Current implementation of unions which are composed solely of strings is quite wasteful, as it causes char[] duplication.

A typical example from ietf-inet-types is ip-prefix. We generate the following fields:

private final Ipv4Prefix _ipv4Prefix;
private final Ipv6Prefix _ipv6Prefix;
private final char[] _value;

We generate the default constructor for use with JMX:

@ConstructorProperties("value")
public IpPrefix(char[] _value)

{ java.lang.String defVal = new java.lang.String(_value); IpPrefix defInst = IpPrefixBuilder.getDefaultInstance(defVal); this._ipv4Prefix = defInst._ipv4Prefix; this._ipv6Prefix = defInst._ipv6Prefix; this._value = _value == null ? null : _value.clone(); }

This means that we have two copies of _value stored: one inside one of the prefixes and one retained via explicit clone.

As it turns out, we do a very similar thing in both normal constructors:

public IpPrefix(Ipv4Prefix _ipv4Prefix)

{ super(); this._ipv4Prefix = _ipv4Prefix; this._ipv6Prefix = null; this._value = _ipv4Prefix.getValue().toString().toCharArray(); }

public IpPrefix(Ipv6Prefix _ipv6Prefix)

{ super(); this._ipv6Prefix = _ipv6Prefix; this._ipv4Prefix = null; this._value = _ipv6Prefix.getValue().toString().toCharArray(); }

While this is generic and works for all types, for strings this ends up being inefficient.

Generated code should be modified so that if a union has only string types, _value field should not be generated and getValue() should be implemented as:

public char[] getValue() {
final String str;
if (_ipv4Prefix != null)

{ str = _ipv4Prefix.getValue(); }

else if (_ipv6Prefix != null)

{ str = _ipv6Prefix.getValue(); }

else

{ throw new IllegalStateException(...); }

return str.toCharArray();
}

Also note that this sort of optimization also applies when a union member is a boolean, as there are only two possible strings accessible via Boolean.toString().