Add Sothera dies-edict regression tests (#5255)

[andriypolanski]

Closes #5255

Summary

• Issue Sothera, the Supervoid? — Whenever a creature I control dies, an opponent should have a choice of creature they control… #5255 mis-associated The End; the report describes Sothera, the Supervoid.
• Sothera's per-opponent exile edict is already on main (feat(parser): Sothera, the Supervoid dies-edict exile + end-step linked reanimation #5404).
• Add issue-specific parse + runtime regression tests for player_scope: Opponent exile when a controlled creature dies.

Test plan

• issue_5255_sothera_dies_trigger_parses_opponent_scoped_exile_edict
• issue_5255_sothera_dies_trigger_exiles_opponent_creature
• sothera_each_opponent_exiles_own_creature (existing)
• sothera_reanimates_only_linked_creature_with_two_counters (existing)

Verification

cargo fmt --all
cargo test -p engine issue_5255 -- --nocapture
cargo test -p engine sothera -- --nocapture

[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!

[matthewevans]

Blocked — the test binary does not compile: this PR registers a module whose file it does not ship. Fix is one line; the linked-issue claim also needs correcting.

🔴 Blocker 1 — main.rs registers a module from your other branch (build break)

crates/engine/tests/integration/main.rs:781 adds:

mod issue_5263_chaos_warp;

but this PR ships only issue_5255_sothera_supervoid.rs. issue_5263_chaos_warp.rs lives in #5600, not here. CI says it plainly:

error[E0583]: file not found for module `issue_5263_chaos_warp`

That single stray line is why all four Rust checks fail (lint, coverage-gate, and both test shards) — the whole integration test binary fails to build, so nothing in it runs, including your Sothera test.

Fix: drop the mod issue_5263_chaos_warp; line from this PR's main.rs. Keep only mod issue_5255_sothera_supervoid;. It looks like this branch was cut from your #5600 branch and picked up its registration; rebasing onto a clean origin/main will also clear it.

🔴 Blocker 2 — Closes #5255 is not supported by the issue

The PR body says Closes #5255, which would auto-close that issue on merge. Two problems:

1. The issue is internally inconsistent. #5255 is titled "Sothera, the Supervoid?" and its Summary describes a dies-edict trigger — but its "Oracle text (verified from card-data.json)" section is for a different card, The End (Cards: the end): "This spell costs {2} less to cast if your life total is 5 or less. Exile target creature or planeswalker…". The Discord extraction attached the wrong card, so the issue does not actually establish the Sothera behavior you are testing.
2. Parser status: has_gaps on that issue — the mechanic is flagged as not fully parsed. A test asserting the fixed behavior is premature until the fix lands.

Fix: remove the closing keyword (use Refs #5255 if you want the backlink). The issue needs re-triage — its Oracle text does not match its title — and should not be silently closed by a test PR.

🟡 Non-blocking

Test placement is correct: crates/engine/tests/integration/ with a mod line, so no no_top_level_test_binaries violation. That part is right.

Recommendation: request-changes. Remove the stray mod issue_5263_chaos_warp;, rebase onto clean origin/main, and drop Closes #5255. Once the binary compiles, the Sothera test can actually be evaluated on its merits — right now it never runs.

[github-actions]

Parse changes introduced by this PR

✓ No card-parse changes detected.

[matthewevans]

Thanks for the quick turnaround on the stray mod — that blocker is genuinely fixed, and CI is fully green on b49e289fb6 (13/13, real test shards).

I'm going to decline this one, though, and I want to be specific about why so it's useful rather than just a "no".

The coverage already exists on main. crates/engine/tests/integration/sothera_supervoid_edict_reanimate.rs (registered at main.rs:619) already covers this exact card, this exact Oracle text, and this exact behavior — il_each_opponent_exiles_own_creature drives the dies-edict through the real resolution path. Your own doc comment notes the new test falls back to "the parsed dies-edict body (same as sothera_supervoid_edict_reanimate)", so we'd be adding a second copy of a test we already run.

The one genuinely new idea is neutralized by its own fallback. Driving the trigger through process_triggers instead of hand-building the ability would be a real improvement in fidelity — that's a seam worth covering. But the test falls back to the parsed body when the trigger path doesn't fire, which means it can pass without ever exercising process_triggers. A test with a fallback can't tell you which path ran, so it can't fail for the reason you want it to. If you want to pursue that, a test that drives process_triggers and fails when the trigger doesn't collect would be genuinely valuable — that's the version I'd merge.

On issue selection — and this is the most useful thing I can give you. Two things to check before opening a regression-test PR:

• Skip issues labeled status:fixed-unreleased. That label means we've already determined the fix landed; a regression test there is testing behavior that already works. (Chaos Warp — Had the AI opponent cast a [[Chaos Warp]] on one of my creatures. #5263 on your other PR had this label too.)
• Skip issues labeled status:needs-repro / classifier:cannot-determine. Sothera, the Supervoid? — Whenever a creature I control dies, an opponent should have a choice of creature they control… #5255 is both — nobody has confirmed it reproduces, and its card extraction is wrong (it pulled The End, not Sothera, as you correctly spotted). Auto-closing it via Closes #5255 would mark a Discord user's unreviewed report as resolved without anyone ever determining whether Sothera actually misbehaves. I'd rather leave it open for real triage.

There are ~19 open bug-labeled issues that are none of those things, and your track record on real engine work is excellent — #5584's ForEachCategory parameterization was the best architectural call I reviewed all week. That's where your time pays off most.

Closing this one; #5255 stays open for triage.