RFR(XS) 8020791: [TESTBUG] runtime/jsig/Test8017498.sh failed to compile

[Yumin Qi]

On 7/19/2013 11:48 AM, Calvin Cheung wrote:
> Thanks Dan and David for the reveiw.
>
> Dan,
>
> I've made all the changes as you've suggested except for the 
> LD_PRELOAD="$MY_LD_PRELOAD" before the java command. The shell tried 
> to execute the LD_PRELOAD="$MY_LD_PRELOAD" as a command and resulting 
> in "No such file or directory" error.
> So I'm keeping the "export" and "unset" of LD_PRELOAD for now.
>
You can put LD_PRELOAD="$MY_LD_PRELOAD" after command, that is
shell cmd LD_PRELOAD="$MY_LD_PRELOAD"

Thanks
Yumin
> updated webrev is at the same location:
>     http://cr.openjdk.java.net/~ccheung/8020791/webrev/
>
> thanks,
> Calvin
>
> On 7/19/2013 7:22 AM, Daniel D. Daugherty wrote:
>> On 7/18/13 6:12 PM, Calvin Cheung wrote:
>>> Please review this small fix for a testcase.
>>>
>>> webrev:
>>>     http://cr.openjdk.java.net/~ccheung/8020791/webrev/
>>
>> test/runtime/jsig/Test8017498.sh
>>     line 54: export LD_PRELOAD=${LD_PRELOAD}
>>         This export will affect every program run after this point.
>>
>>         Please change the variable name to MY_LD_PRELOAD" and don't
>>         export it.
>>
>>     line 57: NULL=NUL
>>     line 58: PS=";"
>>     line 59: FS="\\"
>>         These are not needed for your default case and they are
>>         Windows specific. Your default case is non-Linux so these
>>         are wrong in any case.
>>
>>     line 70: gcc -DLINUX -fPIC -shared -o 
>> ${TESTSRC}${FS}libTestJNI.so ...
>>         Please check the exit status of 'gcc' and do _something_
>>         if the compile fails. You could even pass the test if there
>>         is no compiler.
>>
>>     line 72: # run the java test in the background
>>         Why do you have to run the test in the background?
>>
>>     line 73: echo ${TESTJAVA}${FS}bin${FS}java 
>> -Djava.library.path=${TESTSRC}${FS} -server TestJNI 100 > test.out 
>> 2>&1 &
>>     line 74: ${TESTJAVA}${FS}bin${FS}java 
>> -Djava.library.path=${TESTSRC}${FS} -server TestJNI 100 > test.out 
>> 2>&1 &
>>     line 75:
>>     line 76: # obtain the process id
>>     line 77: C_PID=$!
>>
>>         line 73 will run the echo in background and you're running 
>> java in
>>         the background. That might lead to a race for the C_PID 
>> setting on
>>         line 77.
>>
>>         Lines 73 and 74 would be better as:
>>
>>         cmd="LD_PRELOAD="$MY_LD_PRELOAD" \
>>             ${TESTJAVA}${FS}bin${FS}java \
>>             -Djava.library.path=${TESTSRC}${FS} -server TestJNI 100"
>>         echo "$cmd > test.out 2>&1 &"
>>         $cmd > test.out 2>&1 &
>>
>>     line 80: sleep 1
>>         One second? This is just asking for intermittent failures of
>>         this test. If you decide that you have to run this test in
>>         background then something like this will be better:
>>
>>         count=0
>>         # give the test 60 seconds to pass
>>         while [ "$count" -lt 12 ]; do
>>             sleep 5
>>             grep "old handler" test.out > ${NULL}
>>             if [ $? = 0 ]; then
>>                 echo "Test Passed"
>>                 exit 0
>>             fi
>>             count=`expr $count + 1`
>>         done
>>         kill -9 ${C_PID}
>>         echo "Test Failed"
>>         exit 1
>>
>>     line 82: # reset LD_PRELOAD
>>     line 83: unset LD_PRELOAD
>>         You don't need this if you switch to the MY_LD_PRELOAD logic
>>         that I gave you above.
>>
>> test/runtime/jsig/TestJNI.c
>>     line 45: pthread_attr_t attr;
>>     line 46: stack_t stack;
>>         These variables are not used.
>>
>> Dan
>>
>>
>>
>>>
>>> Bug:
>>> https://jbs.oracle.com/bugs/browse/JDK-8020791
>>> http://bugs.sun.com/view_bug.do?bug_id=8020791
>>>
>>> When the testcase was developed, it was using an incomplete jni_md.h
>>> without the following:
>>> #if defined(SOLARIS) || defined(LINUX) || defined(_ALLBSD_SOURCE)
>>> The default typedef long jlong will be used during compilation of the
>>> native code of the testcase on linux_x64.
>>>
>>> With the above #if defined in the jni_md.h, if "LINUX" isn't defined,
>>> the following typedef will be used and causing the compile error.
>>>     typedef __int64 jlong;
>>>
>>> The fix is to define LINUX (-DLINUX) in the gcc command line.
>>> The change in TestJNI.c is just a minor cleanup.
>>>
>>> Test:
>>>    Ran jtreg on the testcase.
>>>
>>> thanks,
>>> Calvin
>>
>