fix(parser): split bare-"and" "... and each opponent/player <verb>" conjunct

[galuis116]

Fixes #3582.

Summary

When a "you …" clause is joined to a second "each opponent/player <verb> …" clause by a bare " and " (no comma), the second conjunct was silently dropped — only the first half resolved:

• Slitherwisp — "you draw a card and each opponent loses 1 life" → only the draw.
• Curry Favor — "You gain X life and each opponent loses X life, where X is the number of Knights you control" → only the life gain.
• Disinformation Campaign, Bad Deal ("each opponent discards two cards"), Clockwork Fox ("each opponent draws a card").

Root cause

starts_bare_and_clause_lower did not treat "each opponent "/"each player " + a conjugated player verb as a fresh clause start, so the second conjunct never began a new chunk and was swallowed by the first effect. The comma splitter already lists those subjects (so comma-joined / separate-sentence forms work). "each opponent" is intentionally not a bare clause starter so a bare-noun continuation like Goblin Chainwhirler's "deals 1 damage to each opponent and each creature you control" stays a single DamageAll.

Fix (parser-only; no new engine variant)

Add starts_each_player_predicate_clause_lower, modeled on the existing starts_target_continuous_clause_lower sibling, wired as one trailing .or() arm in starts_bare_and_clause_lower. It matches "each opponent "/"each player " only when immediately followed by a conjugated player-action verb (loses/gains/draws/discards/mills/sacrifices/exiles) — an immediate tag on the verb (not take_until), so a bare-noun continuation (no verb) never matches and Goblin Chainwhirler stays one chunk. Once the conjunct is peeled, the existing strip_each_player_subject + per-player scope lowering handles it end-to-end; for Curry Favor the bare-and split emits a ClauseBoundary::Comma, so the "where X is the number of Knights" binding reaches both conjuncts.

No runtime change — the player-scoped effects (LoseLife/GainLife/Draw/Discard/Mill/Sacrifice/Exile under PlayerFilter::Opponent/All) already exist and lower today. CR 102.2 (opponent), CR 119.3 (gain/lose life), CR 121.1 (draw), CR 608.2c (follow the whole text).

Tests

• Splitter unit (starts_bare_and_clause): positive for "each opponent loses 1 life" / "discards a card" / "discards two cards" / "draws a card" / "loses x life" / "each player loses 2 life"; negative for "each creature you control" / "each opponent and each creature you control" / "each opponent's creatures".
• Chunk-split (clause_texts): "you draw a card and each opponent loses 1 life" → two chunks; Goblin Chainwhirler stays a single chunk.
• Full-card parse (parse_effect_chain): Slitherwisp has both Draw and opponent-scoped LoseLife (amount 1); Curry Favor has both GainLife and opponent LoseLife bound to the same X; Bad Deal has the opponent Discard (count 2). Revert-discriminating — without the new arm the conjunct is dropped and these fail.

Verification

cargo +nightly test -p engine --lib: 0 failures (152 sequence + 51 bare_and tests pass, incl. Goblin Chainwhirler no-split and Skulduggery). clippy -D warnings: clean. rustfmt --check: clean. Parser-combinator gate: clean for this diff.

[gemini-code-assist]

Code Review

This pull request introduces a parser helper starts_each_player_predicate_clause_lower to correctly split player-scoped predicate clauses joined by a bare " and " (such as in Slitherwisp or Curry Favor), preventing them from being swallowed by preceding effects. It also adds comprehensive unit and end-to-end integration tests. The feedback suggests expanding the list of recognized player-action verbs (e.g., adding "creates", "puts", "chooses", etc.) to ensure complete sibling coverage for other common card effects.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

[MEDIUM] Missing conjugated player-action verbs in starts_each_player_predicate_clause_lower.

Why it matters: Common player-action clauses like each opponent creates..., each opponent puts..., each opponent chooses..., or each opponent reveals... will be swallowed and dropped because their verbs are not matched by this parser arm.

Suggested fix: Add creates , puts , chooses , reveals , shuffles , and returns to the alt combinator.

        alt((
            tag("loses "),
            tag("gains "),
            tag("draws "),
            tag("discards "),
            tag("mills "),
            tag("sacrifices "),
            tag("exiles "),
            tag("creates "),
            tag("puts "),
            tag("chooses "),
            tag("reveals "),
            tag("shuffles "),
            tag("returns "),
        )),

References

1. L2. Sibling coverage: If a parser arm or string was extended, ensure all sibling variants/verbs are covered. ^(link)
2. Avoid verbatim string equality for parsing Oracle phrases. Instead, decompose compound phrases into modular, reusable parsers and compose them using idiomatic combinator aggregates like nested alt and tag sequences.

[matthewevans]

Approved after deep review. Splits the bare-"and" "X and each opponent/player " conjunct (Slitherwisp, Curry Favor, Bad Deal, Oloro, Gray Merchant — 18-card class) that was being swallowed into the first effect. This is a high-blast-radius change to sequence.rs's clause splitter, so the gating concern was over-split regression — and the guard is tight: starts_each_player_predicate_clause_lower (sequence.rs:1519) fires ONLY when the post-and remainder re-states a fresh player subject (each opponent/each player) + a conjugated player-verb, so same-subject continuations ("gets +1/+1 and gains flying") structurally can't reach it, and Goblin Chainwhirler ("...and each creature you control") is excluded. Three regression negatives prove the over-split doesn't fire (bare-noun continuation, GCW compound, possessive). Plugs into the canonical bare-and .or() cluster (inherits #3509's quote/paren boundary guards); swallowed-clause ratchet moves safe (only adds a split). End-to-end test drives Slitherwisp/Curry Favor/Bad Deal through parse_effect_chain, both conjuncts real + opponent-scoped. Subject parameterized (each opponent/each player). CR 102.2/119.3/121.1/608.2c verified. Enqueuing.

[matthewevans]

Approved & enqueued — nice, tightly-guarded work. One follow-up (non-blocking, for a future PR): re: Gemini's verb-expansion suggestion, I checked the card pool. The speculative verbs (creates/puts/chooses) match zero cards in this structure, so leaving them out is correct scoping. But "returns" is a real 2-card sibling gap: All Hallow's Eve and Underworld Cerberus both have "…and each player returns all creature cards from their graveyard to [the battlefield/their hand]", which the current verb set (loses/gains/draws/discards/mills/sacrifices/exiles) doesn't recognize, so their second clause stays swallowed (no regression — they were already swallowed). Worth adding "returns" in a fast-follow, BUT verify the split actually unlocks them: "each player returns all creature cards from their graveyard to X" is a mass-graveyard-return effect — confirm it parses to a real Effect post-split (your each_player_predicate_conjunct_parses_end_to_end pattern) rather than surfacing an Unimplemented fragment, before adding the verb. Not needed for this PR.