The specialization pass does not specialize for oversaturating arguments
Nobody has claimed this yet.
- Dominant language
- Lean
- Stars
- 9.2k
- Forks
- 990
- Avg merge
- 1d 17h
- Merged PRs (30d)
- 175
Description
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
Specialization does not appear to specialize for oversaturating arguments of a call to a function with a @[specialize] annotation.
Steps to Reproduce
@[specialize]
def foo {α} : Nat → (α → α) → α → α
| 0, f, a => f a
| n+1, f, a => foo n f a
-- The following example demonstrates that `f` is properly specialized:
set_option trace.Compiler.saveBase true in
/--
trace: [Compiler.saveBase] size: 6
def foo._at_._example.spec_0 x.1 x.2 : Nat :=
cases x.1 : Nat
| Nat.zero =>
let _x.3 := 1;
let _x.4 := Nat.add x.2 _x.3;
return _x.4
| Nat.succ n.5 =>
let _x.6 := foo._at_._example.spec_0 n.5 x.2;
return _x.6
[Compiler.saveBase] size: 2
def _example n : Nat :=
let _x.1 := 5;
let _x.2 := foo._at_._example.spec_0 n _x.1;
return _x.2
-/
#guard_msgs in
example {n} := foo n (· + 1) 5
-- The following example demonstrates that while `f` is properly specialized, `a` is not:
set_option trace.Compiler.saveBase true in
/--
trace: [Compiler.saveBase] size: 6
def foo._at_._example.spec_0 x.1 x.2 _y.3 : Nat :=
cases x.1 : Nat
| Nat.zero =>
let _x.4 := x.2 _y.3;
let _x.5 := x.2 _x.4;
return _x.5
| Nat.succ n.6 =>
let _x.7 := foo._at_._example.spec_0 n.6 x.2 _y.3;
return _x.7
[Compiler.saveBase] size: 4
def _example n : Nat :=
fun _f.1 x.2 : Nat :=
let _x.3 := 1;
let _x.4 := Nat.add x.2 _x.3;
return _x.4;
let _x.5 := 5;
let _x.6 := foo._at_._example.spec_0 n _f.1 _x.5;
return _x.6
-/
#guard_msgs in
example {n} := foo n (fun f a => f (f a)) (· + 1) 5
Expected behavior: foo is specialized not only for (fun f a => f (f a)) but also for (· + 1).
Actual behavior: foo is specialized for (fun f a => f (f a)) but not for (· + 1).
Versions
Lean 4.26.0-nightly-2025-10-22
Target: x86_64-unknown-linux-gnu
Impact
Add 👍 to issues you consider important. If others are impacted by this issue, please ask them to add 👍 to it.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by running the minimal reproducer and examining the specialization pass behavior for the two examples. Done means the second example specializes both the higher-order function and the oversaturating (· + 1) argument, matching the expected trace.
Written by the indexing model from the issue text.
Assessment
- Domain
- compilers
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 38/100