Handling of byte arrays bigger than 65536 bytes appears to be broken. In the attached test case, where a byte array of 70000 bytes is being preloaded, the following is being thrown:

Caused by: java.lang.NullPointerException
at org.jboss.marshalling.river.RiverUnmarshaller.doReadObjectArray(RiverUnmarshaller.java:1369)
at org.jboss.marshalling.river.RiverUnmarshaller.doReadArray(RiverUnmarshaller.java:1412)
at org.jboss.marshalling.river.RiverUnmarshaller.doReadObject(RiverUnmarshaller.java:291)
at org.jboss.marshalling.river.RiverUnmarshaller.doReadObject(RiverUnmarshaller.java:158)
at org.jboss.marshalling.AbstractUnmarshaller.readObject(AbstractUnmarshaller.java:81)
at org.infinispan.container.entries.ImmortalCacheEntry$Externalizer.readObject(ImmortalCacheEntry.java:121)

The code shouldn't have gone through the doReadObjectArray() path since it's a byte array.

The problem comes from the differences writing byte arrays smaller and bigger than 65536:

} else if (len <= 65536)

{ write(unshared ? ID_ARRAY_MEDIUM_UNSHARED : ID_ARRAY_MEDIUM); writeShort(len); write(ID_PRIM_BYTE); write(bytes, 0, len); }

else

{ write(unshared ? ID_ARRAY_LARGE_UNSHARED : ID_ARRAY_LARGE); writeInt(len); write(ID_PRIM_BYTE); write(bytes, 0, len); }

For less than 65536, 2 bytes are used for lenght, whereas for longer, 4 bytes. The reading part appears to always read two bytes, so when reading long byte arrays, it takes the 3rd byte of the length which is not ID_PRIM_BYTE and so it fails.

Indeed, here's the reading part:

case ID_ARRAY_LARGE_UNSHARED: {
if (unshared != (leadByte == ID_ARRAY_LARGE_UNSHARED))

{ throw sharedMismatch(); }

final int len = readUnsignedShort();
return doReadArray(len, unshared);
}

It appears it should be reading an unsigned int.