DirectArgumentsMarshalNullableTest failure for s390x

[nealef]

1. Version of JNA and related jars
   • From git master
2. Version and vendor of the java virtual machine
   • openJDK 11 (Eclipse openJ9)
3. Operating system
   • Linux
4. System architecture (CPU type, bitness of the JVM):
   • s390x (64-bit, big endian)
5. Complete description of the problem
   • DirectArgumentsMarshalNullableTest test fails:

    [junit] Testcase: testIntegerType(com.sun.jna.DirectArgumentsMarshalNullableTest):	FAILED
    [junit] Basic non-null call expected:<42> but was:<0>
    [junit] junit.framework.AssertionFailedError: Basic non-null call expected:<42> but was:<0>
    [junit] 	at com.sun.jna.ArgumentsMarshalNullableTest.testIntegerType(ArgumentsMarshalNullableTest.java:214)
    [junit] 	at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
    [junit] 	at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
    [junit] 	at java.base/jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)

6. Steps to reproduce

• Build using ant 1.10.6

I inserted some debug code into DirectArgumentsMarshalNullableTest and into native/testlib.c to see what was being processed:

1. Value of Int32.fromNative/toNative -

    [junit] from: com.sun.jna.ArgumentsMarshalNullableTest$Int32$1@7567860b
    [junit] to:   com.sun.jna.ArgumentsMarshalNullableTest$Int32$2@ae757b28
    [junit] from: com.sun.jna.ArgumentsMarshalNullableTest$Int32$1@7567860b
    [junit] from: com.sun.jna.ArgumentsMarshalNullableTest$Int32$1@7567860b
    [junit] to:   com.sun.jna.ArgumentsMarshalNullableTest$Int32$2@ae757b28
    [junit] to:   com.sun.jna.ArgumentsMarshalNullableTest$Int32$2@ae757b28

2. Value of arg in returnInt32Argument

    [junit] int32arg: 0000002a
    [junit] int32arg: 00000000
    [junit] int32arg: 00000000
    [junit] int32arg: 0000002a
    [junit] int32arg: 00000000

Note: All other tests pass.

[matthiasblaesing]

Could you do a bisect to see when this problem was introduced? I build the library in a qemu-chroot but the speed is only enough for that. If you can run it on real hardware, this should speed up bisecting considerably. Thank you.

[nealef]

Will do. This was the 1st time I’ve tried this level of JNA the earlier versions I worked on (e.g. 4.4) didn’t have that test. If you need an account for doing s390x stuff I can provide one on a CentOS 7.6 clone. I also have a Jenkins slave available if you want to do CI (and use Jenkins of course). Neale

[matthiasblaesing]

Am Mittwoch, den 05.06.2019, 08:56 -0700 schrieb Neale Ferguson:

Will do. This was the 1st time I’ve tried this level of JNA the earlier versions I worked on (e.g. 4.4) didn’t have that test. If you need an account for doing s390x stuff I can provide one on a CentOS 7.6 clone. I also have a Jenkins slave available if you want to do CI (and use Jenkins of course).

If you want to tackle it yourself - great I'm willing to help. If you prefer me to have a look, an account would be great. Thank you for the offer! Matthias

[nealef]

Send me your public key. .

[matthiasblaesing]

The fix for the issue was merged to master. The problem was not in the unittest, but a real problem in the direct dispatch code path.