Bug 348947

Description Miles Parker 2011-06-09 15:17:56 EDT

This is an issue that has come up during final Eclipse Indigo release build. Please see this cross-platform email for reference:

[Me]

OK, uncle. I have responsibility for org.eclipse.emf.java issue. Basically, I need to consume this at runtime, and it has only been available at end-user build-time, so I've needed to provision it.

David, there are four things I can do at this point:

1. Add back in that damn about.html file. I have no idea why it disappeared.
2. Modify my rmaps so they point to the Eclipse hosted
3. Modify my plugin to remove the *one* class that depends on it, removing that functionality.
4. Give up and try again next year.

I reconstructed the history on this one a bit. First, the dependency itself is actually pre-Eclipse, that is before AMP joined the Eclipse platform, 2007. So naturally having the dependency wouldn't have been discussed by PMC at that time and I had no choice but to provision it myself. Second, for builds I initially had had the dependency on the EMF CVS but I changed that on 7/28/09:

"Added dependency on org.eclipse.emf.java to amp to maintain consistency and for ease of client build."

As this was about the time I was trying to get the CBI builds going, my guess is that I had trouble somehow with collecting from the modeling CVS build. Or I may have thought that I "should" be maintaining it in my dependencies just as we are required to for Orbit dependencies. God knows, and she doesn't care.

There may have also been an extremely subtle issue that effected why I needed to maintain it separately. I know that there were some class loading issues that I was trying to solve at some point. I may have not wanted ecore reexported as that seems to be only difference between the two.

 org.eclipse.emf.ecore;visibility:=reexport,

The only other difference is in -- of course -- the about.html files. I actually removed that reference. I have *no* idea why except perhaps for (a stray mouse-click during maintenance?)

So, let me know the best course of action at this point.

-Miles

On Jun 9, 2011, at 7:29 AM, Kenn Hussey wrote:

We actually build the bundles in question as part of the EMF build but don't intended for them to be installed from a repository (instead, their source is embedded in the examples installer, to be extracted as projects into a user's workspace). The bundles are in fact available in the EMF repository, so I'm not sure why they couldn't be consumed from there by AMP.

The about.html files have been in place in the EMF codebase all along. The reason they didn't appear in the aggregated bundles must have to do with a difference in the AMP build. There are other subtle differences with the way these bundles are built by EMF and AMP (e.g., the way the qualifiers are generated) which suggests another potential concern with code being redistributed by another project...

Kenn

On Thu, Jun 9, 2011 at 9:25 AM, Wayne Beaton <wayne@eclipse.org> wrote:
This isn't explicitly covered by the EDP. In practice, however, this is generally considered acceptable.

If the project is copying the code from another project's repository into their own repository, then a fork has occurred, and the IP team needs to be informed (primarily for tracking purposes). In a general sense, projects are permitted to take the output from other projects and include it in their own distributions. If the project they are consuming is in the incubation phase (does not apply here), then they are obligated to include incubation branding in their own downloads.

FWIW, almost every project distributes something from another project at some level or another.

There is no formal requirement to call this out in release review documentation. But that would definitely be nice to see.

As for responsibilities, I respectfully suggest that AMP and EMF share responsibility to resolve the issues. I include AMP in this because they have taken on code with no implicit warranty (this is based on assumptions stemming from the statement that "it wasn't available from EMF"). I include EMF because, well, it's their code. It seems to me that a patch submitted to EMF by the AMP team should clear an "about files" problem up pretty quickly. 

Hopefully this doesn't turn into one of those hard lessons. If, ultimately, EMF cannot provide the necessary assistance to resolve issues, AMP is left with the responsibility. A project not shipping pieces of their code is a huge red flag for consumers. Early and ongoing communication is important in this sort of situation.

In terms of detection, there's only so much that we can do. We depend very heavily on committers and project leads to do the right thing (and provide assistance where we can).