Register String(byte[]) constructor for runtime JNI access

[jaikiran]

More than one user in the Quarkus project has reported that they run into the following exception when building a native image for a Quarkus application:

java.lang.NoSuchMethodError: java.lang.String.<init>([B)V 
 at com.oracle.svm.jni.functions.JNIFunctions$Support.getMethodID(JNIFunctions.java:1113) 
 at com.oracle.svm.jni.functions.JNIFunctions$Support.getMethodID(JNIFunctions.java:1098) 
 at com.oracle.svm.jni.functions.JNIFunctions.GetMethodID(JNIFunctions.java:411) 
 at java.io.UnixFileSystem.canonicalize0(UnixFileSystem.java) 
 at java.io.UnixFileSystem.canonicalize(UnixFileSystem.java:178) 
 at java.io.File.getCanonicalPath(File.java:626) 
 at java.io.File.getCanonicalFile(File.java:651) 
 at sun.security.provider.FileInputStreamPool.getInputStream(FileInputStreamPool.java:84) 
 at sun.security.provider.NativePRNG$RandomIO.<init>(NativePRNG.java:388) 
 at sun.security.provider.NativePRNG$1.run(NativePRNG.java:190) 
 at sun.security.provider.NativePRNG$1.run(NativePRNG.java:130) 
 at java.security.AccessController.doPrivileged(AccessController.java:82) 
 at sun.security.provider.NativePRNG.initIO(NativePRNG.java:129) 
 at sun.security.provider.NativePRNG.<clinit>(NativePRNG.java:93) 
 at com.oracle.svm.core.classinitialization.ClassInitializationInfo.invokeClassInitializer(ClassInitializationInfo.java:375) 
 at com.oracle.svm.core.classinitialization.ClassInitializationInfo.initialize(ClassInitializationInfo.java:295) 
 at java.lang.reflect.Constructor.newInstance(Constructor.java:490) 
 at java.security.Provider.newInstanceUtil(Provider.java:176) 
 at java.security.Provider$Service.newInstance(Provider.java:1894) 
 at java.security.SecureRandom.getDefaultPRNG(SecureRandom.java:290) 
 at java.security.SecureRandom.<init>(SecureRandom.java:219) 
 at java.util.UUID$Holder.<clinit>(UUID.java:101) 
 at com.oracle.svm.core.classinitialization.ClassInitializationInfo.invokeClassInitializer(ClassInitializationInfo.java:375) 
 at com.oracle.svm.core.classinitialization.ClassInitializationInfo.initialize(ClassInitializationInfo.java:295) 
 at java.util.UUID.randomUUID(UUID.java:147) 
 

The previous report of this was in [1] but reproducing that wasn't working out. The more recent report is here [2] and has more clear instructions on how to reproduce this. Looking at the code in these stacktraces and then looking at the JRE code, it comes down to a code path in jni_util.c which can lead to a call to the java.lang.String constructor which just takes the byte[] as a param (as uses the platform encoding internally). This is the place in jni_util.c where that call happens https://github.com/openjdk/jdk11u-dev/blob/master/src/java.base/share/native/libjava/jni_util.c#L732 :

/*If the encoding specified in sun.jnu.encoding is not endorsed
  by "Charset.isSupported" we have to fall back to use String(byte[])
  explicitly here without specifying the encoding name, in which the
  StringCoding class will pickup the iso-8859-1 as the fallback
  converter for us.
 */
jmethodID mid = (*env)->GetMethodID(env, strClazz,
                                    "<init>", "([B)V");
if (mid != NULL) {
    result = (*env)->NewObject(env, strClazz, mid, bytes);
}

The commit in this PR registers this constructor for JNI runtime access to prevent this exception.

[1] quarkusio/quarkus#10682 (comment)
[2] quarkusio/quarkus#19082

[jaikiran]

/cc @zakkak

[zakkak]

LGTM! Thanks @jaikiran

[jaikiran]

Hello @cstancu, would you be able to help review and merge this one, if this looks fine?

[pokusak]

Hi, @cstancu @zakkak would it be possible to merge this one?

[jaikiran]

Hello @dougxc, would you be able to help to have someone review this PR please?

[dougxc]

@peter-hofer can you please review and merge this if it looks good.

[pokusak]

@dougxc , @peter-hofer what needs to be done to approve this single line change?

[jaikiran]

Thank you @dougxc and @peter-hofer for helping in the review and merge.