`refine'` elaborated to a type-incorrect term

[llllvvuu]

Prerequisites

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

• [x] Check that your issue is not already filed: https://github.com/leanprover/lean4/issues
• [x] Reduce the issue to a minimal, self-contained, reproducible test case. Avoid dependencies to Mathlib or Batteries.
• [x] 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

Playground

/-- info: Try this: ne_of_beq_false (Eq.refl false) -/
#guard_msgs (info) in
example : ¬0 = 0 := show_term by
  refine' ne_of_beq_false (by rfl : ↑_ = _)

/-- info: Try this: ne_of_beq_false (sorryAx ((0 == 0) = false) true) -/
#guard_msgs (info) in
example : ¬0 = 0 := show_term by
  refine ne_of_beq_false (by rfl : ↑_ = _)

/-- info: Try this: ne_of_beq_false (sorryAx ((0 == 0) = false) true) -/
#guard_msgs (info) in
example : ¬0 = 0 := show_term by
  refine' ne_of_beq_false (by rfl : _ = _)


/-- info: Try this: ne_of_beq_false (sorryAx ((0 == 0) = false) true) -/
#guard_msgs (info) in
example : ¬0 = 0 := show_term by
  refine' ne_of_beq_false (rfl : ↑_ = _)

The difference between the first and second examples seems attributable to withAssignableSyntheticOpaque.

Context

This was discovered while doing a tactic proof and not seeing an error message at the tactic.

Steps to Reproduce

1. Build the above code or open it in an LSP-enabled editor or web playground

Expected behavior: The tactic fails in all cases, thus elaborating with sorry

Actual behavior: The tactic succeeds in the first case, elaborating to a type-incorrect term

Versions

4.10.0-rc2

Impact

Add :+1: to issues you consider important. If others are impacted by this issue, please ask them to add :+1: to it.