Unwanted Updates when Installing From Updatesite

[tgoethel]

Dear Eclipse community,
I am working with Eclipse 2023-06 (Version: 2023-06 (4.28.0), Build id: 20230608-1333), currently. Recently, I installed an RCP feature, say MyFeature, using an additional updatesite and experienced unexpected behaviour. At the time of installation of this feature, I had version 5.0.1 of google.inject installed in Eclipse.

To reproduce the behaviour, I installed 'MyFeature' in two ways in separate Eclipse instances:

1. When I install it with checkbox "Contact all updatesites ..." disabled, installation is successful and google.inject is still at version 5.0.1

2. When I install it with checkbox "Contact all updatesites ..." enabled, installation is successful, but also google.inject is updated to version 7.0.0.
   
   

I checked that MyFeature is really only available on one updatesite.

As google.inject is not needed to be upgraded as Way 1 shows, I would expect Way 2 to also leave google.inject as it is.

Is my expectation wrong or is there a bug in Eclipse?

Thanks and best wishes,
Thomas

[merks]

We don't know what you have installed and we don't know which update sites are available for updates. Probably this one is available:

That site does have com.google.inject version 7.0.0.

There are a bunch of things you may have installed and that may need updating that require that higher version:

[tgoethel]

(I would actually not have any issues with implicit unnecessary updates. However, google.inject 7.0.0 has a dependency to sun.misc, which I do not have installed)

[merks]

It's not clear from the details provided if the update is unnecessary. That dependency comes form the JRE:

In any case, p2 does have a tendency to install the latest version of things, depending of course on the version range constraints on that dependency.

[tgoethel]

If it was a necessary update, wouldn't it be necessary as well when disabling the checkbox "Contact all updatesites ...", thus, resulting in non-successful installation of "MyFeature"?

[merks]

In general, p2 will take into account the requirements of your feature (the thing(s) being installed or updated) along with the entire set of installable units IUs available from all enabled update sites and will then create a solution that satisfies all those requirements. And, as I said, p2 has a tendency to prefer newer versions over older versions. So yes, it might well update bundles that do not strictly require an update.

[tgoethel]

Thanks for the clarification!

[laeubi]

P2 uses a SAT solver and as SAT is a NP-hard problem usually algorithms implementing such only guarantee that they find ONE solution but not necessarily THE solution in regard to some not formulated constraints, so for a given input there might be different options that satisfy the problem.

So if you check Contact all updatesites you effectively adding more things a solver can choose and that might give different results (even though for that kind of problem if they use a randomized start it is even not guaranteed that the same input always give the same output).

So given what you provided, it seems at least valid to use version 7 here over version 5, so the question is more why is this update "unwanted"? Just because one could find another solution? If yes, this is not a bug (in eclipse).
If it is unwanted because something now fails this is not a bug in Eclipse but in the provided metadata of the failing code (e.g. it should have contained its package imports to not include the next major version).

[laeubi]

That should be fixed in the latest eclipse release (2023-09) and you can just ignore the problem and proceed in older versions.

[tgoethel]

Ok, thank you very much!

[laeubi]

I get an error, because some features cannot be loaded that use google.inject.

Can you provide the error maybe? Its really hard to guess whats going wrong otherwise. Also make sure to at best use the latest Eclipse Release (2023-09) when possible.

[tgoethel]

I am very sorry, but I was too quick when posting my previous post that I also deleted in the meantime. When ignoring the validation error, things work nevertheless as expected.

[merks]

That's what I thought because sun.misc is present in a normal JDK/JRE...