log4cplus / Bugs / #343 Crash on app exit

Václav Haisman - 2017-07-10

labels: --> java

assigned_to: Václav Haisman
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Václav Haisman - 2017-07-10

labels: java --> java, v1.2.0
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Václav Haisman - 2017-07-10

I am not sure what to think about this. Could you please try HEAD of 1.2.x branch or 2.0.x branch? There have been some initialization and shutdown related changes since the 1.2.0 release. If just using newer log4cplus does not help, could you please try and run your app/java under a debugger with Microsoft symbols path set up so that we can see more of the contex, if at all possible?

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Ralf Sieger - 2017-07-11

Tried it with log4cplus-1.2.1-rc1
Same result. But small difference in the call stack.

I've also attached the MSVC Debugger with log4cplus symbols and got:

First-chance exception at 0x7560A9F2 in javaw.exe: Microsoft C++ exception: Concurrency::scheduler_resource_allocation_error at memory location 0x46BFEF34.
Ausnahme (erste Chance) bei 0x58E2591B (jvm.dll) in javaw.exe: 0xC0000005: Zugriffsverletzung beim Lesen an Position 0x00000010

hs_err_pid6884.log

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Václav Haisman - 2017-07-11

Thank you for the additional information, I think I now understand the situation. We have seen this before. The issue is with how are C++11 synchornization primitives implemented in Visual Studio.

As a stop gap measure, you can modify the log4cplus source like the following diff, for 1.2.x branch:

diff --git a/include/log4cplus/config/win32.h b/include/log4cplus/config/win32.h index 8e5175cd..aa4b0202 100644 --- a/include/log4cplus/config/win32.h +++ b/include/log4cplus/config/win32.h @@ -162,7 +162,7 @@ # endif # if _MSC_VER >= 1700 # define LOG4CPLUS_HAVE_CXX11_ATOMICS -# define LOG4CPLUS_WITH_CXX11_THREADS +//# define LOG4CPLUS_WITH_CXX11_THREADS # endif #endif

This will however not apply on 2.0.x branch because I have removed the non-C++11 implementations of synchornization primitives. I will either have to re-introduce them or...I do not know what else yet.

I am keeping this open to remind me about this issue.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

labels: java, v1.2.0 --> java, v1.2.0, c++11, threads, mutex
Description has changed:

Diff:

--- old
+++ new
@@ -1,24 +1,27 @@
 log4cplus:
-- 1.2.0
-- Visual Studio 2012
-- config: release and debug
-- windows 10, i7 4770
+
+* 1.2.0
+* Visual Studio 2012
+* config: release and debug
+* windows 10, i7 4770

 client:
-- 32 bit Java 7 application
-- log4cplus as dll
-- log4cplus.lib and log4cplusD.lib
-- multibyte chars
-- see attached logfiles
-- client DLL compile settings: no MFC, no ATL, multibyte charset, /MDd and /MD
+
+* 32 bit Java 7 application
+* log4cplus as dll
+* log4cplus.lib and log4cplusD.lib
+* multibyte chars
+* see attached logfiles
+* client DLL compile settings: no MFC, no ATL, multibyte charset, /MDd and /MD

 We use log4cplus in an custom C++ dll which itself gets dynamically loaded in java program.
 during the runtime everything works ok.
 At application exit we see an java hotspot error (see logs). 
-This happens with and without calling log4cplus::Logger::shutdown();
-The problem also occurs if nothing but log4cplus::initialize(); and log4cplus::Logger::shutdown(); are called.
+This happens with and without calling `log4cplus::Logger::shutdown();`
+The problem also occurs if nothing but `log4cplus::initialize();` and `log4cplus::Logger::shutdown();` are called.

 From the hotspot error it looks like an cleanup issue:
+
 ~~~
 Stack: [0x46760000,0x467b0000],  sp=0x467aee5c,  free space=315k
 Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)

Václav Haisman - 2017-07-11

This is basically the issue: https://stackoverflow.com/questions/22402791/visual-studio-2013-stdmutex-and-the-dreaded-windows-loader-lock

I wonder, could you try it with Visual Studio 2015 or 2017?

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Václav Haisman - 2017-07-11

I have pushed bd27044f35 to 1.2.x branch to work around this for 1.2.x. It should actually be OK on 2.0.x branch as only Visual Studio 2015 and later is supported there.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Ralf Sieger - 2017-07-12

Hi Václav,
I've tried your patch with log4cplus-1.2.1-rc2 and it seems this did the trick!
Thank you very much for the quick fix!

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Václav Haisman - 2017-07-12

labels: java, v1.2.0, c++11, threads, mutex --> java, v1.2.0, c++11, threads, mutex, fixed

status: open --> closed
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Crash on app exit

Logging Framework for C++

Group

Searches

Help

#343 Crash on app exit

Discussion