Runtime exponential in number of copies of unused variables

Prerequisites

Please put an X between the brackets as you perform the following steps:

• Check that your issue is not already filed:
  https://github.com/leanprover/lean4/issues
• Reduce the issue to a minimal, self-contained, reproducible test case.
  Avoid dependencies to Mathlib or Batteries.
• Test your test case against the latest nightly release, for example on
  https://live.lean-lang.org/#project=lean-nightly
  (You can also use the settings there to switch to “Lean nightly”)

Description

When adding copies of a propositional parameter, which depends on a large proof term of a specific form, to a theorem, the running time of exact sorry seemingly increases exponentially.

Context

See https://leanprover.zulipchat.com/#narrow/stream/270676-lean4/topic/Performance.20issue

Steps to Reproduce

def help (A : Type) : ([] : List A).length % 1 = 0 → Prop := fun _ ↦ True
set_option profiler true

--the running time of "exact sorry" seems exponential in the number of h1
theorem slow {A : Type} (p : Bool)
  (h : ∀ (hp : ([] : List A).length % 1 = 0), help A hp → help A hp)
  (h1 : help A (by cases p <;> omega))
  (h1 : help A (by cases p <;> omega))
  (h1 : help A (by cases p <;> omega))
  (h1 : help A (by cases p <;> omega))
  (h1 : help A (by cases p <;> omega))
  -- (h1 : help A (by cases p <;> omega))
  -- (h1 : help A (by cases p <;> omega))
  (hp : ([] : List A).length % 1 = 0) : help A hp := by
  refine h (by cases p <;> omega) ?_
  exact sorry
--the issue disappers after removing some of the "cases p",
--or replacing "omega" by "simp"
--or replacing A by ℕ
--or replacing "exact sorry" by "exact trivial"

A variant crashing the VS Code infoview is

def help (A : Type) : ([] : List A).length % 1 = 0 → Prop := fun _ ↦ True
set_option profiler true
set_option diagnostics true

example (A : Type) (p : Bool) : ([] : List A).length % 1 = 0 := by
  let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  -- let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  -- let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  -- let h1 : ([] : List A).length % 1 = 0 := by cases p <;> omega
  omega
--also, replacing let by have fixes the issue

Expected behavior: The running time is linear or at least polynomial in the number of copies of h1.

Actual behavior: The running time seems exponential in the number of copies of h1.

Versions

4.11.0-rc2

Impact

Add 👍 to issues you consider important. If others are impacted by this issue, please ask them to add 👍 to it.

Start with the minimal slow reproduction in the issue on Lean 4.11.0-rc2, using set_option profiler true and varying the number of repeated h1 parameters. Compare exact sorry with exact trivial, omega with simp, and the let versus have variants; done means the runtime no longer grows exponentially and the VS Code infoview variant does not crash.