The lazy loading model is likely to cause problem to the UI.
Isn't this problem of loading repos in the IU more general (e.g I add a site)
Could we have some of the sub repos of a composite be loaded eagerly?
Version change has been released
Actual change request to be released
Lazy Collectors, Collector Status
Metadata interfaces
Attendees
Susan Franklin
Ian Bull
Thomas H.
Pascal R.
Minutes
New version class (bug 233699)
Simon enabling patch is released.
Thomas mentioned that the proposal is getting close to be complete and he has an implementation of it mostly ready. We will probably look into releasing this in the new year.
Pascal requested that before the new implementation of version is released a complete test suite testing the OSGi version semantics needed to be added.
Actual change request (bug 218055)
Pascal will be releasing change to the ProvisioningPlan later today.
The API on RequestResult will be evolved to provide explanation support when we get proper explanation
Explanation support (Bug 200380)
Jed is reporting that some good progress is being done on the explanation work.
Current changes are contained into sat4j.pb, but it may be extracted into a separate bundles for ease of maintenance for the SAT4J team
The current approach for explanation is built directly on top of the opb file (no direct interfacing with SAT4J). One worry there was bug reporting. Pascal mentioned that from a SAT4J model, it is possible to generate an opb file. Jed will investigate.
Jed also mentioned that he would like to see if the optimization function could be reworked to be a least perturbation one over an eager update one like we do today (the current eager behavior is the source of some problems)
Composite repos
Susan has renewed her worries about composite repos lazily loading sub repos.
This represents a contract change from what we had agreed to in 3.4. Changing this contract can result in a lot of work that needs to be planned for. We need to do it intentionally, not accidentally. For example this changes the assumption on how the filtering is done and where we make the user wait.
This is a time bomb for the UI which now assumes that a lot of the repos are being loaded in the background before loading the UI.
Given the state of affair, we will make composite repos load sub repos eagerly for now.
This is an interim situation that may last until 3.5 ships
A similar problem can be exhibited by remote repos like those Ian and Thomas are talking about.
A solution to this problem may have to do with changes for bug 256411
Lazy collectors (bug 256411)
Ian has been looking in making Collectors to be lazy.
He thinks that most collectors could be changed. However a problem arise with Collectors that do work on their accept method. He has explored the path of Acceptors but think this is not possible
He will look at adding a Status on Collectors in order to communicate back problems while using the collector (e.g. time out, etc)
New patch being added soon by Ian and to be reviewed by Jeff
Pascal has requested that before committing we have tests reading non p2-impl of IUs and writing them as p2-IUs and also have a test case that mix and match p2-IUs with non p2-IUs and try to install and persist
