I've read that using reinterpret_cast<> should rarely (if ever) be used. So, what is the proper thing to do in a case like this...

Code:
uint64_t u = ; // Some unsigned int64 calculation
int64_t* i = &u;
This, of course, generates an error.
error: invalid conversion from 'uint64_t*' to type 'int64_t*'

Code:
int64_t* i = static_cast<int64_t*> (&u);
And this generates the following error.
error: invalid static_cast from type 'uint64_t*' to type 'int64_t*'

Code:
int64_t* i = reinterpret_cast<int64_t*> (&u);
This works but is considered bad practice since reinterpret_cast is really analogous to the old 'c' casting style (i.e. (int64_t*) &u In other words, there are no CPU instructions generated, it is simply a compiler directive indicating the bits are to be interpreted as the new type.

It appears the solution would be this.
Code:
int64_t* i = static_cast<int64_t*> (static_cast<void*> (&u));
And this is what I have done for quite some time when a situation like this arises. However, what is this really doing? Isn't this analogous to the reinterpret_cast (aside from the fact that compiler instructions are really being generated)? But anything can be cast to a void*, and then from a void* to any other type*. So I assume the underlying bits are not translated to and from the intermediate void* type. Is this true? If so, then it appears you are accomplishing nothing as the internal bits for the data type are not translated.

Is this really the appropriate thing to do? If anyone can help enlighten me it would be greatly appreciated. Thanks.