applied: DecoratedCospan's derived Clone/Debug bound D, not D::Apex — uncloneable for every decoration that exists

[tsondru]

Found during #289 (G2-T3), r6 application, 2026-08-22 — while hand-writing PartialEq for DecoratedCospan. The PartialEq case was fixed there; Clone and Debug carry the identical defect and were left alone as out of scope.

The defect

catgraph-applied/src/decorated_cospan.rs:137:

#[derive(Clone, Debug)]
pub struct DecoratedCospan<Lambda, D>
where
    Lambda: Eq + Copy + Debug,
    D: Decoration,
{
    pub cospan: Cospan<Lambda>,
    pub decoration: D::Apex,   // <-- D::Apex, not D
}

derive bounds the type parameters, not the field types. So it generates impl<Lambda, D> Clone for DecoratedCospan<Lambda, D> where Lambda: Clone, D: Clone, requiring D: Clone — but D is a marker that is never stored. The field is D::Apex.

No Decoration marker in the workspace implements Clone:

marker	derives	visibility
PetriDecoration (petri_net.rs:613)	Debug	pub
Trivial (decorated_cospan.rs:371)	Debug	test-module-private
Counter (decorated_cospan.rs:457)	Debug	test-module-private
Circuit (tests/, examples/)	none	local
LocalTrivial (tests/decorated_cospan.rs:98)	none	local

So DecoratedCospan is uncloneable for every decoration that exists — including PetriDecoration, the only one a downstream consumer can name. Measured on audit/G2-T3: cloning a DecoratedCospan<char, Trivial> gives E0599, "the method clone exists for struct DecoratedCospan<char, Trivial>, but its trait bounds were not satisfied … Trivial: Clone is not satisfied".

Debug has the same shape but bites less: PetriDecoration, Trivial and Counter do derive Debug, so it only fails for Circuit / LocalTrivial.

Why this is the same bug #289 already fixed once

#289's r6 pass added PartialEq to this type and hit exactly this. A #[derive(PartialEq)] there compiles but applies to no decoration in the workspace — measured, three E0369s. The fix that shipped was a hand-written impl bounded on D: Decoration and D::Apex: PartialEq rather than D: PartialEq. Clone and Debug need the same treatment and did not get it.

Work

Replace #[derive(Clone, Debug)] with hand-written impls asking only what is stored:

impl<Lambda, D> Clone for DecoratedCospan<Lambda, D>
where
    Lambda: Eq + Copy + Debug,
    D: Decoration,
    D::Apex: Clone,
{ /* ... */ }

Mirror for Debug. Follow the PartialEq impl that shipped in #289 — it is the worked example, in the same file.

Falsify: add a test that clones a DecoratedCospan<char, Trivial> (and Debug-formats a DecoratedCospan<_, Circuit>). Confirm it fails to compile against the derives before the fix — this is a trait-bound defect, so the falsification is a compile failure, not a red assertion — then passes after. Record the E0599 text in the commit.

Consider while there: whether the markers should simply derive Clone/Copy/Debug as unit structs. That also fixes the symptom and is a one-line change per marker, but it leaves the wrong bound in place, so the next marker anyone adds reintroduces it. The impls are the fix; deriving on the markers is a workaround.

Severity

Minor-to-important. No in-tree caller clones a DecoratedCospan, so nothing is broken today — but it is a public generic type that cannot be cloned for any decoration that exists, and the failure lands on the consumer as a confusing E0599 about a bound they never wrote. Group: applied; no dependency on the audit sweep's remaining rows.

[tsondru]

Probe (2026-09-02, base afd76e0, orchestrator; cargo check -p catgraph-applied --test zz_probe_348, throwaway source + log kept under ledger/OB-348/): reproduced — .clone() on DecoratedCospan<char, PetriDecoration<char>> (from PetriNet::to_decorated_cospan) and on DecoratedCospan<usize, LocalTrivial> both fail E0599 ("the method clone exists … but its trait bounds were not satisfied"); format!("{:?}") on the LocalTrivial one fails E0277 ("LocalTrivial doesn't implement Debug"). At afd76e0: #[derive(Clone, Debug)] at decorated_cospan.rs:117; PetriDecoration derives Debug only (petri_net.rs:700-701); Decoration::Apex: Clone + Debug + PartialEq (decorated_cospan.rs:80); the hand-written PartialEq at :129-138. rg -n 'clone\(\)' over decorated_cospan.rs, petri_net.rs, tests/decorated_cospan.rs, tests/hypergraph_laws.rs finds no clone of a whole DecoratedCospan.

Rulings (owner, 2026-09-02):

1. Replace #[derive(Clone, Debug)] on DecoratedCospan with hand-written impl Clone and impl Debug bounded exactly like the shipped PartialEq — Lambda: Eq + Copy + Debug, D: Decoration and nothing more (D::Apex: Clone + Debug is already in the trait). Markers untouched.
2. Pin: a test that clones and {:?}-formats DecoratedCospan<char, PetriDecoration<char>> and a local no-derive marker, asserting the clone == the original; the falsification is the compile failure under the derives — the E0599 / E0277 texts recorded in the stage file, not a runtime red.
3. Files: catgraph-applied/src/decorated_cospan.rs and new or existing files under catgraph-applied/tests/ only.

[tsondru]

Ruling at the delta stop (owner, 2026-09-02; ledger/OB-348/40-review-delta.md M-r2): shadow-struct byte oracle — in t2_7, a locally declared #[derive(Debug)] struct DecoratedCospan { cospan: Cospan<usize>, decoration: () } compared byte-for-byte with the real type under {:?}, {:#?} and {:>90?}. Fix r2 ships unreviewed by the loop cap; the PR body says so.

[tsondru]

Closed by #401 (squash 75725aa): DecoratedCospan gets hand-written Clone and Debug bounded Lambda: Eq + Copy + Debug, D: Decoration (the shipped PartialEq bounds); markers untouched; pins on PetriDecoration and a no-derive marker with the compile-failure falsification recorded, plus a shadow-struct Debug byte oracle. Record: PR body + ledger/OB-348/.