summaryrefslogtreecommitdiff
path: root/cppuhelper
diff options
context:
space:
mode:
authorStephan Bergmann <sbergman@redhat.com>2014-12-09 12:51:40 +0100
committerStephan Bergmann <sbergman@redhat.com>2014-12-09 13:17:03 +0100
commit6ddde10b4006ece33bc358a391a13e108a35f6fa (patch)
tree2a1b66691b588c2cb931934133fc911b052ae61b /cppuhelper
parent4c66ce8ec8f67d3bc2db5c79de68f4bda26d2e60 (diff)
rhbz#1036877: Join Java AsynchronousFinalizer thread well before exit
AsynchronousFinalizer was originally added as 870a4401c05beec3d31c1f6055a64591edd0a9d9 "INTEGRATION: CWS mtg1: #i57753# Avoid long-running finalize methods" referring to <https://issues.apache.org/ooo/show_bug.cgi?id=57753> " Fix JNI-UNO bridge so that the JVM doesn't run out of memory when a destructor locks the SolarMutex." It is unclear to me how relevant "If JVMs are getting more mature and should no longer have problems with long-running finalize methods, this class could be removed again" really is in practice. After all, advice on hotspot-gc-devel is to avoid finalize() if possible (<http://mail.openjdk.java.net/pipermail/hotspot-gc-dev/2014-June/010215.html> "Re: History of finalizer execution and gc progress?"). So stick with this approach of home-grown draining for now (where a home-grown approach using PhantomReferencens would need a dedicated draining thread, too, so would not have much benefit over the existing code in practice). Timely termination of AsynchronousFinalizer threads is achieved by using a dedicated thread per bridge and joining it in the remote bridge's dispose() resp. the JNI environment's new java_env_dispose. Change-Id: Idcef2dbf361a1de22f60db73828f59e85711aea7
Diffstat (limited to 'cppuhelper')
-rw-r--r--cppuhelper/source/component_context.cxx16
1 files changed, 16 insertions, 0 deletions
diff --git a/cppuhelper/source/component_context.cxx b/cppuhelper/source/component_context.cxx
index 16642e9133fb..53b5eb000468 100644
--- a/cppuhelper/source/component_context.cxx
+++ b/cppuhelper/source/component_context.cxx
@@ -731,6 +731,22 @@ void ComponentContext::disposing()
for ( ; iPos != iEnd; ++iPos )
delete iPos->second;
m_map.clear();
+
+ // Hack to terminate any JNI bridge's AsynchronousFinalizer thread (as JNI
+ // proxies get finalized with arbitrary delay, so the bridge typically does
+ // not dispose itself early enough before the process exits):
+ uno_Environment ** envs;
+ sal_Int32 envCount;
+ uno_getRegisteredEnvironments(
+ &envs, &envCount, &rtl_allocateMemory, OUString("java").pData);
+ assert(envCount >= 0);
+ assert(envCount == 0 || envs != nullptr);
+ for (sal_Int32 i = 0; i != envCount; ++i) {
+ assert(envs[i] != nullptr);
+ assert(envs[i]->dispose != nullptr);
+ (*envs[i]->dispose)(envs[i]);
+ }
+ rtl_freeMemory(envs);
}
ComponentContext::ComponentContext(