https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96111
--- Comment #1 from Klaus Rudolph <lts-rudolph at gmx dot de> --- The error massages are valid as as got by an answr on SO ( https://stackoverflow.com/questions/62791460/checking-type-of-attribute-with-concepts ) [expr.prim.req.compound]/1.3 If the return-type-requirement is present, then: Substitution of template arguments (if any) into the return-type-requirement is performed. The immediately-declared constraint ([temp.param]) of the type-constraint for decltype((E)) shall be satisfied. E is our expression, namely n.value. Now, decltype(n.value) is char or int, that's because decltype has a special rule for class member access and id expressions. But decltype((n.value)) is char& or int&. The value category is encoded in the type of decltype when dealing with a general expression (such as a parenthesized class member access). BUT: That we see an ICE on current trunk is a bug!
