https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95607
Bug ID: 95607
Summary: Inconsistent treating of default argument
instantiation as immediate context
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: akrzemi1 at gmail dot com
Target Milestone: ---
Consider the following three declarations in exactly this order:
```
template<typename T> int f(T, T = "") ;
int f(...) ;
template<typename T>
concept has_f = requires (T v) { f(v); };
```
If I evaluate expression `has_f<int>` it returns `false`. This is quite
surprising because other compilers, like Clang give me a hard error because in
order to test the satisfaction of the concept it has to perform *default
argument instantiation*. But we could call it "implementation divergence".
Anyway, GCC apparently considers the default argument instantiation to be *in*
immediate context.
But when I make the call `f(1)` I get a hard error. The first overload cannot
work because the default argument cannot be instantiated. The second one would
wort, but it is not even considered. So it looks like in this case GCC treats
default argument instantiation as *not being in* immediate context.
This is two different behaviors in two different contexts. It is confusing, and
difficult to build a mental model that would help anticipate when we get a hard
error and when we get a SFINAE-detectable one.