Adding a new pointer doesn't trigger reevaluation of schema expressions that could reference it

[msullivan]

If we do something like

create type T;                                                            
create type P { create multi link l -> T; };                              
insert T;                                                                 
insert P { l := T }; insert P { l := T };                                 

create function cnt_l() -> int64 using (select count(T.<l));              
select cnt_l();                                                           

create type Q { create multi link l -> T; };                              
insert Q { l := T };                                                      
select cnt_l();

we get back 2, as expected, from the first call, and then 2 again from the second call.

This is because cnt_l doesn't get updated when a new link named l gets created. (Note that banning use of backlinks without a type intersection does not help us here, unless we are willing to go as far as banning things like T.l[is Object] (or any other type that doesn't directly have the link)).

This example uses functions but I think there are other potential snags---maybe we could fail to trigger a cycle error in an access policy?

--

As far as backportable ways to fix this, we might be stuck finding all existing links with the same name and doing a _propagate_expr_refs on the expr references to all of them. Ew! For future versions we can probably reduce the pain a bit and track specifically which what backreferences are used by expression, by name. (We can't do that in 2.0 because it would require adding more fields to the schema)

[msullivan]

There is another issue related to this, which is that if a new required pointer is created, or a pointer is made required, that could invalidate existing rewrites/triggers/defaults that use it