JNA 4.5.1 chooses wrong libjnidispatch.so when using Oracle Java 8 on armhf

[joelbygger]

I am very green on Java and JNA, please be patient if nomenclature is wrong.

• Fails with JNA 4.5.1
• Works with JNA 4.5.0
• Ubuntu 14 LTS
• 32 bit i.MX6 ARMv7l (armhf, hard float)
• Can be reprodued with "Hello world" from this repo, https://github.com/java-native-access/jna/blob/4.5.1. /www/GettingStarted.md
• Compile with maven, downloads JNA via "net.java.dev.jna". Create a jar-with-dependencies, libjnidispatch.so exists for 30 something arhitectures.
• Same result independent on what I compile with (Oracle or OpenJDK). OpenJDK can always run the jar, Oracle always fails.
• No issues with OpenJDK Java 8
• Fails with Oracle Java 8:

$ java -XshowSettings:properties -version
...
sun.arch.abi = gnueabihf
...
java version "1.8.0_172"
Java(TM) SE Runtime Environment (build 1.8.0_172-b11)
Java HotSpot(TM) Client VM (build 25.172-b11, mixed mode)

also tried with:

java version "1.8.0_73"
Java(TM) SE Runtime Environment (build 1.8.0_73-b02)
Java HotSpot(TM) Client VM (build 25.73-b02, mixed mode)

• When using flag "-Djna.boot.library.path=" and pointing to specific JNA it works.

Output on success:

$ java -Djna.debug_load=true -jar hello-jar-with-dependencies.jar 
Looking in classpath from sun.misc.Launcher$AppClassLoader@1f96302 for /com/sun/jna/linux-arm/libjnidispatch.so
Found library resource at jar:file:/some/path/hello-jar-with-dependencies.jar!/com/sun/jna/linux-arm/libjnidispatch.so

Output on failure:

Looking in classpath from sun.misc.Launcher$AppClassLoader@1909752 for /com/sun/jna/linux-armel/libjnidispatch.so
Found library resource at jar:file:/some/path/hello-jar-with-dependencies.jar!/com/sun/jna/linux-armel/libjnidispatch.so
Exception in thread "main" java.lang.UnsatisfiedLinkError: /tmp/jna--873636902/jna2423240844359937896.tmp: /tmp/jna--873636902/jna2423240844359937896.tmp: cannot open shared object file: No such file or directory

My conclusion:
Wrong architecture (armel) is choosen when using JNA 4.5.1.

[matthiasblaesing]

I assume the Oracle binaries suffer from the bug, that the binary is not marked to be hardfloat. Please see #920 for an explanation. You found the sun.arch.abi property. It might be worth to create patch, that checks this property to work around the oracle limitation - I'd appretiate such a patch.

[matthiasblaesing]

Ok - I checked raspbian and the oracle 8 JDK and both show the problem. The flags header does not contain the hard-/softfloat markers. So it looks as if the toolchains of both builds don't support adding the flag.

I added an arm EABI build attributes parser to the detection engine and this works also on older binaries.

I tested:

• raspbian OpenJDK 7
• Oracle JDK 8 for ARMHF on rasbian
• openjdk 7 on debian stable arm (hardfloat) and armel (softfloat) on qemu emulation

@joelbygger please see if this fixes the issue for you
@kunkun26 as you authored the changes in #863 I'd like you to review/test this change. It should be safe, but I don't routinely test on arm.

You can either build from the PR branches (#960, #961) or use the attached zip, which contains binaries for the 4.5.X and master branch (upcoming 5.0.0).

jna-test.zip

[matthiasblaesing]

I merged the changes too master and created new snapshots:

https://oss.sonatype.org/content/repositories/snapshots/net/java/dev/jna/jna/5.0.0-SNAPSHOT/jna-5.0.0-20180527.175730-1.jar
https://oss.sonatype.org/content/repositories/snapshots/net/java/dev/jna/jna/4.5.2-SNAPSHOT/jna-4.5.2-20180528.165716-1.jar