RFR: 8267796: vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/TestDescription.java fails with NoClassDefFoundError

[Alex Menkov]

On Thu, 24 Feb 2022 12:44:18 GMT, Alex Menkov <amenkov at openjdk.org> wrote:

> The fix adds workaround in hs201t001a class like we have in hs201t001 test to avoid class loading while the test do single stepping/pop frame.
> Also fixed a number of issues in the test.

test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/TestDescription.java line 74:

> 72:  *
> 73:  * @run main/othervm/native
> 74:  *      -agentlib:hs201t002=pathToNewByteCode=./bin,-waittime=5,-verbose

Adds verbose output for the test and agent

test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/hs201t002.cpp line 48:

> 46: 
> 47: static jint testStep;
> 48: static int redefineNumber;

not used

test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/hs201t002.cpp line 127:

> 125:         nsk_jvmti_setFailStatus();
> 126:     }
> 127:     if (!NSK_JVMTI_VERIFY(jvmti_env->Deallocate(newClassBytes))) {

proper resource deallocation

test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/hs201t002.cpp line 334:

> 332:         jclass klass;
> 333: 
> 334:         NSK_DISPLAY2("\n\n>>>> Exception %s in thread - %s\n",

getClassName() and getThreadName() used the same global buffer to return results. So thread name overrides class name and in log we see:
>>>> Exception hs201t002Thread in thread - hs201t002Thread
instead of 
>>>> Exception Lnsk/jvmti/scenarios/hotswap/HS201/hs201t002a; in thread - hs201t002Thread

test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/hs201t002.cpp line 382:

> 380: /* ============================================================================= */
> 381: 
> 382: int readNewBytecode(jvmtiEnv* jvmti, jint &newClassSize, unsigned char* &newClassBytes) {

The function returns loaded data instead of using global variable for correct resource management

test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/hs201t002.cpp line 429:

> 427: /* ============================================================================= */
> 428: 
> 429: const char* getThreadName(jvmtiEnv* jvmti_env, JNIEnv* jni_env, jthread thread) {

The function now returns JVMTI allocated buffer instead of global buffer

test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/hs201t002.cpp line 468:

> 466: /* ============================================================================= */
> 467: 
> 468: const char* getClassName(jvmtiEnv *jvmti_env, JNIEnv* jni_env, jobject object) {

The function now returns JVMTI allocated buffer instead of global buffer

test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/hs201t002.cpp line 668:

> 666:     NSK_DISPLAY0("Enable events\n");
> 667: 
> 668:     enableEvent(jvmti, JVMTI_EVENT_CLASS_LOAD, testedThread);

This code executed on agent initialization, the test has no chance to set testThread.
Moved to hs201t002.setThread()

test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/newclass/hs201t002a.java line 29:

> 27: 
> 28:     public hs201t002a () {
> 29:         System.out.println("Current step: " + hs201t002.currentStep); // Avoid calling classloader to find hs201t002 in doInit()

This is the fix for the failure, the rest are test improvements

-------------

PR: https://git.openjdk.java.net/jdk/pull/7607