Add a warning to LateContext::get_def_path

[blyxyas]

Preventing anyone from doing the same error as rust-lang/rust-clippy#15043 fixed

[rustbot]

r? @jdonszelmann

rustbot has assigned @jdonszelmann.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[compiler-errors]

Suggested change

	/// Note that this is kinda expensive because it has to
	///
	/// Note that this is kinda expensive because it has to

[compiler-errors]

or use a diagnostic item if the intention was to match an item by its path.

[compiler-errors]

I'm not sure how this connects to rust-lang/clippy#15030. That looks like some test harness problem?

[compiler-errors]

I'm also somewhat skeptical that this function is that notable. Grabbing the symbols for every segment in a def path doesn't seem like that much work in the grand scheme of things. Of course, if this is being done in a tight loop it may be expensive, but we use similar a to compute symbol names and stuff even on the good path.

Can you explain more about what led you to this function being expensive?

[blyxyas]

Oops, I meant rust-lang/rust-clippy#15043, 15030 is the one I'm working on right now. I'll edit that.

In our case, get_def_path was the 3rd heaviest function, we used it in a semi-tight loop, but still, it was accounting for more time than late_lint_mod.

Can you explain more about what led you to this function being expensive?

The call graph is like this:

get_def_path -> AbsolutePathPrinter::default_print_def_path -> core::fmt::write -> <Ty as core::fmt::Display>::fmt -> <FmtPrinter as Printer>::print_type -> ... -> rustc_interface::passes::early_lint_checks -> Running the early lint checks, somehow?

I haven't really looked into why pretty-printing a type would execute early lint checks. Maybe it has something to do with the pretty-printer calling tcx.visible_parent_map.

---

As to the necessity of this little warning, I've looked very briefly into other uses of get_def_path and a lot seem to be optimizable away with other, cheaper and more specific functions. So at least warning the user to think about alternatives I think it's necessary.