[
https://issues.apache.org/jira/browse/CAY-2895?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18020353#comment-18020353
]
Nikita Timofeev commented on CAY-2895:
--------------------------------------
Hi [~soliax] thank you for such a detailed report! I'm going to look into this
soon and hopefully have a fix.
> Incorrect Lazy Pagination Comparison for BigInteger PK
> ------------------------------------------------------
>
> Key: CAY-2895
> URL: https://issues.apache.org/jira/browse/CAY-2895
> Project: Cayenne
> Issue Type: Bug
> Components: Core Library
> Affects Versions: 4.2.2
> Reporter: lallemand
> Assignee: Nikita Timofeev
> Priority: Critical
> Fix For: 4.2.3, 5.0-M2
>
> Attachments: image-2025-09-01-14-24-18-956.png,
> image-2025-09-01-14-24-41-442.png, image-2025-09-01-14-26-02-296.png,
> image-2025-09-01-14-26-35-313.png, image-2025-09-01-14-26-45-722.png,
> image-2025-09-01-14-27-06-177.png, image-2025-09-01-14-27-12-756.png,
> image-2025-09-01-14-27-27-432.png, image-2025-09-15-14-00-38-120.png,
> image-2025-09-15-14-02-55-828.png, image-2025-09-15-14-22-47-825.png
>
>
> Dears,
>
> In the latest stable version ({*}4.2.2{*}), we are encountering an issue with
> the *Lazy implementation* based on {{ObjectSelect<T>.pageSize()}} when the
> table’s *primary key* is of type {*}BIGINT{*}.
>
> ----
> *Steps to Reproduce:*
> # Use {{ObjectSelect<T>}} with pagination ({{{}pageSize{}}}) on a table
> where the primary key column is of type {*}BIGINT{*}.
> # Allow the {{IncrementalFaultList}} in *cayenne-server* to populate
> elements.
> # Insert a new item into the dataset.
> ----
> *Technical Details:*
> * The issue occurs in {*}{{IncrementalFaultList.class}}{*}, specifically
> within the {{fillIn}} method, which initializes {{{}protected final List
> elements{}}}.
> * When processing a *BIGINT* database PK mapping, the type defaults to
> {*}{{Long}}{*}.
> * This works fine until a new item is added. At that point, the method
> {{updateWithResolvedObjectInRange}} is called, which in turn calls
> {{replacesObject}} (for one of six implementations).
> In our case, the class *{{SimpleIdIncrementalFaultList}}* is used.
> Example:
> * {{objectInTheList}} → value: {{{}137{}}}, type: *Long*
> * {{idSnapshot.get(pk.getName())}} → value: {{{}137{}}}, type: *BigInteger*
> *!image-2025-09-01-14-24-18-956.png!*
> Because of the type mismatch, the comparison fails, leading to the following
> exception:
> !image-2025-09-01-14-24-41-442.png!
> As a result, the list contains a *Long* object instead of the expected
> *ObjEntity* inside our lazy implementation{*}.{*}
> ----
> *Proposed Fix / Workaround:*
> I modified the {{replacesObject}} method to handle *BigInteger* correctly.
> !image-2025-09-01-14-26-02-296.png!
> * *First modification (blue section in my test code):*
> _Not recommended._ This forces replacement regardless of match. I added it
> only to bypass an unrelated bug I observed occasionally. This may indicate a
> deeper issue in the pagination implementation that requires further
> investigation.
> * *Second modification (red section in my test code):*
> _Required for correctness._ This ensures that *BigInteger* primary keys are
> compared properly with Long values.
> While this fix works, I believe a more elegant solution should exist within
> Cayenne’s type handling logic.
> ----
> *Test Configuration:*
> * *Databases:*
> Oracle
> !image-2025-09-01-14-26-35-313.png!
> SQL Server
> !image-2025-09-01-14-26-45-722.png!
> * *Entities:*
> {{DbEntity}}
> !image-2025-09-01-14-27-12-756.png!
> ObjEntity !image-2025-09-01-14-27-27-432.png!
> ----
> *Impact:*
> This is a {*}critical issue{*}: pagination becomes completely unusable when
> the primary key is of type {*}BIGINT{*}.
> ----
> *Edit 1:*
> I’ve discovered that the result of this function can sometimes be
> inconsistent if not enough time is allowed for it to complete. This behavior
> seems to be the root cause of the issue highlighted in blue in my previous
> review.
> !image-2025-09-15-14-00-38-120.png!
> Called by
> !image-2025-09-15-14-02-55-828.png!
> When I introduce a slight slowdown in the code execution, the behavior
> appears more stable during the first load. This is related to the issue
> “First modification (highlighted in blue in my test code)" section.
> My environment:
> * Arch Linux
> * Containerized databases (Oracle / SQL Server)
> The problem seems to be that the execution is sometimes “too fast,” which
> leads to certain items not being matched during the initial load. I’ve spent
> hours debugging but haven’t been able to identify the exact culprit. The only
> reliable observation so far is that slowing execution slightly improves
> stability (though this is clearly not a recommended fix).
> This suggests there may be an underlying timing, synchronization, async or
> readiness issue. Further investigation into the root cause would be helpful.
>
> ----
> *NB:*
> *I had to disable this test when using blue change, so I do understand that
> my fixe cannot be used as it, because it's seem to be breaking your logic
> greatly. But without the blue fix, I'm not able to have a stable environement
> on my side when using lazy.*
> !image-2025-09-15-14-22-47-825.png!
>
> ----
>
> *Best regards,*
> Anton L.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
