Hi,
I misunderstood LFV-140, actually it is not problem:
https://shbhar.github.io/lucene-ai-bug-hunt/repro/LFV-130/
This one complains about *after* an EOF happened, the IndexInput behaves
"wrong" under some circumstances. It might be improved, but the trick I
mentioned before with reporting correct exception is fine. Actually once
an IOException happened, an IndexInput should no longer be used because
its state may be corrupted. Plain simple! Maybe we should add
documentation to IndexInput base class that after an IOException the
state of an IndexInput is "undefined" and spurious exceptions may or may
not happen.
If I have a cheap way how to improve that (so it will always throw
EOFException), I will patch it, but otherwise this issue is rejected.
Uwe
Am 29.10.2026 um 17:28 schrieb Shitij Bhargava:
Thank you Uwe.
There are also some apparently unreported bugs that only appeared in
10.6 and were not in 10.5.2 (5 medium priority and others low
priority, based on AI's classification) - I've updated the website to
be able to filter to only these with the "Only bugs new in 10.6"
option if it helps.
I don't want to create unnecessary confusion for everyone at the last
moment (I see 10.6 release is imminent), but if someone can glance at
them for a sanity check, it might be helpful. I'm also prioritizing
understanding these first (starting with
https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-002.html)
Thanks,
Shitij
On Thu, 29 Oct 2026 at 01:06, Uwe Schindler via dev
<[email protected]> wrote:
Hi,
the wrong offset in error message for multi-mmap may be fixed easily, I
will have a look. Actually in the recent version the wrong offsets are
reported through a private method throwing the exception. I can override
that in the subclass. I have to closer look, but I will post a PR once
ready.
Anyways: I will have a look at all the mmap-related bugs.
Uwe
Am 29.10.2026 um 06:27 schrieb Shitij Bhargava:
Quick update:
Refreshed the site with the above 2 bugs to be correctly classified as
previously known instead of new since they were known privately (and
triaged to in-progress and wont fix).
Also Uwe for your feedback in
https://github.com/apache/lucene/pull/16931#issuecomment-6476410908
for unnecessary tests, I'm instructing the agent to be more
disciplined/minimal with a second pass in not just the repro but also
in its unit tests (currently in progress)
Thanks,
Shitij
On Wed, 28 Oct 2026 at 14:45, Shitij Bhargava<[email protected]> wrote:
Thank you Uwe and Chris!
For automatically refreshing, I think it would help if the agent can
just poll 1 link/github issue and keep the website updated with new
triages, fixes, verified/opened issues etc. So I've made it here:
https://github.com/apache/lucene/issues/16937
For the two issues you mentioned, I will work on refreshing the
website for both (like adding a triaged/triaged by column?) and update
here. For LFV-037 (shared arena one) it will update to fix in progress
in #16931 & for LFV-135 (multi-mmap offset) it will say "won't fix".
For future updates just linking to the above issue or triaging there
in natural language would work automatically as well.
If it makes sense I can make it so that an expert in certain area can
search by their name or github handle and it will show a list of bugs
which might be related to their expertise (based on AI's assessment)
instead of having to manually scan the whole table.
Let me know if any other features in the website or the whole process
would help!
Thanks,
Shitij
On 2026/10/28 17:45:52 Uwe Schindler via dev wrote:
Hi,
here is the first PR fixing that issue, thanks to Chris!
https://github.com/apache/lucene/pull/16931
Not sure how to keep track of this resolved issue in your AI report:
https://shbhar.github.io/lucene-ai-bug-hunt/repro/LFV-037/
Uwe
Am 28.10.2026 um 15:16 schrieb Uwe Schindler via dev:
Hi,
thanks for the great web site with all those bugs. I scrolled through
them and of course stopped on the MMapDirectory related ones. The one
with reporting false offsets on random access multi-mmap was
personally known to me, but I never decided to fix it, because the
error message is valid, just a wrong offset reported, which may
prevent easy debugging, but on other hand to have multi-mmap indexes,
the index segment must be larger than 16 GiB and that no longer
happens unless you force merge. Therefor I'd ignore this one.
The shared arena one looks very relevant and I always had the
"feeling" that something is wrong with them. It was also reported to
me privately by Gianluca Graziadei on/after Apache Community over Code
already (or at least what I got from the report is closely related as
it causes similar symptoms). I don't understand the issue fully, so
maybe Chris Hegarthy can help me fixing it.
Uwe
Am 27.10.2026 um 19:42 schrieb Shitij Bhargava:
Hi all,
I work in Amazon advertising where we use Lucene for a lot of our ad
candidate retrieval services. We also develop custom codecs and other
customizations for cases where the stock Lucene options aren't
enough. While developing one of them, I asked an AI agent to find
bugs in my POC implementation, and it ended up finding an unreported
bug in Lucene itself.
I kept going with this approach and it kept finding more unreported
bugs, so I've gone deeper into AI-assisted bug finding with formal
verification and other techniques. As of today the agents have found
592 bugs in Lucene (and counting) that appear to be unreported in
GitHub or the mailing list. It also found (or already knew from its
training data) nearly 200 bugs that have already been reported, which
I'm not counting here. To avoid noise in these findings, the approach
has each potential bug independently and adversarially verified by a
second model (Claude vs GPT) with a minimal unit test repro and
making sure the behavior is not intended from past discussions in
mailing lists, tests, GitHub discussion & commit/PR history.
Many of them seem to be minor or low impact (392 low, 196 medium, 4
high classified by AI), but a few look important. For example:
- After a flush where the only changes are doc-values updates,
openIfChanged returns null and isCurrent() stays true so the update
can be missed:https://github.com/apache/lucene/issues/16900
- HNSW merge-time graph reuse can leave the entry node with no
neighbours, which drops kNN recall to zero in 10.4.0 and later:
https://shbhar.github.io/lucene-ai-bug-hunt/bugs/LFV-005.html
I've posted the full running list with some categorization here:
https://shbhar.github.io/lucene-ai-bug-hunt/. I've excluded a few
that might be security sensitive according to AI, will send them
privately to security@ soon.
I'm working on understanding, triaging and verifying them myself,
then filing issues on GitHub in my own words in line with Lucene's AI
policy (so far I've filed only one). The agents help by preparing
minimal repro tests, diagrams and short explainer videos (not on site
yet) for surrounding context, but understanding and verifying each
bug properly still takes a lot of time, and the community's help
would make a big difference.
Creating this thread to start a discussion on how to validate and
make these findings actually useful, in particular:
- Could folks help in spot-checking some bugs for areas they might be
familiar with?
- Is it beneficial to track this work in an umbrella github issue?
Thanks,
Shitij
--
Uwe Schindler
Achterdiek 19, D-28357 Bremen
https://www.thetaphi.de
eMail:[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail:[email protected]
For additional commands, e-mail:[email protected]
--
Uwe Schindler
Achterdiek 19, D-28357 Bremen
https://www.thetaphi.de
eMail:[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail:[email protected]
For additional commands, e-mail:[email protected]
--
Uwe Schindler
Achterdiek 19, D-28357 Bremen
https://www.thetaphi.de
eMail:[email protected]
