greatly cleaned up along the process).
- In the first iteration starting Tigerstripe headless will basically mean running a “java ….” Command, which will return an ERR code potentially. Hopefully we can provide a quick integration with Ant/Maven as a first step, then think about properly writing a MOJO or an AntTask for better integration.
Ok, how about next week?
Let's touch base early next week (tues, maybe) and set something up. Sound good?
I'm pretty sure ANT could catch this and "bubble it up" to Maven.
When would be a good tine to discuss?
b) I *think* so, I was messing with that a bit but had to move on to other things. The issue is with the way things stand now (and it *may* not even be an issue, I just need to investigate) is that we have tigerstripe wrapped in ANT wrapped in Maven.
I agree, we need to keep an eye on this an try to work it into the build when feasible.
I may be talking nonsense, but I *think* you said something about the build does not detect a plugin failure and will continue attempting to try to build.
I was chatting to Eric today and mentioned this and it seems it could be pretty easy to put some stuff into Tigerstripe that returns a different exit code that you could catch.
so a) Have I got this right?
b) Would something as simple as return code of -1 be sufficient to enable your maven build script to catch and exit?
Not saying this would be in any build until we go open source but we should plan for improvements wherever we can.