Non-virtual interface (Template Method pattern)

[Jon12345]

I was surprised to read "Speaking of virtual functions, should non-virtual interface be promoted? NO." in the "To-do" list.

The guidelines in http://www.gotw.ca/publications/mill18.htm are sound, yet not widely understood.
What's the reason for the "NO"?

[hsutter]

Good catch, thanks.

[cubbimew]

Although I originally submitted that TODO item expecting a "yes" as well, I am not sure it makes a good candidate for a guideline that may be checked by tools: mechanically NVI'ing every virtual function gives us things like std::time_get with seven dummy public forwarders. Is it any different from dummy getters and setters? If the customization points exactly match the public API this seems superfluous.

Something like std::streambuf, where customization points are entirely different from the public API makes more sense as a design guideline, but how could it be formalized?