harmony-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Regis Xu (JIRA)" <j...@apache.org>
Subject [jira] Resolved: (HARMONY-6651) Broken double checked locking in DragSourceContext.java
Date Mon, 20 Sep 2010 08:12:33 GMT

     [ https://issues.apache.org/jira/browse/HARMONY-6651?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel

Regis Xu resolved HARMONY-6651.

    Fix Version/s: 5.0M16
       Resolution: Fixed

Thanks Wendy!

The fix was applied at trunk r998822, please verify.

> Broken double checked locking in DragSourceContext.java 
> --------------------------------------------------------
>                 Key: HARMONY-6651
>                 URL: https://issues.apache.org/jira/browse/HARMONY-6651
>             Project: Harmony
>          Issue Type: Bug
>          Components: Classlib
>    Affects Versions: 6.0M1
>         Environment: Windows Xp
>            Reporter: Wendy Feng
>            Assignee: Regis Xu
>             Fix For: 5.0M16
>   Original Estimate: 3h
>  Remaining Estimate: 3h
> I found broken double-checked locking in modules/nio_char/src/main/java/java/nio/charset/Charset.java
> public abstract class Charset implements Comparable<Charset>{
> ...
> public static SortedMap<String, Charset> availableCharsets() {
>         // Initialize the built-in charsets map cache if necessary
>         if (null == _builtInCharsets) {
>             synchronized (Charset.class) {
>                 if (null == _builtInCharsets) {
>                     _builtInCharsets = new TreeMap<String, Charset>(
>                             IgnoreCaseComparator.getInstance());
>                     _builtInProvider.putCharsets(_builtInCharsets);
>                 }
>             }
>         }
>     ...
>    }
> ...
> }
> The double checked locking idiom is broken in the code above. In current Java memory
model (JMM), the writes initializing the TreeMap object and the write to the _builtInCharesets
field could be reordered by the compiler, processor or memory system. It means the write to
_builtInCharesets can happen before the writes in TreeMap object initialization. 
> Consequence:
> When multiple threads access the code simultaneously, threads can see a non-null _builtInCharesets
field, but the TreeMap can be partially constructed (e.g. The elementData array is not created
yet). Then these threads invoke clone() method on the partially initialized TreeMap, which
will result in undesired behavior, e.g. Throwing NullPointerException. The error doesn't manifest
all the time due to the non-deterministic nature of multi-threading, but it would be too bad
to leave the code depend on the undefined behavior.
> In JDK 5 or above, declaring the _builtInCharesets field as volatile can fix the problem.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message