Restore colour to rustdoc

[ftxqxd]

This not only gives back the colour to structs, traits, and functions, but also restores the long-lost colour to enums. I’ve run linkchecker and there don’t seem to be any more invalid links than usual.
I’m not entirely sure if the way I’ve implemented this is very idiomatic or reliable, but it seems to work.

Closes #17131.
Closes #16712.
Closes #16711.

[alexcrichton]

This may end up causing confusion as it doesn't take into account did.krate, which can be nonzero in the case of inlining documentation from other crates. This could sporadically cause typedefs to become enums or vice versa perhaps.

[ftxqxd]

I think you’re right: rustrt renders Option as a type instead of an enum (e.g., rustrt::args::clone).

[Gankra]

@P1start Can you host a rust documentation instance running this change? We've had some unfortunate layout regressions recently, and I'd like to avoid that.

[ftxqxd]

@gankro I’ve hackily set something up with the documentation generated by this.

[Gankra]

@P1start Awesome, thanks! Documentation looks exactly as broken as always 👍

One thing I noticed is that generics seem weird as uncoloured now. Would it be possible to colour them?

[ftxqxd]

I don’t think that colouring generics would be a good idea, as every other coloured thing is a link to another page. If anything, we should colour typedefs, primitives &c. something other than black to make them stand out more than things which aren’t documented (generics and private types).

[Gankra]

That's a fair point. Perhaps generics could simply be coloured grey to make them visually distinct from keywords and sigils, but otherwise "dead"?

[ftxqxd]

I’ve tried another path that seems to work much better. (The old one was still generating bad cross-crate links sometimes.) I added a bool to rustc::middle::def::Def’s DefTy to indicate whether or not it’s an enum.

[eddyb]

Adding to Def is a definite no.
Have you tried using free functions in middle::ty to automatically handle the cross-crate case?
That is what many rustc passes rely on.

[ftxqxd]

@eddyb Sorry, I don’t know which functions you’re talking about.

[ftxqxd]

@alexcrichton I had to update the last commit because Travis was showing some legitimate errors.