|go ahead... be a heretic|
Perhaps you should read to the end?
What's an obvious way to implement that?
"An obvious way" is very much not "the only way".
But I don't see why changing the first AV with the "sigiled" keys to a HV with indexes to the value AVs shouldt be a problem in C.
Because, as I noted, we don't have sigils for every type of slot in a typeglob. The "in C" part was about there being different value types. To have a value type that could be any of the 6 or so typeglob slot types would be at least complex. It would be complex enough that I would have qualms about that choice.
It turns out that a Perl 5 SV can already be any of those types. But SVs didn't even exist in Perl 4 so they couldn't be how typeglobs were implemented (in Perl 1.0).
The polymorphic SVs are indeed quite complex. The "normal" SV types have much better motivation for how they were done, to my mind, because an SV can transition between those different flavors. Stuffing formats and "IO"s into the polymorphic SV family of structs makes much less sense to me. I couldn't find any part of the Perl source code where this fact was actually useful (I didn't spend that long trying, though) and yet it adds significant complexity, IMHO.
Searching for "PVFM" in the Perl source code mostly finds places where such are not supported.
I'm glad Larry didn't start off trying to implement fully polymorphic SVs in Perl 1.