> No, comparing Bar b1 == Bar b2 will not succeed in your case. Yes, this is different from dmd. Should it be considered a bug/regression? I don't mind reverting this for unions, but it might not always be correct anyway... You're right. I didn't realize we can't actually compare unions in any other way. For the record: As we don't know the 'active' union field we can't do field-by-field comparisons of unions. So the only meaningful way to compare unions is by memcmp. And then we'd have to fill the 'holes' in structs again... I don't know if this corner-case is important enough to really worry about this.