I opened a PR at https://github.com/apache/lucene/pull/12400 with a
fix, I tried to explain in the PR description why AssertingScorer has
this check. Even though it's not documented in BulkScorer#score, I
think it's a good check to keep.
On Wed, Jun 28, 2023 at 6:25 AM Adrien Grand wrote:
>
>
Thanks Patrick, I will look into it this morning.
Le mer. 28 juin 2023, 06:20, Patrick Zhai a écrit :
> Yeah I think that's the commit, I'm definitely not an expert on scorer as
> well so maybe @jpou...@gmail.com could you take a
> look?
>
> Patrick
>
> On Tue, Jun 27, 2023 at 5:34 AM Michael
Yeah I think that's the commit, I'm definitely not an expert on scorer as
well so maybe @jpou...@gmail.com could you take a look?
Patrick
On Tue, Jun 27, 2023 at 5:34 AM Michael McCandless <
luc...@mikemccandless.com> wrote:
> Thanks for digging Patrick!
>
> I sort of think MaxScoreBulkScorer
Thanks for digging Patrick!
I sort of think MaxScoreBulkScorer should be returning NO_MORE_DOCS in this
case? But I'm far from an expert. This may be related to the recent
MAXScore improvements for disjunctions? (
https://github.com/apache/lucene/commit/8703e449cee0693e50a7922a86c1cbc7dcf95d13
The exception was thrown because TimeLimitingBulkScorer passed in a "max"
which is larger than the maxDoc in the segment. And then MaxScoreBulkScorer
directly returns the rangeEnd as the next estimation here