Adrien Grand

> I may add this test case to BasePostingsFormatTestCase, or do you have any other idea on test? 1M documents is too much for a unit test, I was thinking...

@vsop-479 I addressed failures related to formatting but there seems to still be a test failure.

@mikemccand You might want to have a look at this change since (I think) you are one of the most familiar ones with the original code.

I had to revert this change because of test failures, e.g. this seed reproduces on the main branch: ``` gradlew test --tests TestNumericDocValuesUpdates.testSortedIndex -Dtests.seed=4C6E977E1F29E069 -Dtests.locale=khq -Dtests.timezone=Asia/Hong_Kong -Dtests.asserts=true -Dtests.file.encoding=UTF-8 ``` It...

A new PR would be great, thank you for looking into these failures!

> we want a single Field containing a list of key-value pairs or a json formatted Note that you can add one `FeatureField` field to your Lucene document for every...

This is a good point. This limit was introduced with the idea that `FeatureField` would be used to incorporate features into a BM25/TFIDF/DFR score and higher weights than 64 would...

I had missed that the constructor was not publicly available. Am I guessing correctly that your goal is to create custom collectors that work with shapes? Maybe we should make...

I was hoping to get @nknize 's feedback on this since he's been closer to this work, but the API looks ok to me this way.

@uschindler FYI it looks like this change made indexing ~12% faster with `BEST_SPEED` on the stored fields benchmark: http://people.apache.org/~mikemccand/lucenebench/stored_fields_benchmarks.html