I suspect this is from similar root cause of PR101402, PR102033, PR102034 etc. when incorrectly dropping top-level cv-qualifier during calculating template function declarator of which parameter is an array of template function pointer.
Clearly from "#2" of static assert, GCC already gets correct specialization "g<1,int>" signature, still the last statement of specialization declaration causes error with very strange error message. 

clang(https://www.godbolt.org/z/5eM5b4GjG) accepts. MSVC++(https://www.godbolt.org/z/3x9dvsfTh) stumbles like GCC, but a better clear clue.


#include<type_traits>
template<unsigned int N, class T>
void f(const T[N]){}

template<unsigned int N, class T>
using fPtr=decltype(f<N,T>)*;

template<unsigned int N, class T>
fPtr<N,T> af[N]={&f<N,T>}; // even without initialization,error still is

template<unsigned int N, class T>
void g(const decltype(af<N,T>)&){}

static_assert(std::is_same<decltype(af<1,int>),
 fPtr<1,int>[1] >::value, "af is correct"); // #1

static_assert(std::is_same<decltype(g<1,int>),
 void(fPtr<1,int>const(&)[1])>::value, "fun"); // #2

template<>
void g<1,int>(fPtr<1,int>const(&)[1]){}



GCC gives very misleading error message:

error: specialization of 'void g(decltype (af<N, T>)&) [with unsigned int N = 1; T = int; decltype (af<N, T>) = void (* [1])(const int*)]' after instantiation
   21 | void g<1,int>(fPtr<1,int>const(&)[1]){}
      |                                     ^