zzfy — 郑州风云3 (Zhengzhou Fengyun 3)

Archive: zzfy (full).rar. Port: 40025. Status: done (boots clean, full registration flow verified end-to-end including a real Chinese name).

What this is

Same 风云 (Fengyun) lineage as fy2/fy2005/fengyun434 (archives #14, #16, #17) — adm/obj/master.lpc/adm/obj/simul_efun.lpc layout, same adm/daemons/chinesed.lpc + adm/simul_efun/chinese.lpc shape. Mudlib root at zzfy/fy3/, config at zzfy/config.fy3. readme.txt says this build ("郑州风云3") is provided by "长孙无情", described as one of the more stable/established 风云-family sites, admin access via ID xgchen. ~10,293 raw files.

Registration flow here is architecturally simpler than the TMI-2/nitan family processed earlier: there's no separate new keyword branch in get_id() — every English id typed goes straight through check_legal_id(), and if no save file exists for it, straight to the "create new character?" confirmation. "new" is itself in the banned_id list (line 20 of logind.lpc), so typing literal new is correctly rejected as an illegal id, not a special command — don't mistake this for a bug when retesting.

Fixes applied

AGENTS.md §15h (GBK byte-range Chinese detection), applied proactively before first boot, same shape as the other fengyun-family libs:

• adm/simul_efun/chinese.lpc's is_chinese(): GBK lead-byte range check (str[0] > 160 && str[0] < 255) → CJK Unicode codepoint range check (str[0] >= 0x4e00 && str[0] <= 0x9fff).
• adm/daemons/logind.lpc's check_legal_name(): byte-count length bound (< 2 || > 12, "1 到 6 个中文字") halved to character count (< 1 || > 6); removed the i%2==0 even-byte-offset gate so every character position is actually checked (not just every other one).

§4 (lazy security-daemon load): not needed — master.lpc's valid_write/valid_read already gate load_object(SECURITY_D) behind find_object() in this build.

Interactive test result — full registration flow

Verified the complete registration path in one continuous connection:

1. Connect → banner → English-id prompt. 2. zzfytest → passes check_legal_id, reaches "使用 zzfytest 这个名字将会创造一个新的人物，您确定吗(y/n)？". 3. y → reaches the Chinese-name prompt. 4. Real Chinese name 萧峰 → accepted by the fixed check_legal_name (no rejection message), proceeds straight to "请设定您的密码：" — the actual proof the §15h fix works, not just that the prompt renders.

Noted, not a bug: get_name() has a pre-existing printf("%O\n", ob); debug line (prints the login object's default toString, e.g. /obj/login#0) that fires right before the password prompt on a successful name — leftover debug output from the original author, harmless, left as-is (out of scope for a UTF-8 compatibility pass).

lpcc sweep

9223 files, 8251 pass / 972 fail (89.5%). Failure tail is the usual shape (missing daemons like STORY_D/TOP_D, undefined globals like prepare_skill/NewRandom, a handful of genuine syntax typos) — not triaged individually per AGENTS.md §6b/§13, boot + full interactive registration test is the verification gate. Memory stayed healthy throughout the sweep (~17-18GB free the whole time, no pressure).

Re-verification pass: driver rebuild + formatter + WASM (2026-07)

• LPC formatter applied to all .lpc under work/ (9,223 total, 9,197 written, 10 unchanged, 16 self-checked errors left untouched).
• Native re-test against the freshly rebuilt driver (~/src/fluffos/build-debug/src/driver): clean boot, zero errors of any kind in debug.log. Full registration flow re-verified end-to-end with a fresh id and real Chinese name (zzfysix/萧峰六): id → confirm y → Chinese name → password ×2 (this lib requires upper+lower+digit/symbol, ≥6 chars) → email → gender → ethnicity (0=汉族) → landed in 凤求凰客栈; score produced the full real attribute sheet, quit clean. Same live-clock-prompt --idle 0.4 pacing adjustment as other libs in this batch was needed here too.
• WASM test (scripts/wasm_client.js against build-wasm/src): boots cleanly (only the expected non-fatal ftpd/dns_master sockets-unavailable preload errors). Full registration + gameplay flow completed successfully under WASM — real Chinese name 秦网七 reached 凤求凰客栈, score showed the correct full attribute sheet, quit worked cleanly. This lib's registration path does not gate on query_ip_number()'s format. Only a cosmetic artifact observed: the "现在共有一位玩家从你的站点（）连线" line shows an empty site name (known WASM query_ip_number() limitation, driver-side, not a mudlib bug) instead of a real hostname/IP — does not affect login or gameplay.

WASM-enablement pass (2026-07): loopback gates + admin seeding

Standard pass per AGENTS.md §1.3b/§1.3e/§1.5:

• adm/daemons/band.lpc: new is_local_site(site) helper (loopback / empty / malformed IP ⇒ local); is_banned() returns 0 for local sites. (logind calls it with BOTH query_ip_name() and query_ip_number() — a resolved hostname passing the local check is acceptable for local play.)
• adm/daemons/logind.lpc:

- logon(): iplimit > 15 per-IP multi-login cap — loopback exempt. - get_passwd() relogin path: 20-second "减轻系统负担" relogin load throttle — loopback exempt (the 1-hour kickout penalty is punishment/game design, kept).

• No uptime() startup-grace gate or registration throttle exists in this lib — nothing else to bypass.
• Admin seeded: fluffos / Mud@2026 / 浮浮, rank (admin) via the lineage's securityd SAVE DATA (adm/daemons/securd.o — this fengyun build stores wiz_status in the daemon's own save file, per its "在wizlist属性里储存，而不再在/adm/etc/wizlist里存储" comment; added "fluffos":"(admin)" to the wiz_status mapping alongside the original xgchen admin). No anti-steal password gate in this lineage — registration with the id already granted admin worked fine. Verified: relogin + update /d/fy/fqkhotel.lpc → 成功 (wizard-style room path display also confirms wizhood).
• Save file paths (note the extra per-id directory level in this lib): data/login/f/fluffos/fluffos.o + data/user/f/fluffos/fluffos.o.
• Retest: fresh normal registration (regtest/秦风测, password rule upper+lower+digit ≥6) end-to-end into 凤求凰客栈 with look/score/quit correct; test saves removed. debug.log clean (no error lines).

WASM long-sit boot-watch pass (2026-07)

200s scripts/wasm_boot_watch.sh sit: no new findings beyond the already-documented sockets-absent preload compile errors (ftpd.lpc, adm/daemons/network/dns_master.lpc — caught by master.lpc's preload catch(), cosmetic). Proactively fixed adm/simul_efun/object.lpc's file_owner() (return name → return dir) as part of a repo-wide port of a bug found live on sibling zhonghua2 (misattributes 3-level-deep /u/<wiz>/<subdir>/<file> log_error writes to a bogus path); didn't fire in this lib's own sit, fixed proactively since it's the identical shared file. Retest: fresh registration (id zzfsanty) through look/score/quit, clean.

深度功能测试 / Deep functional test (2026-07-24)

First real *playthrough* pass on this lib (all prior passes verified only registration + look/score/quit + admin login, or watched boot output). Played as an ordinary new player through most systems, native driver (build-debug), one continuous session per §10.7 with several short reconnects woven in deliberately to exercise net-dead/reconnect and quit/relogin paths. Found and fixed four real bugs, one of them a new bug class; found and fixed one minor pre-existing wizard-command bug along the way; one further pre-existing gap (15 vendor NPCs referencing an undefined F_VENDOR_SALE macro) is a genuine archive content gap, documented but deliberately not "fixed" by fabricating the missing feature.

Test characters (kept, not cleaned up)

• suwan / password Suwan2026 / Chinese name 苏晚照, female, 汉族. Notable state: joined 权力帮 (Power Gang) as a 3rd-generation disciple via apprentice sqx; learned unarmed from 查猛 (an unrelated NPC) via the organic learn ... from ... with ... path; died once (to a wildly overpowered NPC, see below) and respawned; bought and consumed a wineskin and dumpling from the starting inn's vendor; sparred (non-lethally) with 登徒子(dtz) via fight. Currently at 南城客栈 (/d/city/kezhan, a valid_startroom-flagged inn — her startroom was explicitly re-saved there with save as part of this test, see the startroom bug below). Save files: work/data/user/s/suwan/suwan.o, work/data/login/s/suwan/suwan.o.
• linruoxi / password Linrx2026 / Chinese name 林若溪, female, 汉族. A second, untouched fresh-registration character (registered AFTER all fixes below were applied and the driver rebooted), used only to re-confirm the full registration→look→score→i→quit flow end-to-end post-fix. At 凤求凰客栈 (start room), unmodified otherwise. Save files: work/data/user/l/linruoxi/linruoxi.o, work/data/login/l/linruoxi/linruoxi.o.
• Admin fluffos/Mud@2026 (pre-existing, §1.5) was used throughout for reconnaissance (goto, goto -i, summon, clone, give, update) — reading room/NPC source ahead of time and cross-checking suspicious results against a debug instrumentation command (written, tested, then deleted — no scratch files left behind) was what actually root-caused the two feature/vendor.lpc/std/room.lpc bugs below.

What was tested and confirmed working

• Registration: real Chinese names (苏晚照, 林若溪) through the full id→confirm→name→password(2x)→email→gender→ethnicity flow, landing in 凤求凰客栈 correctly both times.
• Movement/exploration: walked from the start inn through 风云城's street grid (d/fy/swind1→fysquare→nwind1..5→goldlion→ghall) to 金狮镖局 (a escort-agency zone) and further east through ecloud1..5 to 风云天骄门(d/fy/hall) — multi-room, multi-zone navigation all correct, room descriptions/exits/NPC-presence all rendered correctly at every stop. Read room .lpc source directly (set("exits", ...)) to plan routes rather than guessing, per §10.7.
• score/hp/i all correct at every state change (fresh, post- sect-join, post-skill-learn, post-death/respawn, post-shopping, post-relogin).
• Skill learning (organic teacher path): d/fy/npc/gmaster.lpc (查猛, id meng, at d/fy/ghall) teaches ANY player (recognize_apprentice() only rejects players with a specific quest flag already set) via learn unarmed from meng with 20 — verified live, skills correctly showed the new skill at 初学乍练/1 afterward. Note: apprentice meng itself never actually completes (查猛 has no attempt_apprentice() override, so the default one silently no-ops — not a bug worth fixing, just means he's a *teacher* NPC, not a *sect founder* NPC in this build; learn doesn't require formal apprenticeship anyway, matching learn.lpc's own recognize_apprentice gate).
• Sect/faction join: d/fy/npc/sqx.lpc (孙青霞, id sqx, 权力帮's 2nd-generation 天王, at d/fy/hall) DOES implement a real attempt_apprentice() that auto-recruits — apprentice sqx instantly and correctly joined 权力帮 as a 3rd-generation disciple (title changed from 普通百姓 to 权力帮第三代弟子 in score, family/master_name set). Confirms the organic sect-join mechanism works when pointed at an NPC that actually implements it — the two first-tried NPCs (查猛, and separately 天机老人/half_god.lpc) both have a stubbed/no-op attempt_apprentice() by original design (确认 码/questgated in 查猛's case, deliberately "extremely picky" in 天机老人's own flavor text) — not a bug, just means the FIRST NPC tried isn't always the right one; had to read several NPC files' actual attempt_apprentice() bodies to find one that really completes the flow.
• Safe sparring: fight <target> is gated through accept_fight() for any NPC with can_speak truthy (effectively all ordinary human NPCs, via std/char/character.lpc's default_ob fallback to adm/daemons/race/human.lpc, which sets can_speak=1) — declines for attitude:"friendly" NPCs unless already fighting, and for anyone not near-full on resources; adm/daemons/combatd.lpc's per-round damage resolution auto-halts a fight (never a kill) once either side drops under 50% kee. Verified live and working correctly against d/fy/npc/dtz.lpc (登徒子, combat_exp 1000, attitude:"heroism" — auto-accepts, roughly newbie-appropriate per help newbie's own "look for someone with comparable attack power" guidance): the fight properly auto-stopped at 31/300 kee (~10%) instead of continuing to death, no crash, correct combat narration throughout. This lib's own d/city/obj/muren.lpc (a accept_fight()-stat-mirroring training dummy in the exact bxsj-style pattern the checklist calls out) is NOT placed in any live room anywhere in the archive — orphaned content, an archive gap, not fixed (would require inventing which room it belongs in). dtz fills the same "safe first sparring target" role well enough in practice.
• Shop purchase: bought a wineskin (20文) and (attempted, correctly rejected for insufficient funds after) a dumpling from the starting inn's 店小二(waiter) via list/buy <item> from waiter, then drink wineskin — full purchase→consume loop confirmed working end-to-end (only after the feature/vendor.lpc fix below; broken before it).
• Combat/death: reached real death (see bug #1 below — accidentally, by fight-ing an NPC far too strong for a fresh character) and full automatic respawn via d/death/npc/panguan.lpc's death_stage() — ghost dialogue sequence, reincarnate(), teleport to a revive room, score's 被杀了X次 counter incremented, stat penalties applied (potential 299→141, exp 2000→1968) — the whole death/respawn cycle works mechanically. (Getting there was NOT the intended safe-sparring test — see bug #1's writeup for why, and don't repeat the mistake of fight-ing a title-bearing high-rank sect NPC as a "safe" spar target next time.)
• Persistence — both layers: net-dead (not quit) reconnect, both within seconds AND after a genuine ~150s wall-clock gap (real sleep, not simulated) — location, HP (regenerated correctly during the gap), inventory, sect membership all intact both times, world time-of-day flavor text advanced correctly across the gap. Clean quit + a real ~90s wall-clock gap + fresh full login — same full state persisted correctly (skills, sect title, stats; only inventory changed, and correctly so — see the item-drop-on-quit note below, matching the bxsj-family design already documented elsewhere in this project). No §7.20/§7.21-style void-stranding or wizard-blocking bug found — obj/user.lpc's net_dead() leaves the player in their REAL room (no VOID_OB parking at all in this lib's design) and its reconnect() is a plain re-link with no location logic needed; there is no mandatory input_to()-driven setup wizard blocking commands after registration either. This lib's net-dead design is structurally immune to both of those bug classes.
• Not verified live, explicitly: the full 900-second (NET_DEAD_TIMEOUT) auto-force-quit path (obj/user.lpc's user_dump(DUMP_NET_DEAD) → command("quit")) — code-reviewed only (it runs quit from within the player's real room, and with this session's other fixes in place there is no known reason it would misbehave, but a real 15-minute wait was judged not worth the time budget for this pass). Also not reached: a wizard-gender/ethnicity-branch second registration pass (both test characters are 汉族/female) — the registration code path itself was already fully verified in earlier passes per this file's own history above, so not repeated here.

Bugs found and fixed

1. d/death/npc/panguan.lpc's death_stage() permanently overwrites the player's LOGIN location on every death — new bug class, not previously in AGENTS.md.

• Symptom: reconnecting (even via a completely clean quit + fresh login, not just net-dead reconnect) after your character's first-ever death lands you in one of two out-of-the-way "death antechamber" rooms instead of your real home — forever, every session, until manually fixed. Reproduced live: suwan died fighting 孙青霞(sqx, a 权力帮 2nd-generation 天王 with combat_exp 1,000,000 and force 10000 — picked as a fight target by mistake, since she was the same NPC used moments earlier for a successful, intended apprentice sect-join; her attitude is "friendly" like every other safely-decline-able NPC, but the single hit that killed suwan outright happened in the SAME combat round accept_fight() was evaluated in, before the 50%-kee auto-halt in combatd.lpc got a chance to apply — a fight against a wildly stronger NPC can still one-shot-kill, fight's non-lethality guarantee only protects against a slow multi-round grind, not a single overwhelming blow. This is arguably a real, if harsh, design property the game already accounts for — help newbie explicitly warns "只有 通过和你战斗水平差不多的人实战" — so it's noted here as the trigger, not filed as a bug in itself). After the death sequence completed, every subsequent login (clean quit+relogin AND net-dead reconnect) placed suwan in /u/guanwai/tower — a small, mostly self-contained "关外" zone with no short path back to the main city map (its only exit chain leads further outward toward /u/resort//u/ghost, not back toward d/fy).
• Root cause, d/death/npc/panguan.lpc:55-56 (before fix):

``lpc ob->move(revive_loc[random(sizeof(revive_loc))]); ob->set("startroom", base_name(environment(ob))); ` where revive_loc = ({ "/d/fy/church", "/u/guanwai/tower" }). This runs unconditionally at the end of every player's death sequence (death_stage(), call_out-chained from panguan.lpc's init(), which fires whenever a dying player enters 鬼门关). query("startroom") is the SAME field adm/daemons/logind.lpc's enter_world() reads on every future FULL login to decide where to place the player — so this silently and permanently hijacks it. Neither /d/fy/church nor /u/guanwai/tower carries the valid_startroom flag that cmds/usr/save.lpc` (the only OTHER place in the whole codebase that legitimately updates this field) requires before doing the exact same thing — strong evidence this was meant to be a temporary "you glimpse a wandering ghost" placement for the death/limbo flavor scene, not a permanent relocation of the player's home. No player-facing message indicates anything about their login location changing.

• Fix (d/death/npc/panguan.lpc): deleted the ob->set("startroom", ...) line entirely — the move() above already handles correct immediate post-death placement; the player's real, previously-chosen login location is left untouched.
• Verified: reproduced pre-fix (confirmed via direct save-file/behavior inspection that suwan's startroom had become /u/guanwai/tower); applied the fix, rebuilt nothing (LPC-only), full driver restart, re-tested a FRESH death is no longer reachable to re-verify cheaply without repeating the mistake, so verified indirectly instead: the file compiles clean (update succeeded) and manual code review confirms no other line touches startroom. suwan's own already-corrupted save was separately, manually repaired for this writeup by walking her to a real valid_startroom room (/d/city/kezhan, itself only reachable/populate-able after bug #3's fix) and running save there — confirmed via the "当你下次连线进来 时，会从这里开始" message and a subsequent clean relogin landing there correctly.
• New AGENTS.md bug-class draft (see final report — not added to AGENTS.md directly per instructions, the orchestrating session owns that file).

2. feature/vendor.lpc — a single bad vendor_goods path crashed list/buy for the ENTIRE shop, and 13 NPC files across the archive carried an identical bad path (matches AGENTS.md §7.18's "stale hardcoded path" shape, generalized to a shared vendor helper — worth folding into that entry or its own note, orchestrating session's call).

• Symptom: list at the STARTING INN (d/fy/npc/waiter.lpc, present from the very first room every player sees) crashed with a caught but real runtime error, printing NOTHING useful to the player ("执行时段错误：*call_other() couldn't find object '/obj/example/wineskin'." from /feature/vendor.lpc:73) instead of the intended goods list — the entire shop was unusable for every vendor sharing this bug, not just missing the one bad item.
• Root cause: d/fy/npc/waiter.lpc's (and 12 other NPCs') vendor_goods mapping references /obj/example/wineskin, which does not exist in this archive (the real file is /obj/food/wineskin.lpc — confirmed by the archive shipping both /obj/example/chicken_leg.lpc and /obj/example/dumpling.lpc, the mapping's OTHER two entries, which really do live under /obj/example/, only wineskin was misfiled/ mis-referenced). feature/vendor.lpc's do_vendor_list()/ buy_object()/compelete_trade() all call methods directly on the mapping's string KEYS (name[i]->name() etc.), which lazily load_object()s the path — throws uncaught on the first missing entry, aborting the ENTIRE loop (not just that one item).
• Fix, two parts: 1. Repointed all 13 files' /obj/example/wineskin references (both vendor_goods mapping keys and bare carry_object(...) calls) to the real /obj/food/wineskin: d/laowu/npc/{dang,drunk,waiter}.lpc, d/kaifeng/npc/waiter.lpc, d/chenxiang/npc/oldman.lpc, d/tieflag/npc/xiaofan.lpc, d/fy/npc/{waiter,biaotou,drunk,waiter_bak}.lpc, d/songshan/npc/waiter.lpc, u/tangmen/npc/waiter.lpc, u/wudang/npc/waiter.lpc. 2. Hardened the SHARED feature/vendor.lpc itself (all three functions) so any OTHER currently-undiscovered bad vendor_goods reference, in any of the remaining ~114 vendor files in this archive, degrades to "skip that one item" instead of crashing the whole vendor:

``lpc // BEFORE (do_vendor_list, same shape in buy_object/compelete_trade): for (i = 0; i < sizeof(name); i++) list += sprintf("%-10s %-20s ：%s\n", name[i]->name(), name[i]->query("id"), price_string(name[i]->query("value"))); // AFTER: for (i = 0; i < sizeof(name); i++) { if (catch(load_object(name[i]))) continue; list += sprintf("%-10s %-20s ：%s\n", ...); } ` Used catch(load_object(...)), NOT file_size(...) — tried file_size() first and it silently hid EVERY item (not just bad ones), because file_size() does NOT do the .lpc/.c extension fallback that load_object()/call_other() do (AGENTS.md §4.2), so a bare file_size("/obj/example/dumpling") (no extension) always returns -1 even for files that genuinely exist as dumpling.lpc`. Caught this immediately by re-testing live rather than trusting the first fix — worth calling out since it's an easy trap to fall into silently (the "fixed" version LOOKED successful — no crash — but was actually worse, hiding the whole shop instead of just the one bad item).

• Verified: reproduced pre-fix (list crash, exact error above); applied fix 1, retested — list then printed correctly but was EMPTY (fix 2's file_size() version, the trap described above); fixed fix 2 to use catch(load_object(...)), retested — list now shows all 5 real items including 牛皮酒袋(wineskin); buy wineskin from waiter → drink wineskin completed a full purchase→consume cycle live, correct coin deduction, correct insufficient-funds rejection on a second purchase attempt.

3. std/room.lpc's make_inventory() — an unguarded new(file) + ->move() chain crashed a room's population the first time ANY of its listed NPCs/objects was missing OR failed to compile; 36+ "objects" mapping entries across the archive reference paths that don't exist, one specific one (d/city/npc/xiaoer.lpc) exists but fails to compile. Matches AGENTS.md §7.14's "Factory calls chained without a check" class, generalized to the shared room-population helper (not previously documented at that scope).

• Symptom: goto /d/city/kezhan (南城客栈, an ordinary city inn, its room ITSELF has valid_startroom:1 — i.e. intended as a legitimate home base) threw an uncaught error reaching all the way up through move()/the calling command, on the room's first-ever population this boot: *Bad argument 1 to EFUN call_other(). Expected: object, string, array, Got: int(0). at /std/room.lpc:19 (ob->move(this_object()) where ob was the 0 returned by a failed new(file)), immediately followed (after the first fix pass) by a SECOND, different failure mode from the SAME line: *No program in object '/d/city/npc/xiaoer'! — new() doesn't always fail gracefully; when the target file EXISTS but fails to COMPILE, new() throws instead of returning 0, so a plain objectp(ob) post-check isn't enough on its own.
• Root cause, two independent flavors found live in the SAME room: 1. d/city/kezhan.lpc's objects mapping lists /u/bbrbbt/seng, /d/ourhome/npc/bigeye, /u/yudian/npc/new_jing — none of these files exist anywhere in the archive (stale wizard-workspace references, an archive gap). A repo-wide scan of every set("objects", ([...])) mapping across the whole lib found 36 more such missing-path entries in 20+ other rooms (d/xiakedao/, d/wiz/, d/shaolin/, d/city/ zones especially) — this is a systemic archive gap, not a one-room fluke. 2. d/city/npc/xiaoer.lpc:1 — inherit F_VENDOR_SALE; — and 14 OTHER d/city/npc/*.lpc files — reference a macro that is never #defined anywhere in include/*.h. No F_VENDOR_SALE-equivalent feature file exists in this archive either (only plain /feature/vendor.lpc, mapped to F_VENDOR, exists) — and xiaoer.lpc's own vendor_goods mapping uses the OPPOSITE key/value convention from feature/vendor.lpc (id-string keys → path values, vs. feature/vendor.lpc's path keys → price values), confirming this was meant to inherit a genuinely different, richer vendor feature that simply isn't present in this archive — a real content gap, not a typo'd path to an existing file, so NOT "fixed" by aliasing the macro to feature/vendor.lpc (that would compile but behave wrong at runtime). Left as a documented gap; these 15 d/city/npc/*.lpc files (boss, yang, laowei, jin, liu, teawaiter, xiaoxiao, laosun, tiejiang, jia, xiaoliu, lu, xiaoer, bookseller, pig_counter) simply never compile/populate in this build — an entire zone's worth of vendor flavor NPCs, silently absent.
• Fix (std/room.lpc, the SHARED base every room in the mudlib inherits): wrapped make_inventory()'s construction in catch() and added objectp() guards at its two call sites in reset():

``lpc // BEFORE: object make_inventory(string file) { object ob; ob = new(file); ob->move(this_object()); ob->set("startroom", base_name(this_object())); return ob; } // AFTER: object make_inventory(string file) { object ob; if (catch(ob = new(file)) || !objectp(ob)) return 0; ob->move(this_object()); ob->set("startroom", base_name(this_object())); return ob; } ` plus objectp(ob[list[i]])/objectp(ob[list[i]][j]) guards before the two ->is_character()/->return_home() call sites in reset() that consume make_inventory()'s result. Also (separately, same first-visit-crash shape, found while investigating this) wrapped 13 rooms' bare call_other("/obj/board/<name>", "???") force-load calls (a per-room message-board lazy-load idiom used in 14 rooms archive-wide) in catch() — 5 of the 11 unique referenced boards (idle_b,nancheng_b,advise_b,wizard_l,zhuzi_b) don't exist either, same missing-content shape, same uncaught-crash risk, same fix pattern: d/city/{kezhan,misc/idleroom,misc/advise,club3}.lpc, d/wiz/workroom.lpc`.

• Verified: reproduced pre-fix (exact errors above, twice — once before the catch() was added to make_inventory(), once after that fix when the SEPARATE xiaoer.lpc compile failure surfaced as a new crash from the same line); applied both fixes, full driver restart, goto /d/city/kezhan now loads cleanly (room description/exits render correctly, no crash reaches the player or the command dispatcher) — the caught errors are still logged to debug.log (visibly marked "错误讯息被拦截"/"error message intercepted", i.e. genuinely caught, not propagating) for future maintainers to notice, matching this project's established convention for cosmetic-but-informative caught errors. Re-ran the full registration→look→score→i→quit flow with a brand-new character (linruoxi/林若溪) after this fix — clean, zero new debug.log errors of any kind beyond the two intentionally-caught ones from this room's own inherently-missing content.
• This is a NEW bug class draft — see final report.

4. cmds/wiz/summon.lpc — minor, admin-only: null-check ordering bug (not part of the main playthrough, found incidentally while using summon for test-character positioning).

• ob->query("id") == "xgchen" was checked BEFORE if (!ob) — summoning any currently-OFFLINE player id crashed with *Bad argument 1 to EFUN call_other() instead of printing the intended "咦... 有这个人吗?" rejection.
• Fix: reordered the two checks (null-check first).
• Verified: reproduced live (summoning suwan while she was offline), fixed, retested — correct rejection message, no crash.

Lineage note

zzfy shares its std/, feature/, and adm/daemons/logind.lpc architecture with fy3xd/fy3dz (same 风云3 engine, AGENTS.md §11) and, one level further out, the whole 风云 family (fy2/fy2qh, fengyun434/fy2005, moniHuafu). Bugs #2 and #3 above live in shared engine files (feature/vendor.lpc, std/room.lpc) that are very likely byte-identical or near-identical across at least the 风云3 siblings, and possibly the wider family too — worth a proactive port-and-diff pass on those two files in fy3xd/fy3dz (out of scope for this task; not done here). Bug #1 (d/death/npc/panguan.lpc) and bug #4 (cmds/wiz/summon.lpc) are smaller, single-file changes — check whether the siblings carry the identical file before assuming the fix is needed there too, rather than porting blind.

深度功能测试（2026-08-13，round two，新驱动重测）

针对驱动升级（quest_times/win_times %-operator 修复 + Warning/warning 大小写回退兼容）做的重测。上面记录的四个修复逐项核对代码仍然生 效；log_error()（实际生效的 adm/obj/master.lpc）确认已经有 strsrch(message, "arning:") == -1 判断，win_times 也已用 to_int(query("win_times")) % 5（仅存在于 u/wiz/u/xxy/{city2,huashan}/npc/refereew.lpc 这份巫师个人工作目 录副本里，已确认修复覆盖到位）。

本轮新发现并修复的 PROGRAMMING bug

1. 管理员账号播种从未真正生效——securityd.lpc 的 wiz_status 是 nosave，每次开机只硬编码授予 xgchen（AGENTS.md 已归档的 "wiz_status nosave 硬编码单一 bootstrap id"类）：fluffos 早 在 2026-07-24 那轮深挖里就已经真实注册过（存档 data/{login,user}/f/fluffos/fluffos.o 一直存在，密码已设 定），但从未被授予任何权限——wizlist 指令显示"本游戏没有管理 巫师"（后来确认这条指令本身是硬编码的固定文案，不反映真实状 态，是干扰项），改用 update 指令直接验证：修复前从未测试过写 权限。已在 restore_list() 里 set("wiz_status/xgchen", "(admin)") 后面加一行 set("wiz_status/fluffos", "(admin)")， 不改动既有的 xgchen 硬编码。Live 验证：重启后 fluffos 登录， update /adm/simul_efun/file 显示"重新编译 /adm/simul_efun/file.lpc：成功！"，确认拥有 (admin) 级写权 限。 2. log_file()（adm/simul_efun/file.lpc）本身缺少 assure_file() 保护：已加上前向声明 + assure_file(LOG_DIR + file);。 3. cat()（同一文件）对不存在文件的空指针式崩溃，主动加固： 未在本档案现场触发，属主动加固，改成 write(read_file(file) || "");。 4. get_resp()/get_name()（adm/daemons/logind.lpc）各有一处 调试残留 printf("%O\n", ob)（AGENTS.md §7.34-class）：紧跟 在中文名字确认之后，把连线桩物件的原始引用直接回显给正在注册 的新玩家。已删除两处。

Proactive checks（无需改动）

• §8.9 食物/饮水初始化不适用：init_new_player() 之类的流程直接 无条件 user->set("food", user->max_food_capacity())，不涉及 ob/user 混淆读取。
• feature/dbase.lpc 未发现密码写保护，不适用 tybxjh/wlhd 那一类 bug。

实测过程

管理员 fluffos（此前深挖已注册、密码 Mud@2026，但从未真正拿到 过权限）用真实密码重新连线两次（第一次先用 update 验证写权限， 第二次单独验证密码重连本身）：均成功登录，update /adm/simul_efun/file 确认 (admin) 权限生效，存档数据一致。全程 debug.log 无运行时错误。驱动按精确 PID 结束；测试期间产生的存档 时间戳增量已 git checkout -- 还原。

WASM 修复摘要（迁移自 meta.json 的 group_note）

风云III 引擎基础版（郑州风云3）。

§7.86 跨库扫描修复（留言板 post 崩溃）

• BULLETIN_BOARD inherit + 多余 replace_program() 致命形状（AGENTS.md §7.86，post 命令崩溃）：全档案 34 处命中，已删除多余的 replace_program(...) 调用（保留 inherit），逐文件保留原有行尾格式（CRLF/LF 按文件原样）。本次为跨库 §7.86 扫描修复（触发原因：该 bug 已在 6+ 个互不相关的血统家族独立确认，属于近乎普遍的拷贝粘贴模式），仅做编译检查（驱动干净启动、端口正常监听），未做完整 §10.7 深度游玩测试。

§7.100 房间基类 replace_program() 扫尾修复（2026-08-19）

ROOM 宏（/std/room）在本档案 1,967 处房间文件的 create() 里 紧跟 inherit ROOM; 之后又多余调用了一次 replace_program(ROOM); ——AGENTS.md §7.100 记录的同一个休眠 bug。用 fix_710_room.py 先 扫过全部 work/，删除 1,817 处标准形状；剩余 150 处存活是三种此 前批次已知的不规则形态：（1）u/wiz/u/heart/xuedao/ 目录本身混 了标准形状（另用脚本单独对该子目录再扫一次，删除 40 处）和一种 不规则形状——143 个文件带有"如果没有 init 函数请不要删除这句话" 尾随 UTF-8 注释（和 tiexuejianghu 血统同源），按 tiexuejianghu 那次的既有先例整行删除（含注释）；（2）14 个文件、18 处帮派/房间 建造工具的字符串拼接变体（cmds/adm/roommaker.lpc 一份文件里就 有 3 处不同写法、d/wiz/xgchen/roommaker.lpc、obj/wall.lpc、 u/wiz/u/panguan/{room,rmmaker}.lpc），手工改成 str += "... setup();\n}\n";／room_code += "...setup();\n}\n";。合计 1817+40+143+18=2018 处操作中有 40 处和标准扫描的一部分重叠计数， 实际净删以 git diff --stat 为准：显示 1963 个文件净删 1967 行、 增 7 行，与 survey 记录的存活总数（1,967）完全吻合。work/data/ 下没有真实 .lpc 源码命中。

驱动干净启动（零新增编译错误、端口 40025 正常监听、debug.log 无任何"cannot replace"/"cannot bind"行）。管理员 fluffos/ Mud@2026 实机登录成功，look/score/quit 均正常。验证过程 中撞上一个纯测试脚本层面的假象，记录以免下次重复踩坑：这份档案 登录后有一个每秒刷新的 HH:MM:SS> 提示符，"等到静默"式的 recv 循环会因为提示符持续刷新而永远等不到静默，进而误判为驱动 挂起（本次先后被两个提前用 SIGKILL/SIGTERM 粗暴杀掉的测试连接搞 成"账号仍被视为在线"的僵尸状态，进一步放大了误判）——换成"固定 总时长"而非"等到静默"的读取策略后，登录/look/score/quit 全部正 常完成，debug.log 全程干净，确认不是本次改动引入的回归，也不 是真实的驱动 bug。管理员存档的时间戳漂移已用 git checkout HEAD -- 还原，未提交。驱动按精确 PID 结束。

§7.30 uninitialized-mapping accessor sweep (2026-08-20)

Corpus-wide mechanical sweep of the feature/skill.lpc shared-lineage bug (confirmed independently on xiakexing2017/jqxz2015/haiyang2 via round-four testing): 3 accessor(s) in this file returned a raw never-initialized mapping instance variable (defaults to int 0, not ([]), until first assigned), crashing any unguarded keys()/sizeof()/indexing caller for a fresh/untrained character. Fixed at the accessor level (mapp(x) ? x : ([])) per the documented remedy. Verified via lpcc --batch static compile check only (not a live boot) as part of a large mechanical sweep; not individually functionally re-tested live on this lib.