[Patch] Performance improvements for DatasetMerger

[GerdP]

Sometimes JOSM needs much longer than expected to perform an action, eg. when you have a rather complex relation in the selection and use Update Selection. I wondered what goes on and found that DatasetMerger.merge() is very inefficient when the target dataset contains lots of objects, no matter how many object the source has.
Reason: It creates a list of target nodes which is only needed when the source contains an object that has a negative id.

List<? extends OsmPrimitive> candidates = new ArrayList<>(targetDataSet.getNodes());
...
candidates = new ArrayList<>(targetDataSet.getWays());
...
candidates = new ArrayList<>(targetDataSet.getRelations());

Problem: These lines require repeated iterations over SubclassFilteredCollection instances.

[GerdP]

avoid to materialise unneeded copies of SubclassFilteredCollection

[GerdP]

Forget that patch. If we need the copy we need it from the original data.

[GerdP]

In 17136/josm:

see #19898: Performance improvements for DatasetMerger
Avoid to avoid to materialise unneeded copies of SubclassFilteredCollection

[GerdP]

I've only committed a simple improvement. The copies are still created far too often.

[GerdP]

With this patch the copies are only created when needed, and flow control looks better. With downloaded data this is faster, have to try merges of layers with new data.

[GerdP]

Hmm, to be honest I can't find any case where this patch changes performance compared to the already implemented changes in r17136. So I think we can keep the existing code as well.

[Don-vip]

Milestone renamed