Time to parse a scratch file grows with size of codebase

[ChrisPenner]

Using Stew's giant codebase I notice that, even in a completely empty branch, getting a slurp printout after saving a scratch file with a single definition takes about 2s; on a fresh codebase it's instant.

I suspect we're still loading the global PPE or something, but with self-contained namespaces we should probably just not do that.

[ChrisPenner]

I confirmed that we do indeed load the root names, and ALSO find historical names for all labeledreferences, all of which I believe is unnecessary under the new "self-contained namespaces" assumptions.

I'll look at fixing this probably after #3391 or something like it gets merged since it'll be much more noticeable at that point.

[ceedubs]

all of which I believe is unnecessary under the new "self-contained namespaces" assumptions.

Today we talked about a violation of the "self-contained namespaces" assumption that is going to be common. If you update a type, then you will have a bunch of references to the previous (and now unnamed) type until you finish your entire todo list.

[aryairani]

probably fixed with #5041

[ChrisPenner]

I think this was probably already fixed when we removed the global root ppe fallback, but project-roots can't hurt either :)