As far as I understand the C and C++ standard regarding unions, it is technically undefined behaviour if I try to access a particular field of a union that is different to the field that was actually written to. Recently I have been reviewing some old code written by someone else that is similar to the following:
union myunion {
float myfloat;
unsigned int myuint;
};
myunion a;
a.myfloat = 1.01;
myunion b = a;
in a nutshell, I am trying to find out if using the assignment operator on a union in this way is actually well defined behaviour. When I write code like this myself, I endeavour to store the type of the field actively used by a union instance and make sure that I only read or write this value. I am guessing that using the assignment operator above perhaps just results in a bitwise copy (since in this example the fields are of equal size) as stepping through the code in the debugger shows that this is what is happening. This code has been exhibiting subtle bugs so I am keen to find out if there is anything inherently risky in the sense of UB by directly assigning a union to another in this way.
In pre-C++11, union could only contain PODs (or something similar)—the restrictions were so designed that a bitwise copy would work. In C++11, some of the restrictions have been lifted, but if you violate the old restrictions, copy and assignment are deleted (unless you define the operators yourself).