Possible leak in compile function

[mretallack]

Hi,

I have been performing a valgrind check on the avian jvm and I think I may of found a leak in the compile.cpp - compile function.

It appears that stackMap is allocated and then passed to the Frame class, but is not free'ed. I have included a patch to show the changes that I have done. This appears to stop valgrind reporting the issue, however I do not feel that I understand the code enough to say that this is a problem and that the fix is correct.

Thanks Mark Retallack

Index: avian/src/compile.cpp
===================================================================
--- avian.orig/src/compile.cpp
+++ avian/src/compile.cpp
@@ -7030,9 +7030,9 @@ void compile(MyThread* t, Context* conte

             c->restoreState(state);

-            ir::Type* stackMap = (ir::Type*)malloc(
+            ir::Type* stackMap2 = (ir::Type*)malloc(
                 sizeof(ir::Type) * context->method->code()->maxStack());
-            Frame frame2(&frame, stackMap);
+            Frame frame2(&frame, stackMap2);

             unsigned end = duplicatedBaseIp + exceptionHandlerEnd(eh);
             if (exceptionHandlerIp(eh) >= static_cast<unsigned>(start)
@@ -7053,6 +7053,7 @@ void compile(MyThread* t, Context* conte
             context->eventLog.append(PopContextEvent);

             eventIndex = calculateFrameMaps(t, context, 0, eventIndex, 0);
+            free(stackMap2);
           }
         }
       }

[joshuawarner32]

@mretallack Good catch, thanks!

FWIW, looks like it was my fault (replacing THREAD_RUNTIME_ARRAY with malloc because the former doesn't support non-POD types, forgetting the free): 7abbace8fb47f9a98f5323c7aaad21346bcde195

Another option would be to modify THREAD_RUNTIME_ARRAY to allocate it's memory as a char* and reinterpret_cast where necessary. That'll lose some type safety, though.

Your patch looks good. Do you want to make a pull request? Otherwise, I can commit it on your behalf.

[joshuawarner32]

Also, I'm curious how you got avian to run correctly under valgrind? I'm not sure what changed, but it seems within the last year or so, avian stopped working under valgrind - it just crashes in initialization.

[mretallack]

Hi @joshuawarner32 , thanks for the info,

Its probable best if you apply the patch for now.

I am using the libjvm under our own application (using the JNI interface to start Avian) and I don't think I have had any problems running it under valgrind. Not sure if it makes any difference, but we are using i386 instead of x64.

Thanks again.