Address the three nits from the #152 review

[milyin]

Finding 6 of the review on #152 — three small items, plus a follow-up fixing the test Copilot caught.

1. wire_fixed_returns unrequires the peeled Vec element

The shape fold peels three layers (Option, Vec, &) and the no_converter branch unrequired two. With only a Vec<E>-returning function declared, the bare E requirement was left to the adapter's boundary_only_types — which covers it for JniGen today, making it a coupling rather than a property. The peel happens here, so the unrequire belongs here.

a_vec_only_sum_return_drops_the_bare_requirement covers it. The existing sum_return_layers_ride_the_shape_fold cannot: its Option<Reading> fixture unrequires the bare type through a different layer, so it would keep passing if the Vec element were left required.

The test seeds the requirement explicitly (Copilot caught that the first version did not). A scan registers only the top-level return, so Reading was never in required_outputs_scan and the assertion held whether or not the fix was present — it passed while testing nothing. require_output(Reading) before apply_sum_returns reproduces what an adapter that does require the element leaves behind, and a precondition assert pins that the fixture starts out in that state.

Verified to discriminate:

wire_fixed_returns	Result
with unrequire_output(&core)	pass
with that line removed	FAILED

2. The kt_handle_target comment in render.rs

It read as if it fixed a live bug ("so a tag-gated variant handle locks the expression it actually came from"). It does not: build_flat_sum_field returns None for any projection payload, so a tag-gated handle leaf cannot reach that code today. The change is correct and defensive, and the comment now says that — the access template is the leaf's own answer to "where did this come from", and deriving it twice is how the two drift apart.

3. cbindgen.toml trailing newline

Pre-existing, now with added content. Added.

Verified

cargo test --all --all-features (392 lib), cargo fmt --check, clippy --all-targets --no-default-features --all-features -D warnings, examples/regen-check.sh — all clean.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR addresses three follow-up nits from the #152 review, focused on making sum-type return planning less adapter-coupled, clarifying a Kotlin/JNI comment to reflect current behavior, and normalizing an example config file’s formatting.

Changes:

• Ensure wire_fixed_returns drops the required-output scan registration for the peeled Vec element (core) when sums have no whole-value converter.
• Update the kt_handle_target explanation in JNI/Kotlin rendering to describe the current (defensive) rationale and non-reachability for projection payloads.
• Add a focused test for the Vec<sum>-only fixture and add the missing trailing newline to cbindgen.toml.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 1 comment.

File	Description
prebindgen/src/api/lang/jnigen/jni/render.rs	Clarifies comment explaining why handle lock targets use the leaf’s access template.
prebindgen/src/api/core/unfold/tests.rs	Adds a test covering Vec<Reading>-only sum return invariants.
prebindgen/src/api/core/unfold.rs	Extends no_converter unrequire logic to include the peeled core type.
examples/example-cbindgen/cbindgen.toml	Adds trailing newline at EOF.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[milyin]

Correct, and fixed in 9b4699e — this names the caveat the test's own doc comment admitted, which was a sign the test should have been better rather than better-documented.

Seeding is exactly the right shape: require_output(Reading) before apply_sum_returns reproduces what an adapter that does require the peeled element leaves behind, which is the state unrequire_output(&core) exists for. A precondition assert pins that the fixture starts out in it, so the seeding cannot silently stop working either.

let bare: syn::Type = syn::parse_quote!(Reading);
reg.require_output(&bare, &crate::SourceLocation::default());
assert!(
    reg.required_outputs_scan.contains(&TypeKey::from_type(&bare)),
    "fixture precondition: the bare element starts out required"
);

Verified to discriminate, which it previously did not:

wire_fixed_returns	Result
with unrequire_output(&core)	pass
with that line removed	FAILED

The doc comment no longer hedges about pinning-rather-than-catching, because it now catches.

[Copilot]

Pull request overview

Copilot reviewed 3 out of 4 changed files in this pull request and generated no new comments.