[MINDEXER-191] Avoid group rebuilding during incremental updates

[mbien]

Rebuilding of the in-memory group data takes more time than the index update itself since it has to iterate through the full index.

What is worse is that rebuilding happens in a loop for each chunk:

maven-indexer/indexer-core/src/main/java/org/apache/maven/index/updater/DefaultIndexUpdater.java

Lines 428 to 430 in fbf7c0c

	public void addIndexChunk(ResourceFetcher source, String filename) throws IOException {
	loadIndexDirectory(updateRequest, source, true, filename);
	}

The reader has all group data of the extracted incremental update already, while the context should have the group data of the current index.

If we assume that maven group IDs are rarely removed we can simply merge both collections without having to rebuild anything.

• typical maven central index updates are >4x faster without group rebuilding (<70s on an i6700k)
• minor optimizations and deprecation fixes (additional ~10% faster merge)

context:

#202 moved the groups from index to memory storage which is why NetBeans is persisting the group data after maven-indexer did the extraction apache/netbeans#4136. When the index context is loaded the group data is restored by NB. Incremental updates can simply add more groups to that collection.

Other use cases might want to call IndexingContext#rebuildGroups() explicitly, if this isn't done, the context will have only the groups of the incremental update, post update.

---

https://issues.apache.org/jira/browse/MINDEXER-191