Please review correction of regression test Test6929067

[David Holmes]

Pavel Tisnovsky said the following on 05/06/11 19:57:
> David Holmes wrote:
>> Pavel,
>>
>> I'm a bit confused about this test with regards to how it is supposed to
>> work and what combinations of 32-bit and 64-bit are expected. It seems
>> to me that if you run this on a 64-bit platform then it expects to find
>> that TESTJAVA is a 64-bit VM, which may not be the case.
>>
>> I also don't understand this part:
>>
>> ! gcc ${COMP_FLAG} -o invoke \
>> ! -L${TESTJAVA}/jre/lib/${ARCH}/client \
>> ! -L${TESTJAVA}/jre/lib/${ARCH}/server \
>> ! -ljvm -lpthread -I${TESTJAVA}/include -I${TESTJAVA}/include/linux
>> invoke.c
>>
>> does this assume/expect that client will be found on 32-bit and server
>> only on 64-bit because we don't presently ship 64-bit client? In terms
>> of exported symbols I believe the two should be the same. Will gcc
>> complain if it encounters a 64-bit lib if compiling with -m32 and
>> vice-versa?
> 
> Hi David,
> 
> AFAIK gcc does not care if client and/or server libraries is found. When
> .../client/ directory does not exists it simply uses libraries from
> .../server/ directory. I'm not sure in which case gcc could find 64-lib
> using -m32 - on Solaris? (To be honest I have not tested this test on
> Solaris)

Solaris behaviour is not relevant as the test trivially declares success 
on Solaris and Windows.

But I'm very unclear on the behaviour on Linux. As I said it seems to 
only expect to find a 64-bit JVM on a 64-bit capable platform.

David

> Pavel
> 
>> David Holmes
>>
>> Pavel Tisnovsky said the following on 05/06/11 02:05:
>>> Pavel Tisnovsky wrote:
>>>> David Holmes wrote:
>>>>> Dr Andrew John Hughes said the following on 11/17/10 08:55:
>>>>>> On 16 November 2010 11:28, Pavel Tisnovsky <ptisnovs at redhat.com>
>>>>>> wrote:
>>>>>>> David Holmes wrote:
>>>>>>>> This is a fix for 64-bit systems but the invoke program is still
>>>>>>>> being
>>>>>>>> compiled as 32-bit. I'm a little surprised it can then load the
>>>>>>>> 64-bit VM.
>>>>>>> it seems that "invoke" is compiled as 64-bit executable (on RHEL 5
>>>>>>> x86_64 at least):
>>>>>>>
>>>>>>> invoke: ELF 64-bit LSB executable, AMD x86-64, version 1 (SYSV), for
>>>>>>> GNU/Linux 2.6.9, dynamically linked (uses shared libs), for GNU/Linux
>>>>>>> 2.6.9, not stripped
>>>>>> I think David may be referring to the need for -m32 or -m64, where gcc
>>>>>> is configured.  On GNU/Linux platforms, gcc tends to produce a 64-bit
>>>>>> binary on x86_64 by default but on Solaris, it produces a 32-bit
>>>>>> binary.
>>>>> Exactly so - sorry didn't see your reply before my last reply.
>>>>>
>>>>> So in its new form the test will work correctly if run on a 64-bit
>>>>> VM on
>>>>> 64-bit Linux, or a 32-bit VM on 32-bit Linux, but not on a 32-bit VM on
>>>>> 64-bit Linux. (I[m not sure our internal testing ever tries the
>>>>> latter).
>>>>>
>>>>> It also relies on the fact that the Linux JRE/JDK doesn't contain both
>>>>> the 64-bit and 32-bit VMs.
>>>> Hi David,
>>>>
>>>> you are right, we can't assume that find select the right libjvm.so. I'm
>>>> going to fix this test by other way.
>>>>
>>>> Thanks
>>>> Pavel
>>>>
>>>>> David
>>> Hi David,
>>>
>>> here is new version of fix of regression test Test6929067:
>>> http://cr.openjdk.java.net/~ptisnovs/jtreg-runtime-test-6929067-fix/
>>>
>>> Can you please review it?
>>>
>>> Thank you in advance,
>>> Pavel Tisnovsky
>