Fix Vindictive Lich

[mike-theDude]

Summary

Fixes the Vindictive Lich modal-trigger softlock by capping modal choices to the largest set whose combined target slots can satisfy modal target constraints.

Closes #4959.

Files changed

• crates/engine/src/game/ability_utils.rs
• crates/engine/src/game/casting.rs
• crates/engine/src/game/engine.rs
• crates/engine/src/game/triggers.rs
• crates/engine/src/game/engine_trigger_target_tests.rs

CR references

• CR 700.2a-b
• CR 603.3c
• CR 115.1

Track

Developer

LLM

Model: codex-5
Thinking: high
Tier: Standard

Gate A

./scripts/check-parser-combinators.sh exited 0 with no stdout.

Anchored on

• crates/engine/src/game/ability_utils.rs:731 — existing modal per-mode target legality filtering derives target constraints and uses target-slot assignment checks.
• crates/engine/src/game/engine_modes.rs:108 — existing ability-mode selection derives target constraints before routing chosen modes into target selection.
• crates/engine/src/game/engine.rs:5242 — existing pending-trigger modal cleanup removes illegal no-mode triggers from the stack.

Verification

• cargo fmt --all -- --check — clean
• ./scripts/check-parser-combinators.sh — clean
• cargo test -p engine vindictive_lich — 3 passed / 0 failed
• cargo clippy --all-targets -- -D warnings — clean
• cargo test -p engine — 16,045 unit tests passed; 2,588 integration tests passed; 0 failed
• ./scripts/gen-card-data.sh — clean
• cargo coverage — Vindictive Lich: supported: true, gap_count: 0, no parser warnings
• cargo semantic-audit — Vindictive Lich: 0 findings

Scope Expansion

None.

Validation Failures

None.

CI Failures

None.

[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

✓ No card-parse changes detected.

[matthewevans]

Approving — reviewed on head e50f5929.

Card + rules verification (live): Vindictive Lich's Oracle text is "When this creature dies, choose one or more. Each mode must target a different player. • Target opponent sacrifices… • Target opponent discards two cards. • Target opponent loses 5 life." (Scryfall). I confirmed the parser emits ModalSelectionConstraint::DifferentTargetPlayers for that exact reminder phrase (oracle_modal.rs), so this fix is not inert — the real card parses to a modal carrying the constraint and the new cap fires on it. The fixture tests hardcode the constraint, but the end-to-end parse path is present.

The bug is real: per-mode legality (filter_modes_by_target_legality) only proves each mode is individually legal; it can't see that DifferentTargetPlayers makes a 3-mode selection impossible in a duel with one opponent. Before this, the engine would let you choose all three then strand target assignment.

Architecture — single authority across all flows: modal_choice_with_target_assignment_limit is built once and called from all three modal-dispatch sites (activate / pending-trigger / dispatch). That's the correct generalization for a 3-flow cluster rather than three ad-hoc caps. It composes existing primitives (build_chained_resolved, build_target_slots, has_legal_target_assignment_for_ability).

Rules mapping is correct per context: activated abilities (casting.rs) return Err when no legal mode (CR 700.2a — activation rejected), while triggered abilities (engine.rs/triggers.rs) drop from the stack (CR 700.2b + CR 603.3c — removed if no mode chosen). That asymmetry is exactly right.

Tests are discriminating: duel caps max_choices to 1, 3-player FFA caps to 2 (two opponents), and over-selection (SelectModes{[0,1]}) is rejected while [0] proceeds to Priority.

Non-blocking (separate issue, pre-existing): the recognizer at oracle_modal.rs uses a verbatim lower == "each mode must target a different player" match, which is the no-verbatim-string-match pattern CLAUDE.md prohibits — but it predates this PR and is out of scope here. Not a blocker.

Solid engine work.