New mangling rule for deduced default template arguments.

[mizvekov]

As part of https://isocpp.org/files/papers/P3310R0.html, we would like to introduce deduction of default template arguments.

When deducing a template template parameter from a template specialization, we propose to default any extra arguments the parameter side would not consume, thus effectively allowing a template template parameter to bind to a template declaration with less template parameters, even if the extra parameters aren't declared as defaulted in the template declaration.

Consider:

template<class T, class U> struct A {};
A<int, float> v;
template<template<class> class TT> void f(TT<int>);

// OK: TT picks 'float' as the default argument for the second parameter.
void g() { f(v); }

Since 'f' specialized from A<int, float> is different from 'f' specialized from A<int, double>, this will affect mangling and so we need new rules.

This should only affect template template arguments.

A class template specialization A<int, double> which was specialized as written, is still the same class when specialized from a template template parameter bound to A deduced from a different specialization.

I would like guidance with such a rule.

My initial idea would be that when mangling such a template template argument, we follow the template name with a number, which is a starting position from which the first argument is deduced, followed by the same mangling we currently use for a template argument list.

Thus:

template<class T, class U> struct A {};

template<template<class> class TT> void f(TT<int>);

void z() {
  f(A<int, double>());
}

This 'f' specialized with A<int, double> would be mangled as _Z1fITtTyE1A1IdEEvT_IiE.

[rjmccall]

Mangling an adjustment on the template argument seems like the right approach. That particular suggestion seems likely to cause problems because of how it separates the template-name from the template-arguments and just drops an arbitrary numeral into the mangling, but we can surely come up with something. I believe we already specify an adjustment for changing template signatures for template template arguments; it's possible that that would be sufficient here, or maybe we need to be more explicit about the deduction.

[mizvekov]

I believe we already specify an adjustment for changing template signatures for template template arguments; it's possible that that would be sufficient here, or maybe we need to be more explicit about the deduction.

Is this about https://github.com/itanium-cxx-abi/cxx-abi/pull/166, wrt the template-param-decl production?

[rjmccall]

Yes. Thanks for reminding me that I need to put some effort into landing that in the spec.

[mizvekov]

It seems that could work, but it needs to be extended to encode default arguments as well.

Though it seems this could already be required for the proposed resolution of https://open-std.org/jtc1/sc22/wg21/docs/cwg_active.html#1286