[PERF] Instantiations are getting too far

[blyxyas]

As the documentation lints got approved, Ty::intern is now the public enemy number one. Today I analyzed all the occurrences of instantiate* functions and all Ty::interns.

Here is a list of all those calls that exceed 65 while running our test suite. Not all of these can be optimized away, but I'm sure that the greatest culprits can be optimized greatly.

There were lots of lints that use the instantiation functions <50 times in the whole test suite, so I'm pretty sure that (almost) any function that uses them > e.g. 5k times can be heavily optimized away.

For Ty::intern directly, only manual_abs_diff.rs::is_sub_expr is directly optimizable (100% of that), but it "only" creates 131 new types.
For instantiations:

• dereference.rs for_defined_ty 2.25k (calls)
• impl_hash_with_borrow_str_and_bytes.rs.html - 2.66k ..
• only_used_in_recursion - 2.66k ..
• mixed_read_write_in_expression - 20.8k
• needless_borrows_for_generic_args - 580
• large_enum_variant - 500
• ty/mod.rs 2.15k
• from_over_into - 241
• needless_match - 322
• utils/src/lib.rs - 9099
• functions/results.rs - 9660
• functions/ref_option.rs - 10488
• mut_key.rs - 22.8k (@blyxyas)
• non_copy_const - 1001
• methods/mod.rs - 9361
• self_named_constructors - 929
• useless_asref - 191
• path_buf_push_overwrite.rs - 168
• map_clone.rs - 482
• case_sensitive_file_extension_comparisons - 68
• unnecessary_to_owned - 469
• manual_ok_or - 90
• get_first - 105
• implicit_clone 107
• iter_on_single_or_empty_collections - 79
• unit_return_expecting_ord - 90630 (@blyxyas)
• mut_reference.rs - 17.7k
• transmute_undefined_repr - 349
• needless_pass_by_ref_mut - 80
• use_self.rs - 964
• derive.rs - 2.66k
• rest_pat_in_fully_bound_struct - 89
• large_const_arrays - 589

[Jarcho]

What are you referring to when you say Ty::intern since that isn't actually a function anywhere and types involve multiple interners.

Most type interning would come from either normalization or generic instantiation. Neither of those can be avoided, but you might be able to rearrange things to do them less often.

[blyxyas]

"Small" typo, The function that appears in perf is CtxtInterners::intern_ty. The "call graph" goes something like:

                               /- sty.rs::Ty::new
intern_ty <- mk_ty_from_kind <=
                               \-structural_impls.rs::TypeSuperFoldable::try_super_fold_with <= .. <= binder.rs::EarlyBinder::instantiate <= lots of callers

[profetia]

Is there any guide how I can reproduce this? I could try to do some optimizations thereafter