test: stabilize live regression fixtures

AGENTS.md

@@ -41,7 +41,7 @@
      }
    - 同一worktreeでの重複起動は絶対に避ける (被ると競合)。
    - 研究完了前に実装を開始しない。
-    - **すべてのコマンドに確実なタイムアウトを付与** (`test_smoke_80` は子プロセス `cppfm` を fork するため親だけを殺すと孤児化)。例: `timeout --foreground --kill-after=5 120 cmake -B build -G Ninja` / `timeout --foreground --kill-after=5 300 cmake --build build -j2` / `timeout --foreground --kill-after=5 450 ./build/test_smoke_80 ./build/cppfm`。終了確認は `pgrep -a -f 'cppfm --por[t]'` の後、対象が確認できた場合だけ `timeout --foreground --kill-after=5 10 pkill -9 -f 'cppfm --por[t]'` を使う。`ctest -R smoke80 --timeout 450` でも可。
+    - **すべてのコマンドに確実なタイムアウトを付与** (`test_smoke_80` は子プロセス `cppfm` を fork するため親だけを殺すと孤児化)。例: `timeout --foreground --kill-after=5 120 cmake -B build -G Ninja` / `timeout --foreground --kill-after=5 300 cmake --build build -j2` / `timeout --foreground --kill-after=5 600 ./build/test_smoke_80 ./build/cppfm`。終了確認は `pgrep -a -f 'cppfm --por[t]'` の後、対象が確認できた場合だけ `timeout --foreground --kill-after=5 10 pkill -9 -f 'cppfm --por[t]'` を使う。`ctest -R smoke80 --timeout 600` でも可。
 
 4. 並行で開発させ、すべてが終わったあと diff をレビューしマージ
    - `git diff main --stat` と `git log --oneline --graph` で確認
@@ -99,7 +99,7 @@ timeout --foreground --kill-after=5 60 ./build/test_native ./build/cppfm
 timeout --foreground --kill-after=5 30 ./build/test_scoreboard_reset               # 22/22 PASS expected (ctest scoreboard_reset TIMEOUT 30)
 timeout --foreground --kill-after=5 60 ./build/test_spec_wire                     # 392 PASS 0 FAIL 0 SKIP (wire byte-identical lock; 実績は CURRENT_STATE §2)
 timeout --foreground --kill-after=5 30 ./build/test_fuzz                          # 23 PASS 0 FAIL (fuzz 23 cases, TIMEOUT 30)
-timeout --foreground --kill-after=5 450 ./build/test_smoke_80 ./build/cppfm        # 212 PASS 0 FAIL (実績は CURRENT_STATE §2) — 450s (600s under load)
+timeout --foreground --kill-after=5 600 ./build/test_smoke_80 ./build/cppfm        # 225 PASS 0 FAIL (実績は CURRENT_STATE §2)
 pgrep -a -f 'cppfm --por[t]' || true
 timeout --foreground --kill-after=5 10 pkill -9 -f 'cppfm --por[t]' 2>/dev/null || true
 timeout --foreground --kill-after=5 5 sleep 1
@@ -111,7 +111,7 @@ timeout --foreground --kill-after=5 400 python3 tests/soak_test.py --duration 30
 timeout --foreground --kill-after=5 60 python3 tools/bench_chunk_gen.py --view-distance 32 --chunks 4225 --dry --strict  # plan45 O-11 view32
 # nightly 24h (plan45 O-06 — ctest外): nohup python3 tests/soak_test.py --duration 86400 --binary ./build/cppfm > /tmp/soak24h.log 2>&1 &
 ctest -R "native|scoreboard_reset|spec_wire|fuzz" --output-on-failure --timeout 60
-ctest -R smoke80 --output-on-failure --timeout 450   # 600 under load
+ctest -R smoke80 --output-on-failure --timeout 600
 ```
 
 ## 6. Architecture Quick Map

CMakeLists.txt

@@ -487,7 +487,6 @@ if(TARGET cppfm_embedded_resources)
 endif()
 add_test(NAME smoke80 COMMAND test_smoke_80 $<TARGET_FILE:cppfm>
          WORKING_DIRECTORY ${CMAKE_SOURCE_DIR})
-set_tests_properties(smoke80 PROPERTIES TIMEOUT 450)
 
 # plan51 §1–2: generation-safe handle and selective routing invariants.
 add_executable(test_jvm_handles

docs/CURRENT_STATE.md

@@ -82,7 +82,7 @@ artifact boundaries.
 | Official primitive oracle | `INSPECTED` | SHA-1-pinned Mojang 1.21.4 server artifact; VarInt, VarLong, Varint21 frame splitter, and packed BlockPos decoder/packing bytecode reviewed, not executed |
 | Primitive and stream regressions | `PASS` | `test_goal_network_bugs` 79/79; independent signed endpoints, non-minimal forms, Java terminal-payload truncation, generic sixth/eleventh-byte rejection/consumption, strict three-byte Varint21 frame rejection, plaintext/encrypted stream alignment, big-endian fixed-width fields, and packed Position axis limits |
 | Related CTest targets | `7/7 PASS` | `native`, `spec_wire`, `jvm_native_bridge`, `core_safety`, `goal_network_bugs`, `wire_full`, and `fuzz` |
-| Smoke regression after tick-based chat pacing fix | `1/1 PASS` | `ctest --test-dir build -R smoke80 --output-on-failure --timeout 450`; 181.26 seconds, `225 PASS / 0 FAIL`. Reset feedback has a 30-second bounded wait and prints chat/disconnect diagnostics only on failure. Actions for prior SHA `54e7ad9` missed this feedback under the old 8-second wait after observing 120 ticks; the current PR-head Actions result is authoritative for verifying this test-only mitigation. |
+| Smoke regression after tick-based chat pacing fix | `PASS LOCALLY; CI HARDENING IN FOLLOW-UP` | `ctest --test-dir build -R smoke80 --output-on-failure --timeout 600`; last recorded local run was 181.26 seconds, `225 PASS / 0 FAIL`. Reset feedback has a 30-second bounded wait and prints chat/disconnect diagnostics only on failure. A later Actions run exposed that the target-specific `TIMEOUT 450` overrode the workflow-wide `--timeout 600`; the follow-up removes that override and lowers the random-tick stress value while keeping it above the vanilla default. |
 | PR Actions gate | `PASS / SHA-SPECIFIC` | Combined head `bac90dfbce70a8d1c09ddf525b339fac479ff350` passed the full build and non-nightly CTest ([run 35828644414](https://github.com/nico2525nn/cpp-fabricmc/actions/runs/35828644414)); subsequent review changes in this branch still require a new exact-head run. |
 
 MISSING #71's row-specific evidence is now `PASS` in the 90-row coverage ledger
@@ -424,7 +424,9 @@ baseline and protected manifest, a 90-row coverage ledger (`14 PASS`, `38 PARTIA
 with focused regression tests. The PR #1 review completed five hostile rounds
 and a final follow-up;
 the current verdict is `P0=0, P1=0, P2=0, P3=0` unresolved after the focused
-fixes and 8/8 local CTest rerun. Exact-head CI is still pending. Source-order
+fixes and 8/8 local CTest rerun. The exact PR-head Actions run
+[`35865145988`](https://github.com/nico2525nn/cpp-fabricmc/actions/runs/35865145988)
+passed before PR #1 was merged. Source-order
 guards cover security fixes where no authenticated client or external JVM fixture
 is available. Scoped production cleanup measured `-9`
 lines; the additional refactors are recorded in `goal-cleanup-runtime.md` and
@@ -439,3 +441,52 @@ remaining-entry fixture passed `1/1` in `63.93s`, and the final two-launch
 protocol matrix passed with both owned processes returning `0` and no escalation.
 The requested 7200-second soak failed at `t=1200s` with server exit `-9`; no
 accepted long-soak result is claimed.
+
+The first post-merge main Actions run
+[`35868844643`](https://github.com/nico2525nn/cpp-fabricmc/actions/runs/35868844643)
+timed out `smoke80`; a same-SHA rerun completed the suite but exposed one
+`goal_live_remaining` failure. That fixture shrank the world border while the
+player was still far outside it, then issued a teleport; on a slower runner,
+border damage could remove the command target before the teleport feedback was
+observed. Locally, `smoke80` passed all 225 cases in `178.25s`, and the full
+pre-fix non-nightly CTest run passed `55/55` in `522.97s`. The follow-up makes
+the fixture teleport to its damage position before shrinking the border, then
+checks the same border wire and damage effects; the targeted post-change test
+passed `5/5` consecutive launches (`263.24s`).
+
+The PR #3 Actions run
+[`35877465112`](https://github.com/nico2525nn/cpp-fabricmc/actions/runs/35877465112)
+then exposed a separate `smoke80` issue: CTest reported `***Timeout 450.01 sec`
+even though the workflow supplies `--timeout 600`, because the test had its own
+`TIMEOUT 450` property. Its captured output reached the world-management cases,
+then the block-tick scenario using `randomTickSpeed 100`; subsequent disconnect
+and assertion failures were cascading output after the timeout. The follow-up
+removes the target-specific timeout so the shared CTest limit applies and uses
+`randomTickSpeed 10` (still above vanilla's default 3) with the existing 16×16
+crop field. This keeps the real random-tick behavior check while avoiding a
+whole-simulation stress setting in the ordinary regression suite.
+
+The exact-head follow-up Actions run
+[`35883033964`](https://github.com/nico2525nn/cpp-fabricmc/actions/runs/35883033964)
+then completed the expanded CTest suite without the smoke timeout; one separate
+`plan43` assertion (`W-02 m2/h0/s1 window opens`) failed, leaving the other 54
+tests passing. Investigation found that `summonHorseNear` accepted any horse
+spawn observed after its snapshot, including an older horse from a delayed
+initial entity stream. The test now snapshots the highest observed entity ID
+and accepts only a horse with a larger ID. The revised `plan43` test passed five
+consecutive local launches (`130.37s` total); the new PR head still requires its
+own GitHub Actions run before merge.
+
+The following exact-head Actions run
+[`35886973563`](https://github.com/nico2525nn/cpp-fabricmc/actions/runs/35886973563)
+passed configure, build, static checks, and deterministic focused gates, but the
+full CTest regression failed only in `plan43` (`W-02 atk0 lands (HurtAnimation
+eid)`); the other 54 tests passed. This attack-only miss is distinct from the
+earlier window assertion. The W-02 fixture previously began interaction as soon
+as `join()` observed Join Game, before the initial chunk batch marked client
+startup complete. It now waits for `ChunkBatchFinished` before selecting a
+summoned horse and issuing interaction/attack packets, and exits that case
+cleanly if setup fails. This is test-readiness hardening, not a production combat
+change or proof of the failure's server-side cause. The revised `plan43` passed
+10 consecutive local launches (`260.95s` total); the new PR head still requires
+its own Actions run before merge.

docs/VERIFICATION.md

@@ -658,7 +658,7 @@ timeout --foreground --kill-after=5 120 python3 tests/bot_smoke.py --binary ./bu
 timeout --foreground --kill-after=5 400 python3 tools/soak_bot.py --duration 300 --binary ./build/cppfm
 timeout --foreground --kill-after=5 1200 ctest --test-dir build -LE 'nightly|package' --output-on-failure --timeout 600
 timeout --foreground --kill-after=5 300 ctest --test-dir build -R 'native|scoreboard_reset|spec_wire|plan43|flood_net|fuzz|wire_full|gameplay_full|seed_parity|block_hardness_full|redstone_engine_full|mob_stats_full|mining_full|quality_audit|tautology_lint|mcproto_framing|bench|multi_client|bot_smoke|recipes_mirror|recovery|rcon_multi' --output-on-failure --timeout 120
-timeout --foreground --kill-after=5 600 ctest --test-dir build -R smoke80 --output-on-failure --timeout 450
+timeout --foreground --kill-after=5 700 ctest --test-dir build -R smoke80 --output-on-failure --timeout 600
 ```
 
 For cleanup, first inspect and then use exact PIDs; do not paste an unreviewed broad

tests/test_goal_live_remaining.py

@@ -273,8 +273,12 @@ def run(binary: Path, artifact_root: Path) -> dict[str, object]:
         client.command("setblock 63 -60 50 minecraft:air", "Changed the block")
         border_before = len(client.packet_payloads)
         client.command("worldborder center 0 0", "Set world border center to 0.000000, 0.000000")
-        client.command("worldborder set 2", "Set world border to 2.000000 blocks wide")
+        # Move to the eventual damage position while the default border is
+        # still large. Shrinking it first leaves the player far outside the
+        # new border and can kill them before the following teleport command
+        # is processed on a slower runner.
         client.command("tp @s 10 -60 50", "Teleported 1 entity")
+        client.command("worldborder set 2", "Set world border to 2.000000 blocks wide")
         border_packets = client.packet_payloads[border_before:]
         border_packets += client.pump(2.2)
         border_health = [parse_health(payload) for packet_id, payload in border_packets

tests/test_plan43.cpp

@@ -18,7 +18,6 @@
 #include <chrono>
 #include <thread>
 #include <cmath>
-#include <unordered_set>
 
 using namespace cppfm;
 using namespace cpptest;
@@ -187,17 +186,20 @@ static std::int32_t summonHorseNear(TestClient& c, const char* who) {
     // NOTE: /summon takes NO position args (server spawns at player+(2,1,2));
     // extra tokens fail brigadier parse, so send the bare command.
     (void)who;
-    // snapshot known horse eids BEFORE summoning (a warm server answers in
-    // <ms; snapshotting after the send would swallow our own horse).
-    std::unordered_set<std::int32_t> known;
-    for (auto& s : c.spawns()) if (s.type == 63) known.insert(s.eid);
+    // Snapshot the entity-id high-water mark before summoning. A newly joined
+    // client's initial entity stream can still contain older horses after
+    // join() returns; treating any post-snapshot horse packet as ours makes
+    // this test race against that tail of the stream.
+    std::int32_t highestSeenEntityId = 0;
+    for (const auto& s : c.spawns())
+        if (s.eid > highestSeenEntityId) highestSeenEntityId = s.eid;
     c.sendChatCommand("summon minecraft:horse");
     auto deadline = std::chrono::steady_clock::now() + std::chrono::milliseconds(6000);
     while (std::chrono::steady_clock::now() < deadline) {
         auto ss = c.spawns();
         std::int32_t found = -1;
         for (auto& s : ss)
-            if (s.type == 63 && !known.count(s.eid)) found = s.eid; // newest birth wins
+            if (s.type == 63 && s.eid > highestSeenEntityId) found = s.eid;
         if (found >= 0) return found;
         std::this_thread::sleep_for(std::chrono::milliseconds(100));
     }
@@ -248,7 +250,13 @@ static void tUseEntity(ServerProc& srv) {
         char msg[160];
         snprintf(msg, sizeof(msg), "W-02 %s join", k.label);
         char nm[32]; snprintf(nm, sizeof(nm), "P43U%d", (int)(k.mouse * 4 + k.hand * 2 + (k.sneak ? 1 : 0)));
-        CHECK(c.connect("127.0.0.1", srv.port) && c.join(nm), msg);
+        const bool joined = c.connect("127.0.0.1", srv.port) && c.join(nm);
+        CHECK(joined, msg);
+        if (!joined) { c.close(); continue; }
+        const bool chunksReady = waitInitialChunks(c);
+        snprintf(msg, sizeof(msg), "W-02 %s initial chunk batch complete", k.label);
+        CHECK(chunksReady, msg);
+        if (!chunksReady) { c.close(); continue; }
         std::int32_t eid = summonHorseNear(c, k.label);
         snprintf(msg, sizeof(msg), "W-02 %s horse summoned", k.label);
         CHECK(eid >= 0, msg);
@@ -267,9 +275,15 @@ static void tUseEntity(ServerProc& srv) {
     for (int s = 0; s < 2; ++s) {
         TestClient c;
         char nm[32]; snprintf(nm, sizeof(nm), "P43A%d", s);
-        CHECK(c.connect("127.0.0.1", srv.port) && c.join(nm), "W-02 atk join");
-        std::int32_t eid = summonHorseNear(c, s ? "atk1" : "atk0");
+        const bool joined = c.connect("127.0.0.1", srv.port) && c.join(nm);
+        CHECK(joined, "W-02 atk join");
+        if (!joined) { c.close(); continue; }
+        const bool chunksReady = waitInitialChunks(c);
         char msg[160];
+        snprintf(msg, sizeof(msg), "W-02 atk%d initial chunk batch complete", s);
+        CHECK(chunksReady, msg);
+        if (!chunksReady) { c.close(); continue; }
+        std::int32_t eid = summonHorseNear(c, s ? "atk1" : "atk0");
         snprintf(msg, sizeof(msg), "W-02 atk%d horse summoned", s);
         CHECK(eid >= 0, msg);
         if (eid >= 0) {

tests/test_smoke_80.cpp

@@ -270,8 +270,8 @@ static void testBlockBehaviors(ServerProc& srv){
     CHECK(waitBlockUpdate(c,4,-59,0,4333,2000),
           "wheat field age 0 is installed before the crop tick test");
     c.clearChatLines();
-    c.sendChatCommand("gamerule randomTickSpeed 100");
-    CHECK(waitChat(c,"100",2000), "randomTickSpeed 100 command accepted");
+    c.sendChatCommand("gamerule randomTickSpeed 10");
+    CHECK(waitChat(c,"10",2000), "randomTickSpeed 10 command accepted");
     const auto cropDeadline = std::chrono::steady_clock::now() +
                               std::chrono::milliseconds(6000);
     bool grew = false;
@@ -286,9 +286,11 @@ static void testBlockBehaviors(ServerProc& srv){
         }
     }
     CHECK(grew,"wheat random tick with high randomTickSpeed changes the crop state");
-    // The server's vanilla chat-spam budget decays in simulation ticks, not
-    // wall-clock time. 100 random ticks/section can delay the tick thread, so
-    // observe six time-sync packets (120 ticks) before resetting the rule.
+    // Random ticks apply to every simulated section, not only this field. Keep
+    // the elevated value modest (10 vs. vanilla's default 3) so this gameplay
+    // assertion does not become a world-wide CPU stress test on CI. The vanilla
+    // chat-spam budget decays in simulation ticks, so observe 120 ticks before
+    // resetting the rule.
     const int resetAfterTicks = c.counters().timeUpdates + 6;
     CHECK(waitForTimeUpdates(c, resetAfterTicks, 30000),
           "server advances 120 ticks before gamerule reset");