fix(engine): bind Maze of Ith's bidirectional prevent anaphor to its own target

[rykerwilliams]

Summary

Fixes #1094 — [[Maze of Ith]] ({T}: Untap target attacking creature. Prevent all combat damage that would be dealt to and dealt by that creature this turn. — verified via Scryfall, from The Dark, 1994) doesn't untap the creature or prevent it from dealing or taking damage.

Root cause (originally diagnosed by @matthewevans in the issue's coverage-classifier comment): the ability parsed as Untap { target } + child PreventDamage { target: Any } — the "that creature" anaphor was dropped to the default Any instead of binding to the same creature via TargetFilter::ParentTarget. Separately, the clause is bidirectional ("dealt to and dealt by that creature") and the parser had no representation at all for the "dealt by" half (damage the creature deals, not just damage dealt to it).

Effect::PreventDamage's target + damage_source_filter fields combine with AND semantics (game/replacement.rs), so "recipient==X OR source==X" cannot be expressed as one shield. The fix splits the clause into two PreventDamage nodes chained via SubAbilityLink::SequentialSibling — a recipient-scoped "to" shield and a source-scoped-only "by" shield.

Two latent runtime gaps in game/effects/prevent_damage.rs, both required for the split to actually work, not scope creep:

• resolve_source_filter had no arm for a bare ParentTarget damage-source filter (only ParentTargetSlot{index}) — added one.
• source_scoped_prevent's gate only recognized the pre-existing And{ParentTargetSlot,...} shape, so a bare ParentTarget source filter fell through to host_on_targets, which auto-filled valid_card: SelfRef and silently re-imposed the recipient constraint the source-only shield must not have — widened the gate.

Shipped scope

7 cards sharing this exact singular-anaphor grammar ("prevent...dealt to and dealt by <that creature|it|the creature>"): Maze of Ith, Ebony Horse, Elvish Scout, Foxfire, Ith High Arcanist, Maze of Shadows, Delirium.

Deliberately deferred (separate follow-up issues, not silently dropped):

• Self-ref-activated cohort ("this creature" — Deftblade Elite, Urborg Phantom, Moonlight Geist) — would need its own TargetFilter::SelfRef recipient-scoping carve-out in prevent_damage.rs::resolve(), currently untested and out of scope here.
• Plural anaphor (Energy Arc, "those creatures").
• Multi-target + unless-pay branching (Winter's Chill) — verified via trace that parent_target_available is guaranteed false when its inner "prevent...that creature..." clause parses (its preceding sibling clause has a recognized condition.is_some(), which makes the backward target-referent walk bail before reaching the earlier typed target), so the new machinery is a guaranteed no-op there — no guard needed.

Files changed

• crates/engine/src/parser/oracle_target.rs — parse_anaphoric_target_ref, a thin wrapper delegating to the existing parse_target building block (gated on a new parent_target_available: bool).
• crates/engine/src/parser/oracle_ir/ast.rs — parent_target_available: bool field on UtilityImperativeAst::Prevent.
• crates/engine/src/parser/oracle_effect/imperative.rs — threads the field through all 4 construction/consumption sites; factored parse_prevention_amount out for shared reuse.
• crates/engine/src/parser/oracle_effect/lower.rs — try_parse_bidirectional_prevent, the sibling-splitting interceptor.
• crates/engine/src/game/effects/prevent_damage.rs — the two runtime fixes above.
• crates/engine/src/game/effects/mod.rs — first_object_target promoted to pub(crate) for reuse.
• crates/engine/src/game/scenario.rs — add_land_from_oracle, a reusable test-harness builder (no existing helper places a land on the battlefield with parsed Oracle text).
• crates/engine/src/parser/oracle_tests.rs — unit tests for all 7 family cards + negative gate + sibling-clause survival (Foxfire's trailing draw trigger, Delirium's preceding tap/damage clauses).
• crates/engine/tests/integration/maze_of_ith_untap_bidirectional_prevent.rs (new) — discriminating runtime tests driving the real activate/combat pipeline, including a hostile un-Mazed attacker/blocker fixture and a defending-player-life-total fixture.

CR references

CR 615, CR 615.1a, CR 615.11, CR 511.2, CR 608.2c, CR 302.6, CR 609.7, CR 701.26a/b — all verified against docs/MagicCompRules.txt.

Test plan

• Verified all 7 shipped cards' Oracle text via Scryfall (not paraphrases)
• cargo fmt --all clean
• cargo clippy -p engine --lib -- -D warnings clean
• ./scripts/check-parser-combinators.sh (Gate A): 0 violations
• cargo test -p engine --lib: 15964 passed, 0 failed
• cargo test -p engine --test integration: 2563 passed, 0 failed (includes both new Maze of Ith tests)
• Live-revert-and-rerun: independently reverted the parser split, the resolve_source_filter fix, and the source_scoped_prevent gate widening — each flips both integration tests to failing; restored and re-confirmed green
• Went through 4 rounds of independent implementation review (2 planning rounds + 2 implementation rounds) before opening this PR — findings included a misapplied CR 615.5 citation (corrected to 615.1a + 608.2c) and a hand-rolled anaphor recognizer that duplicated the existing parse_target building block (refactored to delegate)

Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com

https://claude.ai/code/session_01XbgwGxbU9NHN9kou9isp8K

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[github-actions]

Parse changes introduced by this PR · 14 card(s), 3 signature(s) (baseline: main 02bb03df2ba4)

14 card(s) · ability/PreventDamage · field target: any target → parent target

Examples: Delirium, Djeru's Resolve, Ebony Horse (+11 more)

7 card(s) · ability/PreventDamage · added: PreventDamage (amount=All, scope=CombatDamage, target=any target)

Examples: Delirium, Ebony Horse, Elvish Scout (+4 more)

1 card(s) · ability/PreventDamage · added: PreventDamage (amount=All, scope=AllDamage, target=any target)

Examples: Kiora, the Tide's Fury

[matthewevans]

APPROVE — maintainer sign-off.

This converts Maze of Ith's Prevent sentence from a single unbound PreventDamage { target: Any, scope: CombatDamage } (an effectively global combat shield) into a correctly-bound anaphor, at the right seam: the anaphor threads ParseContext.parent_target_available → UtilityImperativeAst::Prevent → parse_anaphoric_target_ref → TargetFilter::ParentTarget, and resolves at resolve() through the ability's own inherited target list (first_object_target(ability.targets)), never a board scan. CR 608.2c snapshot semantics hold.

Spot-verified against PR head 25c99c80:

• parse_anaphoric_target_ref reuses parse_target and returns Some only when matches!(filter, TargetFilter::ParentTarget) — it declines (returns None) on a fresh typed filter or the Any fallback rather than guessing. It also short-circuits None when parent_target_available is false.
• resolve_source_filter's bare-ParentTarget arm maps to first_object_target(ability_targets) → SpecificObject { id } (or None) — the parent's own chosen object, no global rescan. Distinct from parse_event_context_ref's "that creature" → TriggeringSource trigger path, which is correctly avoided in this non-trigger context.

Parse-diff confrontation (do not re-litigate the blast radius). Sticky baseline main 02bb03df, 14 cards / 3 signatures:

• 14 cards rebind PreventDamage.target from any target → parent target. This is the whole prevent-with-anaphor class.
• 7 cards additionally gained a CombatDamage "by" node — precisely the bidirectional+combat cards: Maze of Ith, Ebony Horse, Elvish Scout, Delirium, Foxfire, Maze of Shadows, Ith High Arcanist ("...dealt to and dealt by [it/that creature]...").
• 1 card (Kiora, the Tide's Fury) gained an AllDamage "by" node — it is bidirectional but all-damage ("Prevent all damage … dealt to and dealt by that permanent").
• Djeru's Resolve is single-direction ("dealt to it" only) and correctly received only the rebind, no "by" node — verifying the "by" node is gated on the "and dealt by" clause, not applied blanketly.

The blast radius exactly equals the stated class; every gained/changed card is explained.

Test discrimination. maze_of_ith_untaps_and_prevents_both_directions drives the real activation + combat pipeline and asserts BOTH shield directions (Mazed attacker and its blocker each take 0) AND that an un-Mazed attacker/blocker pair in the same combat still trades 2/2. Under the pre-fix global shield the un-Mazed pair would be shielded to 0/0, so those == 2 assertions FAIL on revert — a genuine negative-control discriminator. maze_of_ith_prevents_mazed_attacker_damage_to_defending_player asserts the defending player loses exactly the un-Mazed attacker's 3 (not 2+3=5), discriminating the source-scoped "by" shield.

Bidirectional "by" leg is a SequentialSibling with damage_source_filter = ParentTarget; recipient "to" leg is scoped to the parent-target object. Per-identity scoping, not a global shield. scenario.rs::add_land_from_oracle fills a real gap for the land-with-abilities class (existing helpers hardcode a mana ability, use Zone::Hand, or are creature-only). Fixtures use verbatim Oracle text.

All required checks green. Right seam, house idioms (reuses parse_target/ParentTarget rather than proliferating variants), discriminating runtime tests including a negative-control pair.