[
https://issues.apache.org/jira/browse/LUCENE-9001?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16948283#comment-16948283
]
Adrien Grand commented on LUCENE-9001:
--------------------------------------
Your points are right, which is why I said I was inclined to merge your change.
On the other hand, SetOnce wasn't designed as a general concurrency utility,
and we don't have this conditionally-set use-case that exhibits the issue.
While I have no doubt trySet has use-cases and would be faster than
implementing on top of get and set, this would be dead code to Lucene. I don't
have strong feelings though, there are precedents e.g. not all FixedBitSet
methods are used by Lucene but I wanted to highlight these points in case other
committers have opinions. You seem to feel strongly about introducing trySet
which sounds ok to me. I plan to merge this change in a couple days if nobody
objects.
> Race condition in SetOnce
> -------------------------
>
> Key: LUCENE-9001
> URL: https://issues.apache.org/jira/browse/LUCENE-9001
> Project: Lucene - Core
> Issue Type: Bug
> Reporter: Przemko Robakowski
> Priority: Minor
> Time Spent: 10m
> Remaining Estimate: 0h
>
> There is race condition in SetOnce that can cause code below fail with
> NullPointerException:
> {code:java}
> SetOnce<String> setOnce = new SetOnce<>();
> new Thread(() -> setOnce.set("thread")).start();
> try{
> setOnce.set("main");
> } catch (SetOnce.AlreadySetException e){
> setOnce.get().hashCode(); //possible NPE!
> }
> {code}
> This is caused by 2 separate write operations - 1 for set marker field and 1
> for actual object. So it's possible that marker is already set to true
> (causing AlreadySetException on second write attempt) but object is still not
> set (causing NullPointerException).
> This can be avoided by using single AtomicReference instead to serve both
> purposes.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
