[
https://issues.apache.org/jira/browse/OAK-10812?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17852849#comment-17852849
]
Stefan Egli commented on OAK-10812:
-----------------------------------
{quote}I think that we have to use the parallel minimum instead of the parallel
maximum in this case. {quote}
Right, I think the bug lies in the fact that {{pmax}} was used for {{fromRev}}
*and* {{toRev}} - while it seems like the correct way would have been to use
{{pmin}} for {{fromRev}} and {{pmax}} for {{toRev}}.
Except, in the current state of
[PR#1465|https://github.com/apache/jackrabbit-oak/pull/1465] that's not what
the PR now contains. Instead of switching pmax to pmin it now does a read-only
conditioned ignore of minRevision. I think that following the above mentioned
conclusion, it should actually be a simple flip of
{noformat}
- fromRev = fromRev.pmax(minRevisions);
+ fromRev = fromRev.pmin(minRevisions);
{noformat}
Btw, what might also be useful is to have that new diffManyChildrenReadOnlyMode
test method in two variants (and then maybe call it different) : one for
read-only and one for read-write. I understand the read-write one wouldn't be
able to reuse clusterId 1 - as is currently coded in
diffManyChildrenReadOnlyMode. So it would probably use clusterId 3. I have
tested a read-write variant of that method and it indeed fails with the current
code (that applies only to read-only mode - but in fact, the bug is independent
of read-only mode).
> DocumentNodeStore#diffManyChildren(...) may produce incorrect results in
> readonly mode
> --------------------------------------------------------------------------------------
>
> Key: OAK-10812
> URL: https://issues.apache.org/jira/browse/OAK-10812
> Project: Jackrabbit Oak
> Issue Type: Bug
> Components: documentmk
> Reporter: Manfred Baedke
> Assignee: Manfred Baedke
> Priority: Major
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
