mikemccand commented on PR #13656:
URL: https://github.com/apache/lucene/pull/13656#issuecomment-2296609036
> Have you followed what @epotyom and I found so far in #13657?
Yes! And it's completely insane... it might be time to turn on
`-XX:+PrintAssembly` and watch how hotspot is spe
gsmiller commented on PR #13656:
URL: https://github.com/apache/lucene/pull/13656#issuecomment-2291752889
Totally @mikemccand . Have you followed what @epotyom and I found so far in
#13657? It's wild. Just moving the initialization of the array list before or
after the call to createWeight
mikemccand commented on PR #13656:
URL: https://github.com/apache/lucene/pull/13656#issuecomment-2291583847
Thanks @gsmiller -- looks like this did indeed fix the nightly benchy e.g.
[CombinedTerm](https://home.apache.org/~mikemccand/lucenebench/CombinedTerm.html),
[TermQuery](https://home.
gsmiller commented on PR #13656:
URL: https://github.com/apache/lucene/pull/13656#issuecomment-2291395658
Investigation continues on how #13568 could possibly have caused a
regression to various nightly benchmark tasks, but I've confirmed that this
patch fixed the issues (see
[Term](https:
gsmiller merged PR #13656:
URL: https://github.com/apache/lucene/pull/13656
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@lucene.ap
gsmiller commented on PR #13656:
URL: https://github.com/apache/lucene/pull/13656#issuecomment-2289336382
Because this change is small (just undoing a small bit of #13568 in
IndexSearcher) and appears to correct the performance regressions the nightly
benchmarks demonstrated, I'm going to m
gsmiller commented on PR #13656:
URL: https://github.com/apache/lucene/pull/13656#issuecomment-2289330332
Also, benchmarks (wikimedium10m) comparing this PR to the tip of main seems
to indicate it reverses the performance impact we observed in the nightly runs:
```
gsmiller commented on PR #13656:
URL: https://github.com/apache/lucene/pull/13656#issuecomment-2289311663
Benchmarks (`wikimedium10m`) comparing this change to
`9b481f76f77b40a5061fb3b70763e4b4a06d31d0` (last commit on main before #13568)
show no performance regression, which is what we're
