shbhar commented on issue #16937: URL: https://github.com/apache/lucene/issues/16937#issuecomment-6695504892
It's been over a week since we started tracking this exercise, so I want to take the time to summarize progress from the community. ## Status by findings First, thank you to everyone who has picked up a finding and spent their time in fixes and reviews. Here is a table summarizing the findings and their status so far: | Finding | What | Since | Status | Upstream | Comment | |---|---|---|---|---|---| | [LFV-155](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-155.html) | `DoubleDocValuesStats` starts its max at `Double.MIN_VALUE` | 6.4.0 (2016) | Fixed (11.0) | #16971 by Simranjeet Singh, reviewed by Uwe Schindler and Shitij Bhargava, merged by Uwe Schindler | | | [LFV-663](https://github.com/shbhar/lucene-ai-bug-hunt/tree/main/repro/LFV-663) | `BrazilianStemmer` drops its "logias" rule | 1.9 (2006) | Fixed (11.0) | #16972 by Simranjeet Singh, reviewed by Uwe Schindler and Shitij Bhargava, merged by Uwe Schindler | | | [LFV-037](https://github.com/shbhar/lucene-ai-bug-hunt/tree/main/repro/LFV-037) | MMapDirectory loses shared-arena grouping | 9.12.0 (2024) | Fixed, backported to 10.6.1 | #16931 and backport #16945 by Chris Hegarty, reviewed by Uwe Schindler | Found independently | | [LFV-297](https://github.com/shbhar/lucene-ai-bug-hunt/tree/main/repro/LFV-297) | `RescoreTopNQuery` throws instead of rescoring | 10.3.0 (2025) | Fixed | #16926 by Greg Miller, reviewed by Egor Potemkin | Found independently | | [LFV-288](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-288.html) | `UnifiedHighlighter.Builder#withFlags` is ignored | 9.1.0 (2021) | Fixed | #16983 by Christoph Büscher, reviewed and merged by Luca Cavanna | Found independently | | [LFV-001](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-001.html) | NRT reader reported as current after doc-values updates | 7.0.0 (2017) | PR open | #16923 by Di Wu (for #16900, opened by Shitij Bhargava), reviewed by Shitij Bhargava | | | [LFV-013](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-013.html) | Concurrent doc-values updates can make a merge throw an NPE | 7.0.0 (2017) | PR open | #16989 by Jim Ferenczi | | | [LFV-002](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-002.html) | Merge can lose an acknowledged doc-values update | 10.6.0 (2026) | PR open | #16988 by Jim Ferenczi | | | [LFV-700](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-700.html) | `BlendedInfixSuggester` loses a suggestion when a duplicate arrives | 4.7.0 (2014) | PR open | #16993 by Simranjeet Singh | | | [LFV-1079](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-1079.html) | Vector-similarity value sources ignore the function in `equals` | 9.7.0 (2023) | PR open | #16974 by Simranjeet Singh, reviewed by Shitij Bhargava | | | [LFV-044](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-044.html) | Doc-values range query with a skip index returns wrong hits | 10.6.0 (2026) | PR open | #16896 by kkewwei | Found independently | | [LFV-358](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-358.html) | Sequential DrillSideways skips a two-phase check | 5.3.0 (2015) | PR open | #16981 by Eugene Rizhkov | Found independently | | [LFV-284](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-284.html) | Classic QueryParser: a stopword before AND changes results in 10.6 | 10.6.0 (2026) | Issue open, no replies yet | #16950, opened by Shitij Bhargava | | | [LFV-130](https://github.com/shbhar/lucene-ai-bug-hunt/tree/main/repro/LFV-130) | Seek past EOF on an mmap input after a read hit EOF | - | By design (docs to follow in #16951) | Triaged by Uwe Schindler, who opened #16951 | | ### In terms of numbers | | Oct 28 (this issue opened) | Nov 6 | |---|---:|---:| | New findings | 693 | 1,040 | | With a full write-up | 313 | 1,035 | | Already known | 244 | 253 | | Fixed upstream (from the list + independently) | 0 | 2 + 3 | | Open fix PRs (from the list + independently) | 0 | 5 + 2 | | Good first fixes | not tagged yet | 270 | | New in 10.6.0 (not in 10.5.2) | not tagged yet | 23 | ## Website improvements These findings are not useful unless we optimize their presentation and make it easier for contributors to make decisions, repro, design fixes etc. So website improvements are as important as the findings themselves or more. Here is a summary of what changed: 1. **Searchability with deep linking** If you're only interested in mmap related issues, you can search and share/link to it directly with something like https://shbhar.github.io/lucene-ai-bug-hunt/?q=mmap. Any filters are also similarly deep linkable and chainable like for "high priority Vector search issues": https://shbhar.github.io/lucene-ai-bug-hunt/?priority=high&q=vector 2. **New in 10.6.0** Since 10.6.0 just got released, it might be useful to scan issues the agent found that appear to be regressions from 10.5.2 and if confirmed, fix them in 10.6.1, the deep link for such issues is like: https://shbhar.github.io/lucene-ai-bug-hunt/?new106=1 3. **New "Good first fix" label** there are a lot of issues which are straightforward fixes that are ideal for new contributors (like #16972) and can help grow the community, so there is a new filter on the website to directly go there: https://shbhar.github.io/lucene-ai-bug-hunt/?gff=1 4. **Much more coverage for AI generated explainer docs** The explainer docs ([example](https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-001.html)) include much more detail than the simple repro docs we had, which can be very useful in quickly understanding the finding (even if complicated) including with diagrams, the area around the finding, history of the behavior across versions, suggested fixes etc. For the findings I have verified and reviewed it helped me a lot and I hope it is the same for others 5. Misc usability improvements like clear coloring of high/medium/low impact ## Reflecting on the agent workflow and process so far Thanks to @sjs004 for starting a discussion on how we can use a [similar exercise in OpenSearch](https://github.com/opensearch-project/OpenSearch/issues/23551). It made me think more and articulate what has and hasnt worked well so far and how to improve this process both for the benefit of Lucene itself but also make it more transferable to other projects. I've put my current thinking here: https://shbhar.github.io/lucene-ai-bug-hunt/howto.html with the core principles, methods and an initial draft of [md instructions](https://github.com/shbhar/lucene-ai-bug-hunt/blob/main/HOWTO.md) which can help in getting started. Such an exercise still needs human judgement to steer the workflow properly for the individual needs of a project (which can be very different from Lucene!) so these are just starting points, and Im sure there are opportunities for improvements. I hope the OpenSearch exercise, discussions and reflections from the community on what works and does not can help us improve the process for Lucene too both for the models and tools available today and as they change over time. As always, any feedback or suggestions on how to improve the process/techniques to have greater benefit for Lucene and the community is always welcome! -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
