With #60966, we can avoid lang items and manual path matching for lints, and instead mark them in the code. But I think there are some places that could still be ported to the new system:

• The InvalidValue lint uses match_def_path for transmute -- somehow, using a diagnostic item did not work (is_diagnostic_item returns false).
• Aren't Rc and Arc lang items just for diagnostics? Here's where they get used:

  rust/src/librustc_typeck/check/expr.rs

  Lines 879 to 880 in f49f388

  	try_alt_rcvr(&mut err, rcvr_t, lang_items::Arc);
  	try_alt_rcvr(&mut err, rcvr_t, lang_items::Rc);

  
  And, indirectly, at
  

  rust/src/librustc_mir/borrow_check/error_reporting.rs

  Lines 720 to 723 in f49f388

  	if ty.is_rc() {
  	format!("an `Rc`")
  	} else if ty.is_arc() {
  	format!("an `Arc`")

  
  So, I am not sure if those should be lang items or diagnostic items now? No language semantics is attached to them, the uses above are just in error messages from what I can see.
• There's also no language semantics attached to MaybeUninit; it is a lang item because it gets used in layout construction for generators. Doesn't really fit "diagnostics item" either, though. (EDIT: decided this makes more sense to remain a lang item)

@oli-obk @Centril any thoughts?