"Gottlob Frege"
Top posting because I'm going to remark about this thread in general, but this message is a fine example.
Andy seems to be speaking (mostly) about int_<>, integral_c, etc (I'll just say the incorrect 'int<>') and plus; David is talking about transform.
I think, but can't say for sure, that Andy wants plus to work 'as expected' ^^^^^^^^^^^^^^^^^^ and always return int<>, not just 'concept-identical' to int.
Andy, please take note: what you want is still unclear to people other than me. Since "as expected" is vague, and your statement seems a little too sweeping, I'll say it this way: I (Dave) think, but can't say for sure, that Andy wants plus to always return an int_ specialization when all the inputs are specializations of int_. I think, but can't say for sure, that he also wants a different specification for the behavior of plus, and all the other arithmetic operators.
Maybe it does, I'm not sure. And/or he wants transform ***when applied to int<> with plus, etc*** to return int<>.
David, says, that ***in general*** transform can't make the return type "match" (in Andy's stricter sense) the inputs.
Maybe it could for the special case of int<>? IMO, that would be quite hard, as the point of the templates is to be general.
I don't think we can do it. However, it's certainly possible to make
the *elements* of the resulting container be int_ specializations
whenever the operation has the form:
transform