`elabTermWithHoles _ _ _ (allowNaturalHoles := true)` produces invalid terms

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

The refine' tactic (and the elabTermWithHoles function it calls) can produce terms which do not type-check

Context

[Broader context that the issue occurred in. If there was any prior discussion on the Lean Zulip, link it here as well.]

Steps to Reproduce

Run

import Lean

theorem foo : False := by
  refine' of_decide_eq_true (by rfl : ↑_ = _)

Expected behavior: An error is emitted somewhere on the last line

Actual behavior: The error appears on the foo token

This can be further minimized to

import Lean

open Lean Elab Tactic

theorem foo : False := by
  run_tac 
    let (e, _) ← elabTermWithHoles (← `(of_decide_eq_true (by rfl : ↑_ = _))) (some <| .const ``False []) `foo (allowNaturalHoles := true)
    Meta.check e

or

import Lean

open Lean Elab Tactic

theorem foo : False := by
  run_tac 
    let e ← Meta.withAssignableSyntheticOpaque <|
      elabTermEnsuringType (← `(of_decide_eq_true (by rfl : ↑_ = _))) (some <| .const ``False [])
    Lean.logInfo m!"Produced {e}"
    Meta.check e

where the Meta.check fails.

Versions

Lean 4.18.0-rc1
Target: x86_64-unknown-linux-gnu

Additional Information

[Additional information, configuration or data that might be necessary to reproduce the issue]

Impact

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

Run the minimal reproducer using elabTermWithHoles, elabTermEnsuringType, and Meta.check, then trace how refine' handles the generated holes. Compare the reported location with the expected error on the final tactic line; done means the produced term is rejected at the correct location and does not pass Meta.check as invalid.