`NonLinearScale` triviality

[rbrugo]

Hi, I've a design question for you: I can't understand why NonLinearScale must be a trivial type. I've read the code, but I couldn't find any point where this triviality is exploited. I mean, yes, this way a unit_t must a trivial type, but with this restriction I can't make an unit based on, for example,

• boost::multiprecision numbers such as cpp_int and float128
• std::complex<T>
• std::valarray<T>
• any user-defined type which initialize to zero its storage

Even without this restriction, a unit_t based on a trivial type remains trivial, so removing this constraint doesn't seems to have real downsides; it seems to me that default_constructible is sufficient for a type to be used with a NonLinearScale. Am I missing something? It would be possible to remove the constraint safely?

Thanks

[nholthaus]

I don't think this is an issue in 3.0.