Details
-
Type:
Bug
-
Status:
Closed
-
Priority:
Major
-
Resolution: Fixed
-
Affects Version/s: 1.0
-
Fix Version/s: 1-1-beta-2
-
Component/s: None
-
Labels:None
-
Environment:Example: High Traffic Webapp, context-scoped container with non-lifecycle singleton
-
Number of attachments :
Description
Currently CachingComponentAdapter doesn't do anything to make sure the underlying ComponentInstance is only created once.
I'd like to get some comments on how we should handle this. I think the safest way is to just make CachingComponentAdapter.getComponentInstance synchronized but there might be some people who don't like that (which I can understand). However everything else might leak.
I got a testcase to show the problem. (Requires Doug Lea's concurrent libary since I wanted a safe countdown 
Activity
Thomas Heller
made changes -
| Field | Original Value | New Value |
|---|---|---|
| Attachment | PicoRaceTest.java [ 12331 ] |
If you do:
pico.registerComponent(new SynchronizedComponentAdapter(new CachingComponentAdapter(new ConstructorInjectionComponentAdapter("slow", SlowCtor.class))));
it passes (put the sync adapter at the outer level)
Aslak Hellesøy
added a comment - If you do:
pico.registerComponent(new SynchronizedComponentAdapter(new CachingComponentAdapter(new ConstructorInjectionComponentAdapter("slow", SlowCtor.class))));
it passes (put the sync adapter at the outer level) Use a SynchronizedCA, see unit tests.
Jörg Schaible
added a comment - Use a SynchronizedCA, see unit tests. Jörg Schaible
made changes -
| Status | Open [ 1 ] | Closed [ 6 ] |
| Fix Version/s | 1-1-beta-2 [ 11243 ] | |
| Resolution | Fixed [ 1 ] |
peter royal
made changes -
results may vary depending on your system. On one of my systems the test-case passes fine until I adjust the Thread.sleep to about 500. On my main machine I can run it with Thread.sleep(2) and it will fail.