zhongjidiyu — 终极地狱之轩辕传说 (Ultimate Hell: The Legend of Xuanyuan)

Archive: 终极地狱.rar (archive #78 of ~100). Port: 40072. Status: done (boots clean, full registration flow verified with two real Chinese names in separate continuous connections, post-login commands look/score/quit all confirmed producing correct real output).

What this is

Self-extracts to a nested revive/ root (not top-level) containing a raw readme (说明文件.txt, GBK-encoded, filename itself needed a manual re-encode pass after extraction — see "Extraction gotcha" below) that reads (translated): "This is the bottom-layer source code released by doing... released for study purposes, not for running a live server with." Live banner reads "终极地狱之轩辕传说" ("Ultimate Hell: The Legend of Xuanyuan"), MUD_NAME in the raw config is hell.local. adm/single/master.c's own header credits the full lineage: for ES II mudlib / original from Lil / rewritten by Annihilator (11/07/94) / modified by Xiang for XKX (12/15/95) / updated by Doing Lu for hell (2K) — this is the same "ES II" lineage already seen across ~20 prior libs (es1_win/esI/xkx2001/rzrmud/xo/ bmxkx2001/kxkj/yueyingqiyuan/wuhanzhan/haiyang2/ yanhuangwuhun/yhyxs/xuanjianlu/shenzhou/shenmo/ kxkj1/etc.), specifically closest in file layout (adm/single/{master,simul_efun}.c, adm/daemons/, adm/simul_efun/) to xkx2001(#25)/rzrmud(#23)/bmxkx2001(#45). securityd.c's own header credits "[email protected]" then "Modified by Xiang for XKX" then "Updated by DOing Lu for ULTRA HELL" — same chain. MudOS v22pre11-era target.

IMPORTANT — cross-check flagged for the main session: two OTHER archives sharing the "终极地狱" name are being processed concurrently by sibling agents in this same batch: archive #79 (终极地狱之爱若幽兰1.166正式版.rar) and archive #80 (终极地狱-指间mud版服务端.rar). This agent could not cross-check against them (concurrent extraction). The main session should diff/md5sum core files (master.c/chinese.c/securityd.c/logind.c) across all three once all are done to determine whether they're the same game at different snapshots, forks of one lineage, or unrelated codebases that happen to share a title — per this project's repeated lesson (shujian2008 vs sjpl2, xianlvqiyuan vs xlqy_new2007, etc.) that similar Chinese titles are NOT a reliable lineage signal on their own.

Extraction gotcha

The archive's own readme file extracted with a GBK-encoded filename that unrar's locale handling left as raw undecodable bytes (˵\udcc3\udcf7\udcceļ\udcfe.txt — Python surrogate-escaped). Renamed it to its correct decoded name (说明文件.txt) via a small Python os.fsencode/gb18030-decode script rather than leaving the mojibake filename in work/.

Fixes applied (proactive, per AGENTS.md catalog, before first boot)

• §15h (is_chinese/check_legal_name): adm/simul_efun/chinese.lpc's is_chinese() had the classic GBK-byte-range check (str[i] < 161, alternating-byte-offset check) — replaced with a CJK Unicode codepoint range test. adm/daemons/logind.lpc's check_legal_name(name, maxlen) halved its GBK-byte-calibrated bound (maxlen → maxlen / 2, and strlen(name) < 2 → < 1) and a separate combined surname+given-name length guard (strlen(fname) < 4 → < 2, matching the message's own "至少要有两个汉字" wording).
• §15h deep named.lpc fix (same shape as the nitan-family/chidi/dtsl precedent): adm/daemons/named.lpc's PATH(name) macro (name[0..1] → name[0..0]), all strlen(name) < 2 guards → < 1, the sliding-window similar-name check (name[i..i+3]/name[i..i+5] 2/3-char GBK windows → name[i..i+1]/name[i..i+2] UTF-8 windows, loop bound l - 4 → l - 2, combined guard < 4 → < 2, inner guard i+6<=l → i+3<=l).
• §15n (custom securityd.lpc ACL blocking the driver's own compile-time source loading): added an explicit case "load_object"/"recompile_object"/"include": return 1; at the TOP of valid_read, before the ! objectp(user) branch that would otherwise error() on a non-object/non-string user (a real risk here — this securityd's error-throwing branch is unguarded, unlike some sibling libs' softer return 0). Also softened both valid_read's and valid_write's error("...Invalid argument type of user.") calls to return 0 (deny) as defense-in-depth so a driver-internal check can never crash this apply outright.
• §14 (valid_override 3-arg upgrade): master.lpc's valid_override widened from 2-arg to varargs int valid_override(file, name, main_file), checking main_file == SIMUL_EFUN_OB/MASTER_OB too — relevant since simul_efun.lpc composes ~10 #included fragment files under /adm/simul_efun/.
• §8d/§15o insurance: added master::get_include_path() (prepends the compiling file's own directory) even though no live <local.h>-style bug was found on inspection (all local headers in this lib — herb.h, fish.h, medicine.h, wedding-ring.h — are already #included with quotes, not angle brackets) — cheap, free insurance matching established practice for any lib with a genuinely custom securityd.
• §15ae, TWO separate instances (private add_action-dispatched functions silently never firing): 1. feature/command.lpc's central command_hook(string arg) was private nomask — dropped private, kept nomask. This is the CORE post-login command dispatch; without this fix EVERY typed command (look, score, anything) would silently do nothing. 2. New second instance found via the mandated grep sweep: adm/npc/luban.lpc (the "鲁班" NPC, a house-building/contractor feature) registers 18 of its own commands via add_action (do_stop/do_answer/do_desc/do_show/do_changename/ do_changeid/do_changetype/do_changedesc/do_finish/ do_withdraw/decide_withdraw/do_demolish/do_help/do_list/ do_type/do_agree/do_reject/do_delete), every one declared private int name(...) (both forward declarations and real definitions). Fixed all 18 by dropping private from both the declaration and definition of each. This is an optional gameplay feature (build-a-house via the 鲁班 NPC), not on the critical registration/command path, but matches the exact same bug shape — worth fixing since it's mechanical and cheap. Confirms the project's standing instruction to grep every add_action-dispatched hook for private is not just a formality — this lib had it twice.
• §15s + §15aa combined (adm/simul_efun/message.lpc): the local message(mixed arg, string message, mixed target, mixed exclude) override is defined near the BOTTOM of the file, but tell_object/tell_room/shout/write/say (all textually earlier) call the bare identifier message(...) first — since message is also a real driver efun name, this silently binds those earlier calls straight to efun::message(), bypassing the override. Added a varargs void message(...) forward declaration at the top of the file (matching the real definition's signature) before any caller. Also fixed the real definition to coerce a falsy/omitted exclude (varargs default 0) to ({}) before delegating to efun::message(), since this driver's real efun rejects a raw int 0 for that argument (void | object | object * only).
• §15w (log_error() broadcasting compile *warnings*, not just real errors, to whoever's online): gated master.lpc's log_error() player broadcast on the message NOT containing the substring "warning:" (matching this driver's actual lowercase diagnostic prefix), still logging everything to /log/log regardless.
• §15p (proactive DNS/intermud daemon exclusion): removed /adm/daemons/network/dns_master from adm/etc/preload before the first boot attempt. (Checked cmds/usr/mudlist.c's inline DNS_MASTER call per §15ab's "can still bite even when excluded from preload" — already self-guarded with if (! find_object(DNS_MASTER)) return notify_fail(...), confirmed harmless.)
• §3 counterexample (blanket \bstatic\b→nosave sed colliding with a pre-existing "static/..." log-path naming convention): this lib has ~54 log_file("static/XXX", ...) call sites (crash dumps, security logs, promotion logs, per-command usage logs, etc.) that the blanket sed rewrote to "nosave/XXX". Reverted all of them back to "static/ via a scoped grep -l '"nosave/' | xargs sed -i 's/"nosave\//"static\//g' pass (safe here since 100% of the matches were this exact naming convention, confirmed by diffing against the raw pre-conversion source first).
• Directories created (none of /log, /log/static, /log/user, /binaries existed anywhere in the raw archive, despite being written to by log_file()/assure_file()-less call sites): mkdir -p work/log work/log/static work/log/user work/binaries before first boot, per §15ah's "missing save-data directory silently aborts a write_file() with no catch()" lesson. libs/zhongjidiyu/log/ (sibling of config.fluffos, the driver's own CWD-relative "log directory") also created.

New/notable bugs found (not yet in AGENTS.md's numbered catalog)

1. virtuald.lpc's "virtual object" fallback treats file_size() == -1 (nonexistent) as truthy "file exists" — compile_object(file) walks file's path components and does if (file_size(name + ".lpc")) { ob = name->query_virtual_room(...); }. Since file_size() returns -1 for a missing path (a nonzero int, so truthy), this ALWAYS "succeeds" into calling name->query_virtual_room() for a name that doesn't exist as an object either — which triggers the driver to lazily load_object(name), which fails (file doesn't exist), which calls master::compile_object(name) AGAIN as the driver's standard "couldn't find this file, try a virtual object" fallback, which calls virtuald.compile_object(name) again... recursing until Inherit chain too deep: > 30 aborts it. This isn't cosmetic: it fired continuously (every NPC-patrol heartbeat tick referencing a room in the missing /d/city//d/city3 zones — see gap #2 below) and, before it was fixed, appeared to be eating enough eval-cost/CPU each tick that the FIRST test session's character landed with a corrupted room state (fell back to START_ROOM which itself doesn't exist either, see next item). Fix: file_size(name + ".lpc") >= 0 (a real, existing file) instead of a bare truthy check. This is a genuine, driver-lineage-wide engine bug independent of missing content — worth checking any other ES-II-family lib with a virtuald.lpc/"virtual object" fallback for the same bare-truthy-file_size() shape. 2. Missing zone content, confirmed real archive gap (§13): /d contains ONLY death/, pk/, register/ — every city/sect zone (/d/city, /d/city3, /d/wudang, /d/xuedao, /d/beijing, /d/shaolin, /d/xingxiu, /d/heimuya, dozens more) referenced by START_ROOM (/d/city/kedian), cmds/chat/enterworld.lpc, npcd.lpc's patrol tables, mapd.lpc's map data, and 48 clone/board/*_b.lpc bulletin boards is genuinely absent — this really is the "bottom-layer source code... for study" skeleton release its own readme describes, not a full world pack. Also affects a whole sub-feature: /data/room/moman/* (a "house" template set, 绿竹苑) that #includes /d/room/roomnpc/roomnpc.h (also missing, defines is_owner()/owner_is_present()) — 9 files affected, all under the same missing /d/room/ tree. Because START_ROOM itself (/d/city/kedian) doesn't exist, logind.lpc's own fallback-to- START_ROOM path is dead code that always fails too — but this is masked in practice because REGISTER_ROOM (/d/register/entry, which DOES exist, home to NPCs 水笙/狄云) is what a genuinely fresh (! registered) player actually uses, so the missing /d/city never surfaces during ordinary registration. Documented, not fabricated. 3. d/register/npc/diyun.lpc unguarded factory-chain call (§15e): carry_object("/d/city/obj/cloth.lpc")->wear() in this NPC's create() — /d/city/obj/cloth.lpc is part of the missing /d/city zone (gap #2), so carry_object() correctly returns 0, and the unguarded ->wear() on it threw Bad argument 1 to EFUN call_other() every time this NPC's create() ran (i.e. every time /d/register/entry, the REAL register room, reset) — which, because the error was inside a catch(load_object(startroom)) in logind.lpc, made the ENTIRE register-room load look like it failed, sending every new character to the (also nonexistent) START_ROOM fallback instead of the real, populated register room. This was the single highest-impact fix in this pass — before it, characters landed with a broken/empty environment and look/score appeared to produce no output at all. Fixed with the standard if (objectp(cloth)) cloth->wear(); guard; the NPC simply goes unclothed (acceptable, matches precedent). 4. Genuinely broken/orphaned, never-loaded daemon files (not fixed, not reachable in normal play — same category as §12's "orphaned non-LPC file", just for an incomplete .lpc instead): adm/daemons/ cruised.lpc ("漫游精灵", cross-server player roaming) uses vfd/ socket_info without ever declaring them, references STREAM_BINARY without #include <net/socket.h>, and has a CAHNNEL_D typo (for CHANNEL_D) — none of CRUISE_D is even #defined anywhere, and nothing else in the lib references it; not in preload. adm/daemons/ quest/girl.lpc similarly references undeclared rcv_npcs/ supply_objs globals and is never loaded by anything else either. Both read as genuinely incomplete/abandoned files from the original author, consistent with the archive's own "study source, not a full release" disclaimer. 5. Pre-existing typos, fixed (unrelated to encoding/rename, confirmed via raw-byte diff against the original archive): - inherit/condition/damo.lpc:225: if (! intp(cnd["level"]) || ! intp(cnd["remain"] || ! stringp(cnd["id"]))) — a misplaced closing paren (should close after cnd["remain"], not after cnd["id"]). - Same file, a genuine corrupted byte in the RAW archive (not our conversion): a stray invalid 0xf7 byte immediately before a closing quote at raw byte offset 7414 (inside "发作身亡了") made the whole file fail strict GB18030 decoding, so convert_lib.sh fell back to iconv -c (lossy), whose "eats an adjacent real byte" recovery heuristic then corrupted a LATER, unrelated string literal ("长长的吁" @ line 278) into 5 invalid UTF-8 bytes — the exact xo_final/tianxia pattern documented in AGENTS.md's Encoding section. Fixed by stripping the single stray byte from the RAW bytes directly (confirmed via bytes.decode('gb18030') raising UnicodeDecodeError at that exact offset) and re-decoding the whole file cleanly, rather than patching the already-corrupted UTF-8 output. - adm/simul_efun/util.lpc never #include <ansi.h> despite using NOR/HIY/HIW/HIR/HIG/HIC/WHT/BCYN throughout — silent in a real boot only because simul_efun.lpc (which #includes this file) already pulls in ansi.h itself first (§6b: works fine composed, fails standalone under lpcc). Added the #include directly for correctness/self-sufficiency (harmless, header-guarded). - inherit/misc/quest.lpc's set_information(string key, string info) — narrower than QUEST_D's own mixed info, and 5 quest template files (clone/quest/{explore,judge,search,shen,supply}.lpc) pass a function-pointer closure ((: ask_npc :)), not a string — widened to mixed info (matches AGENTS.md's already-cataloged "quest set_information signature" bug class, e.g. yhyxs).

Confirmed NOT needed (checked, not present)

• §4 (lazy load_object(SECURITY_D) recursion in master.lpc's valid_read/valid_write): not present — both applies only do find_object(SECURITY_D), never load_object().
• §8c (this_player()-override footgun in valid_read/valid_write): not present — this securityd's applies use the user parameter directly (as passed by the driver/master), never silently substitute this_player().
• §15l (master.lpc destructing SIMUL_EFUN_OB in create()): not present — create() is just a write() banner line.
• §8e (tail() missing-efun): no tail( calls anywhere.
• §15x (hardcoded MUD_PORT mismatch): no such constant — connect() in master.lpc doesn't check the port at all; versiond.lpc reads the port dynamically via get_config(__MUD_PORT__).
• §15r (check_config.lpc-style driver-version self-check): no such file inherited into master.lpc/simul_efun.lpc.
• §15u (dormant "phone-home license check"): none found in securityd.lpc (this one's fully readable, no obfuscated destructive function).
• §15ag (ed_start/ed_cmd/query_ed_mode, __OLD_ED__ gap): no code calls any of these (only doc/efuns/* reference pages and a stale data/versiond.o save blob mention the strings).
• Uppercase .C files (§ shenmo precedent): none found (find work -name '*.C' empty).
• §15ad-style non-fatal versiond.lpc config-ID mismatch: PRESENT but confirmed harmless by the same reasoning as yhyxs — setup() sets version_ok = 1 synchronously (this is a local release server) BEFORE the deferred call_out("in_server", 2) that does the broken socket_bind() (using get_config(__MUD_PORT__), which this driver build reports as an empty string here, so port = "" + 12 becomes the string "12", and socket_bind() rejects a string). Logged once per boot as *Bad argument 2 to socket_bind() Expected: int Got: "12". — cosmetic, version_ok is already correct by the time it fires.

Testing methodology note (worth flagging for future libs)

This lib sets user->set("env/prompt", "time") for every new player — a live, once-per-second clock prompt re-drawn from the character's own heart_beat() (inherit/char/char.lpc), independent of the driver's write_prompt()/process_input() applies. scripts/mudclient.py's --idle pacing logic resets its "quiet" timer on ANY received byte, including these once-a-second heartbeat prompt redraws — with the default/ larger --idle values used successfully on most other libs (1.0–3.5s), the heartbeat's constant ~1s drip meant the "idle" condition was never satisfied, so queued --send lines for look/score/quit were silently never actually transmitted at all (confirmed via -dconnections driver debug output: zero get_user_command() calls for those lines). This looked EXACTLY like the §15ae "commands silently do nothing" failure mode and cost significant diagnostic time (traced through add_action, exec()/interactive-transfer, process_input()/APPLY_PROCESS_INPUT, even a raw byte-level socket script) before the actual cause — a test-tool pacing artifact specific to libs with a persistent per-second heartbeat prompt, not a mudlib bug — was found. Fix for testing: use a SHORT --idle (0.3–0.5s, shorter than the heartbeat period) on any lib with a live/clock-style env/prompt, or use a raw socket script that sends after a fixed wait rather than an idle-detection loop. Worth promoting to AGENTS.md as its own catalog entry for the next lib with a similar live-prompt feature.

Registration flow shape

get_id (English id, 3-10 letters) → confirm_id (y/n, only for a NEW id) → get_surname (Chinese, ≤2 chars, optional/blank-skippable) → get_name (Chinese, ≤2 chars, combined surname+name must be ≥2 chars, checked against NAME_D->invalid_new_name() for banned Jin Yong character names and near-duplicate existing names) → new_ad_password/ confirm_ad_password (a SEPARATE "admin/recovery password", ≥5 chars) → new_password/confirm_password (the regular login password, ≥5 chars) → get_type (1-5, character "build" archetype, blank defaults to 5/balanced) → get_gender (m/f) → enter_world(). No hidden BIG5/GB prompt, no client-version gate, no anti-flood throttle found in band.lpc (only is_banned(site) — a real IP-ban list, not a per-connection registration throttle).

Interactive test result

Full registration + post-login flow verified in two separate continuous connections with real Chinese names "秦风" (male) and "林枫" (female, type 耐力型), both reaching the actual populated starting room "世外桃源" (with NPCs 水笙/狄云 present, exits west/east/north/south). Both sessions then sent look (re-displayed the correct room description), score (correctly responded "还没有出生呐，察看什么？" — a legitimate game-rule message, since neither test character completed the additional register <email> + "投胎做人" reincarnation step the unregistered banner describes; this is real command execution, not a silent failure), and quit (correctly responded "欢迎下次再来！" and disconnected). Zero debug.log errors in the final clean runs beyond the confirmed-harmless versiond.lpc socket_bind cosmetic line. command_hook's private removal (§15ae) confirmed working via this real post-login-command test, per the project's standing policy.

lpcc sweep

1169 files, 1072 pass / 97 fail (91.7%). All 97 failures triaged into already-understood categories, none require further fixing:

• 48: couldn't find object '/d/...' — the missing zone content (gap #2 above), bulletin boards referencing nonexistent rooms across dozens of sects/cities.
• 32: the /data/room/moman/* house-template module's missing /d/room/roomnpc/roomnpc.h (is_owner/owner_is_present) — same root cause, part of the same missing /d/room/ tree.
• ~11: adm/daemons/cruised.lpc (6 distinct errors) + adm/daemons/quest/ girl.lpc (2 errors) — genuinely broken/abandoned, never-loaded daemon files (item #4 above).
• 1: adm/simul_efun/message.lpc's message_system(string message) — its OWN parameter is named message, shadowing the function name it then tries to call (message("system", ...)) — a real-looking bug that is confirmed a lpcc-standalone-compile artifact per §6b (this file is #included into simul_efun.lpc, not a standalone object; the real driver boot never hits this — verified zero hits for this error text in a clean boot's debug.log). Not fixed, to avoid risking a real behavior change to a confirmed non-issue.

Files touched (for reference)

adm/simul_efun/chinese.lpc, adm/daemons/logind.lpc, adm/daemons/ named.lpc, adm/daemons/securityd.lpc, adm/single/master.lpc, feature/command.lpc, adm/simul_efun/message.lpc, clone/user/user.lpc, adm/etc/preload, adm/daemons/virtuald.lpc, d/register/npc/diyun.lpc, adm/npc/luban.lpc, inherit/condition/damo.lpc, adm/simul_efun/ util.lpc, inherit/misc/quest.lpc, plus the ~30-file "nosave/ → "static/ revert (§3 counterexample) and directory creation (log/, log/static/, log/user/, binaries/).

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

• LPC formatter applied to all .lpc under work/ (1,169 total, 1,154 written, 12 unchanged, 3 self-checked errors left untouched).
• Native re-test against the freshly rebuilt driver (~/src/fluffos/build-debug/src/driver): clean boot, only the pre-existing benign versiond.lpc/socket_bind() message in debug.log (§15ad). Full registration flow re-verified end-to-end with a fresh id/real Chinese surname+given-name (zjdytest/秦风), admin+regular password, character-type + gender selection, reaching 世外桃源; look (re-showed room), score (correctly returned "还没 有出生呐，察看什么？" — expected pre-"投胎" message, per the documented design, not a bug), quit (clean disconnect) all confirmed correct. Same --idle 0.5 live-clock-prompt workaround from the original pass (§15an) was still needed and still works.
• WASM test (scripts/wasm_client.js against build-wasm/src): boots (only the expected sockets-package-unavailable-class preload messages), but every login is disconnected immediately after the banner, before the id prompt ever appears — root-caused via the harness's raw log: versiond.lpc genuinely uses socket_create()/socket_bind()/socket_listen() (a real cross-server version-sync feature), which are hard "Undefined function" compile errors under WASM's sockets-less build, leaving /adm/daemons/versiond as a program-less object; logind.lpc's logon() then calls VERSION_D->is_version_ok() with no catch() (unlike the adjacent catch(MUDLIST_CMD->main()) one line above it), which throws *No program in object '/adm/daemons/ versiond'! uncaught — the driver's own new_conn_handler() treats this as a failed logon() and disconnects the connection outright (confirmed via new_conn_handler: logon() on object clone/user/ login#0 has failed, the user is disconnected. in the raw log, appearing immediately after the versiond error). This is the exact same shape independently found on zhonghua2 (a different lineage) in this same pass — evidently a common idiom across several of these archives (an unguarded VERSION_D->is_version_ok() call right after a guarded catch(MUDLIST_CMD->main())). Root cause is entirely the WASM sandbox's missing sockets package colliding with a genuinely socket-based feature this lib relies on — confirmed NOT a mudlib bug (native boots and completes full registration/login cleanly, verified above) — so, per the standing instruction for WASM-mode limitations, not patched. This is a distinct failure mode from the documented query_ip_number() gotcha (this lib's logon() doesn't even reach an IP-format gate before hitting this one) — noted separately for whoever next triages WASM playability across the "hell"/ES-II family. (Superseded by the 2026-07 WASM-enablement pass below — now patched.)

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

Standard pass per AGENTS.md §1.3b/§1.3c/§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.
• adm/daemons/logind.lpc logon() (~line 67) and get_id() (~line 130): both VERSION_D->is_version_ok() gates guarded with find_object(VERSION_D) — absent versiond (WASM: sockets package missing ⇒ program-less object ⇒ uncaught throw killed every login before the id prompt) now means "version ok". This was the documented WASM login blocker.
• adm/daemons/closed.lpc heart_beat() and adm/daemons/questd.lpc start_all_quest(): same find_object(VERSION_D) guard (both preloaded).
• The iplimit > 3 per-IP cap in logon() is already inside #if 0 (disabled upstream); no uptime()/anti-flood gates exist otherwise — nothing else to bypass.
• NEW pre-existing bug found & fixed — re-login was broken for ALL accounts: clone/user/user.lpc's restore() override validates a "sec_id" integrity checksum with crypt(calc_sec_id(1), sec_id) != sec_id, but its own save() stores set("sec_id", calc_sec_id()) — the PLAINTEXT stub value "none", never a crypt() of it. The invariant can never hold, so every re-login printed 无法读取你的数据档案 and destructed (registration itself worked, which is why earlier passes — which only tested registration+look/score/quit in one session — never hit it). Fix (user.lpc ~line 153): also accept the plaintext form save() actually writes (&& sec_id != calc_sec_id(1)). Verified: plain player and fluffos both re-login fine now.
• Admin seeded: fluffos / 浮云, rank (admin) (top of wiz_levels) via adm/etc/wizlist. Two-password lineage: 管理密码 Admin@2026, 普通密码 Mud@2026 (must differ, same deviation as zhonghua2, documented in README). Registered as plain player first, wizlist entry added afterwards + restart (consistent with the lineage's anti-steal conventions). Verified: 目前权限：(admin) banner, wizard-style room display, update /d/register/entry.lpc → 成功.
• Retest: fresh registration (regtest/秦风) end-to-end into 世外桃源 + re-login as regtest (exercising the fixed restore path) + look/score/ quit all correct; test saves removed. Final boot's debug.log clean except the known §15ad versiond socket_bind() cosmetic error. (Longer-running boots also surface pre-existing quest-daemon heart_beat errors — Bad argument 1 to call_other / Bad argument 2 to present() in /clone/quest/* / /adm/daemons/quest/* — caused by the skeleton archive's missing zones, present before this pass, unrelated to the VERSION_D guards.)
• Second pre-existing bug found & fixed (backported from sibling zjdyaryl, same symptom confirmed here): first login after every boot had ALL commands dead (even quit → 什么？), working from the second login on. logind.lpc check_ok() lazily loads MESSAGE_D (adm/daemons/network/messaged.lpc) whose create() → startup_udp() → socket_bind(fd, "<string>") throws uncaught (§15ad get_config port-ID mismatch), aborting the first login's remaining setup. Fixed: messaged.lpc create() wraps startup_udp() in catch(), and check_ok() wraps MESSAGE_D->find_chatter() in catch() (also required under WASM where messaged fails to compile entirely). Verified: first login after a fresh boot now runs update successfully.

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

200s scripts/wasm_boot_watch.sh sit: the quest-daemon heart_beat() "Bad argument 1 to call_other"/"Bad argument 2 to present()" errors (clone/quest/judge.lpc, .../search.lpc, .../supply.lpc, .../shen.lpc) reappeared, matching the already-documented pre-existing "skeleton archive's missing zones" class (recorded above, unrelated to the VERSION_D guards) — confirmed still the same already-known class, no new fix needed. ftpd.lpc/versiond.lpc sockets-absent preload errors also match the already-documented §1.3c class (caught, cosmetic). One genuinely new proactive fix: this lib shares inherit/item/combined.lpc (private void destruct_me(), call_out-self-invoked when a money stack hits amount 0) and adm/simul_efun/object.lpc (file_owner() return name bug) with sibling zhonghua2, where both fired live during that lib's sit — ported both fixes here proactively (dropped private on destruct_me; file_owner() now return dir) even though neither triggered visibly in this lib's own 200s window. Retest: fresh registration (id zjdsanb) into 世外桃源, look/score(pre-投胎 message)/quit clean, no regressions.

深度功能测试 / Deep functional test (round two, per AGENTS.md §10.7)

Native driver only (~/src/fluffos/build-debug/src/driver config.fluffos from libs/zhongjidiyu/), scripts/mudclient.py --idle 0.5-0.6 throughout (this lib's live per-second clock prompt requires it, §15an/AGENTS.md §8.3 item 1 — reconfirmed still necessary). Read help/newbie and help/intro first, in full, before touching the client — they document the actual intended flow precisely: 水笙 register <email> → decide → walk to one of 世外桃源's 4 exits and out to pick a 性格/character trait → out again into 阎罗殿 (Hall of Yama) → wash (roll 4 innate stats) → born <地名> (choose birthplace, triggers reincarnation and moves the player to their starting zone). Safe sparring is the ordinary fight command ("较量" — per help/intro, stops before either side dies); skill learning is bai (apprentice) then xue (learn) from an NPC teacher; sect/faction join happens the same way, via bai-ing a sect NPC.

One continuous session, two test characters (both left in place as evidence): hufengbo/秦风 (male, 猛士型, admin password Admin@2026, regular password Mud@2026) and muyunqing/林枫 (female, 耐力型, same password scheme). Both completed real registration through register+decide+personality-room+wash end-to-end, confirmed via look/score/i at every step. Admin account fluffos/浮云 (seeded by an earlier pass) also used, with goto, to reach content a fresh player account cannot (see below).

Confirmed still correctly applied, not re-diagnosed as new:

• §7.14's file_size()-boolean-context fix in adm/daemons/virtuald.lpc (file_size(name + ".lpc") >= 0, not a bare truthy check) — read the live file, matches the documented fix exactly.
• §8.3a's private nomask command_hook fix in feature/command.lpc — confirmed working LIVE via an NPC's own command() self-call, not just player-typed commands: 陆天抒 (the 光明磊落 personality-room NPC) correctly greets and speaks via init() → command("hi ...") / command("say ...") the instant a fresh character walks in, and 地藏王's own command("kick ...")/command("say ...") self-calls in 阎罗殿 fired correctly too. The 18-handler adm/npc/luban.lpc fix wasn't independently re-verified live (费 the feature is gated behind the missing /d/city zone, see below) but the file itself still shows all 18 handlers correctly private-free.

Bug found #1 (NEW) — a failed born corrupts a real account into a permanently no-environment login, and enter_world()'s own fallback chain doesn't actually catch it. This is the highest-impact finding of this pass. Root cause, two parts, both fixed:

1. d/register/yanluodian.lpc's do_born() persisted startroom/born/born_family to the player's SAVE data unconditionally, BEFORE checking whether the destination object actually resolved (obj = load_object(dest)/find_object(dest)). Since every born destination in this skeleton archive points into a missing zone (/d/city, /d/beijing, etc. — none exist, see gap #2 above), literally every born attempt "fails" (prints "牛头一呆， 搔搔头说：怎么好像有问题...") — but the state was ALREADY written by then. This flips enter_world()'s own !stringp(user->query("born")) safety net (which would otherwise keep an unborn player parked safely in BORN_ROOM/阎罗殿 on every future login) to false, so it stops applying. 2. adm/daemons/logind.lpc's enter_world() then tries to resolve the now-persisted (broken) startroom. Its own fallback logic only wrapped catch(load_object(startroom)), not the subsequent user->move(startroom) — and load_object() on a path that simply doesn't exist (with no virtual-object fallback) returns 0 without throwing, so the catch() sees "no error" and falls into the UNGUARDED user->move(startroom). feature/move.lpc's move() does call_other(dest, "???") as its very first destination-resolution step, which genuinely throws (*call_other() couldn't find object '/d/city/kedian'.) for a nonexistent path — uncaught here, aborting enter_world() mid-function. Confirmed live: reconnecting as the already-"born" hufengbo printed "你无法进入这个世界，可能你的档案 出了一些问题" and then look returned "你的四周灰蒙蒙地一片，什么 也没有" — the exact §7.22 "no environment at all" symptom, but via a distinct mechanism (neither an eval-cost abort nor a compile error — a plain unguarded call_other() inside an un-catch()-wrapped move()). score still worked (doesn't need an environment), which is what made this state look survivable at a glance — look/movement were the ones silently dead. This lib's own hardcoded START_ROOM fallback (/d/city/kedian) is ALSO one of the missing paths, so even the "else" branch's intended recovery was reaching for a target that doesn't exist either.

Fix, two parts (both live-verified — see below): - d/register/yanluodian.lpc: moved the !objectp(obj) failure check to immediately after obj is computed, in BOTH the plain-string and the family/世家 branches, before any me->set(...) call. A failed born now leaves the account's startroom/born fields untouched, exactly as if born had never been attempted. - adm/daemons/logind.lpc's enter_world(): wrapped the actual user->move(startroom) in catch() too (not just load_object()), and added a THIRD fallback level — if even START_ROOM fails to load/move into (true on this specific archive), fall back to BORN_ROOM (阎罗殿, always present) instead of leaving the player with no environment at all.

Live verification: before the fix, reconnecting as hufengbo (who had already borned to a missing destination in a prior session) reproduced the crash exactly as described, with *call_other() couldn't find object '/d/city/kedian'. in debug.log. After the fix (driver restarted to pick it up), the SAME already-corrupted hufengbo account now lands in 阎罗殿 via the BORN_ROOM fallback (kicked onward to 虚空/void by 地藏王's own already-correct "you're already born, get out" NPC logic — a real, pre-existing, working game mechanic, not a bug), and look/score/movement/quit all work normally from there — no more no-environment state. Separately, a FRESH character (muyunqing) that ran wash+born <missing dest> AFTER the fix correctly stayed in the "not yet born" state (score → "还没有出生呐，察看什么？", still standing in 阎罗殿 with 地藏王 still willing to hand over the newbie book) on both the immediate next command and a later real reconnect — confirming the root-cause fix (part 1) prevents the corruption from happening at all for any new account, while the defense-in-depth fix (part 2) also protects any account that manages to reach a broken startroom by some other path. No new debug.log errors from any of this.

Doesn't match any existing catalog entry exactly — closest are §7.14 ("missing post-registration destination room... guard the move with load_object()/fallback to START_ROOM") and §7.22 (the "no environment at all" symptom shape) but the specific mechanism (a catch() that only wraps the probe call, not the actual unguarded move() right after it, PLUS a state-mutation-before-validation ordering bug in the caller) is new. Worth a full write-up as a candidate new §7.x entry — see final report.

Bug found #2 (matches AGENTS.md §8.3a's addendum exactly, not new) — feature/action.lpc's eval_function() was still private. Found live: apply() with insufficient permission: ... function: eval_function, origin: internal, needs: private, has: hidden fired in debug.log during ordinary play (a delayed-effect call_out against our own test character's body). This is the SAME file and SAME function name as the xuanjianlu-derived precedent already in §8.3a's addendum ("the shared 'delayed status effect' primitive used by 130+ kungfu-skill/drug files"). The lib's own "WASM long-sit boot-watch pass" note (above) proactively ported §8.3a's addendum fix for inherit/item/combined.lpc's destruct_me() and adm/simul_efun/object.lpc's file_owner() from sibling zhonghua2, but missed this THIRD instance — feature/action.lpc itself apparently wasn't part of that proactive sweep. Fixed identically: dropped private, kept the function otherwise unchanged. Proactively grepped the whole lib (every call_out("name", ...) site cross-referenced against a same-file private declaration of name) for any other instance — found none: every other flagged hit is a call_out target defined in a standalone daemon/NPC/command file that is never inherited elsewhere (confirmed via grep -rl 'inherit.*"<path>"'), so the DECL_PRIVATE→DECL_HIDDEN-once-inherited mechanism doesn't apply to them.

Bug found #3 (matches AGENTS.md §7.25 exactly, not new, but a genuinely new instance/location) — inherit/room/room.lpc's shared make_inventory() had no guard, crashing the ENTIRE afterlife zone's first-ever visit. d/death/npc/wgargoyle.lpc ("白无常") and d/death/npc/bgargoyle.lpc ("黑无常") both call set_skill("dodge", ...) in their own create() — and this archive's /kungfu/skill/ directory is entirely absent (not just the /d/city-style zone rooms already documented in gap #2 — the base skill-file tree itself is missing, a broader gap than previously written up). feature/skill.lpc's set_skill() correctly error()s on a missing skill file, but that error() fires DURING the NPC's create(), which new() propagates as an uncaught throw — and inherit/room/room.lpc's make_inventory() (ob = new(file); ob->set(...); ob->move(...); return ob;, called from every room's reset(), lib-wide) had zero catch() around any of it. Live-reproduced via admin goto /d/death/gate (DEATH_ROOM, where every player death sends them via feature/damage.lpc's die()): the room's FIRST-EVER reset() this boot aborted entirely, so goto silently failed to move the admin there at all (landed back at start with only a debug.log trace: *F_SKILL: No such skill (dodge), blamed at wgargoyle.lpc's create() via room.lpc's make_inventory()). Same shape independently confirmed for bgargoyle.lpc in /d/death/gateway.

Fix: wrapped make_inventory()'s new()/set()/move() chain in catch(), returning 0 on any failure (destructing a partially-constructed object first if new() itself did return one); and guarded the ONE call site in reset()'s case 1: branch that dereferenced the result without an objectp() check (ob[list[i]]->is_character()) — the default: (multi-object) branch already continued past the equivalent dereference for a freshly-(re)- made entry, so it didn't need the same guard. Live-verified this unblocked the ENTIRE death/afterlife zone, not just the one room: after the fix, admin goto successfully reached and looked at all 8 death-zone rooms (gate/鬼门关, road1/road2/road3/鬼门大道, inn1/小店, inn2/黑店, god1/天堂, god2/圣殿, block/空房间, gateway/酆都城门) with zero further errors beyond the two now-gracefully-caught missing-skill NPCs — this whole zone is otherwise 100% present and intact in the archive, just unreachable-on-first-visit before this fix. Matches §7.25's established pattern and fix precisely; the missing /kungfu/skill/ tree itself is left alone as a genuine archive content gap (not fabricated) — 白无常/黑无常 simply never spawn, same graceful-degradation outcome §7.25 already prescribes for this shape.

Confirmed content gap, broader than previously documented: the /kungfu/skill/ directory (referenced by include/globals.h's SKILL_D(x) macro, "/kungfu/skill/" + x) does not exist AT ALL in this archive — not one basic skill file (dodge, force, unarmed, etc., all named in help/newbie's own example cha output) is present, beyond the previously-documented missing /d/city-style zone rooms. This means even IF a player could reach a populated zone, xue/learn-style skill acquisition would have nothing to actually load. Documented here, not fabricated — matches the archive's own "study source, skeleton release" disclaimer.

Safe-sparring / skill-teacher / sect-join — confirmed NOT reachable live, by design of this skeleton archive, not a bug. Every currently-existing room in the archive (d/register/*, d/pk/*, d/death/*) explicitly sets no_fight: 1 — the fight command (§ help/intro's "较量") is real, present, and correctly wired (cmds/std/fight.lpc exists and is a normal add_action command), but there is no reachable room anywhere in this archive where combat is actually enabled. The bai(apprentice)/xue (learn) skill-teacher path and every sect-join NPC live under the missing /d/<sect> zones (§13/gap #2) and, per the finding above, even the underlying skill FILES they'd teach don't exist either. d/pk/ ("屠人场" — a PK tournament arena, 乌老大's sign/join commands) is a distinct, unrelated feature (scheduled PvP, not the newbie-doc's safe fight spar) and is itself a dead end too (west exit → /d/changan/..., missing). Consistent with, and extends, this lib's own already-documented "skeleton archive" gap — not attempted to fabricate.

Shop purchase / combat / death — explicitly NOT tested live, and cannot be, given the above. No shop, no fightable NPC, and no death trigger is reachable by an ordinary player account in this archive's current content. Code review only: feature/damage.lpc's die() calls me->move(DEATH_ROOM) (/d/death/gate, confirmed to exist and — after Bug #3's fix — load cleanly) and DEATH_ROOM->start_death(me); this path was not exercised by an actual player death since no live combat path exists to trigger it.

quit + debug.log grep, both unclean (net-dead) and clean-quit reconnect, both prompt and after a real wall-clock gap: all tested, all clean.

• Clean quit (both test characters, multiple times): correct "欢迎 下次再来！" message, unequipped-item drop logic runs without error (both characters' items were all equipped, so the drop loop's message("vision", ..., environment(me), ...) call — which is NOT itself guarded by an environment(me) check, only by whether any items were actually dropped — was never exercised with a null environment; noted as a theoretical residual risk but not reproduced, see final report).
• Every quit immediately followed by a same-second reconnect attempt correctly hit the documented 30-second quit-retention lockout ("你距 上一次退出时间只有N秒钟，请稍候再登录。") — confirms this project's own established §10.7 caveat about the lockout window in practice.
• Prompt reconnect after an UNCLEAN disconnect (letting mudclient.py's connection simply close without sending quit, simulating a network drop): correctly printed "重新连线完毕" and resumed in the exact same room with no state loss — this lib's net_dead() (clone/user/ user.lpc) does NOT use a VOID_OB-parking pattern at all (confirmed by reading the code: the player object is simply left in its current environment, heart_beat disabled, and a 900s user_dump call_out scheduled) — so the §7.20 "void-parking without a location-restore path" bug class structurally cannot occur here; reconnect() just clears the net_dead flag and re-enables commands in place. No location-restore step is even needed.
• Reconnect after a clean quit AND a genuine ~50s+ real wall-clock wait (past the lockout): went through the FULL fresh-login code path (login banner + "你上次光临终极地狱是 ... 连接的" last-visit message, not the short-circuit "重新连线完毕"), landed back in the exact same room/state (muyunqing: still 阎罗殿, still un-born, 地藏王 handed the newbie book again since it wasn't in inventory) — full persistence confirmed correct.
• The full NET_DEAD_TIMEOUT (900s / 15 minutes) user_dump force-quit-after-abandonment escalation path was NOT exercised live (would require a 15-minute uninterrupted wait) — reviewed via code instead (clone/user/user.lpc's user_dump()/DUMP_NET_DEAD case: announces the timeout to the room, then command("quit") with a force_quit() fallback if that fails) and looks correct/unremarkable, but this specific path is unverified live, per §10.7 item 6's explicit-fallback allowance.

Files touched, this pass

• libs/zhongjidiyu/work/d/register/yanluodian.lpc — do_born(): check objectp(obj) before persisting startroom/born/born_family state, in both branches (Bug #1, part 1).
• libs/zhongjidiyu/work/adm/daemons/logind.lpc — enter_world(): wrap user->move(startroom) in catch() too (not just load_object()), add a BORN_ROOM last-resort fallback beyond START_ROOM (Bug #1, part 2).
• libs/zhongjidiyu/work/feature/action.lpc — eval_function(): dropped private (Bug #2, matches §8.3a's addendum).
• libs/zhongjidiyu/work/inherit/room/room.lpc — make_inventory(): wrapped in catch(), returns 0 on failure instead of propagating; reset()'s case 1: call site guarded with objectp() before dereferencing (Bug #3, matches §7.25).
• Test character saves left in place as evidence: data/login/h/hufengbo.o + data/user/h/hufengbo.o (秦风, the account that reproduced Bug #1 pre-fix and was re-verified recovering post-fix), data/login/m/muyunqing.o + data/user/m/muyunqing.o (林枫, registered entirely after the fix, confirms no regression for a clean account). data/user/f/fluffos.o (admin) timestamp-touched by ordinary login, no content change beyond last_on.

WASM 修复摘要（迁移自 meta.json 的 group_note）

尽管标题相似，但和 hell 那对"终极地狱"没有关系。状态已从过时的 limited 修正——这份档案自己的 README 和 group_note 里从未记录过任何缺陷说明，本轮重新测试也没有发现：管理员登录（fluffos/Mud@2026）干净正常，'目前权限：(admin)'。另外还发现并修复了一个和手足档案 zjdyaryl/zjdyzj 共有的真实 bug：clone/user/login.lpc 的 query_save_file() 只防范了 id 不是字符串的情况（!stringp(id)），没有防范空字符串的情况——对 "" 取 id[0] 会回传整数 0，而 sprintf 的 %c 格式会拒绝这个值（'Incorrect argument to type %c, must be valid UTF8 char'），在任何 id 被输入之前，第一次自动呼叫 query_save_file() 就会崩溃。已通过额外检查 id == "" 修复。

§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 43 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 sub-threshold instance (2026-08-20)

Found during the §7.100 tail-sweep (below the original 166-lib survey's

=100-occurrence threshold, never checked). 39 live

replace_program(ROOM); occurrences across 38 files: 18 clone/megazine/*.lpc rooms, clone/misc/void.lpc, a 12-file d/pk/turen*.lpc cluster, a 6-file d/death/*.lpc cluster, plus this lib's own room-building tool clone/misc/roommaker.lpc (both a standalone line and a separate string-builder template line, both fixed) — so newly-built rooms were inheriting the bug too. A full post-fix grep confirms 0 live occurrences remain. Verified via a clean native driver boot (zero new debug.log errors, port listening, killed by exact PID after ~8s).

§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): 5 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.

Shop + 拜师 (2026-09-04 librarian loop) — cheap-confirm, no re-boot

The existing deep-test body already records, with concrete evidence, that this archive is a skeleton: every shipped room is no_fight: 1, shop/sect NPCs and skill files live under missing /d/<sect> trees, and "No shop, no fightable NPC" was an explicit live-test conclusion rather than a heading-only TODO. This pass re-read that section and did not re-boot. Still no reachable shop or bai path to exercise.

深度功能测试（§10.7，2026-10-04）— negative range ends

scripts/lpc_fix_negative_ranges.py: 4 line(s) in 1 file(s) count a range from the end with < (x[a..-1] -> x[a..<1], x[-2..] -> x[<2..]; KB 06 §7.209). In this driver a negative constant end gave "" / ({ }) and a negative start the whole value, so each of these returned the wrong slice; the compile warning A negative constant as the second element of arr[x..y] is gone from the files that compile. Range lvalues (s[0..-1] = text, the prepend idiom) are left alone. A paired compile of the changed files at HEAD and in the working tree showed no new error.

深度功能测试（§10.7，2026-10-06）— 与 zhonghua2 同源的三处编译警告

• include/npc/indian.lpc 与 masia.lpc 共用全局名 ma_sname/ma_pname；两个头文件被同一个守护进程（npcd 等）一起包含时，后包含的印度名表覆盖马来名表，generate_ma_name() 生成的是印度名字。印度表改名 in_sname/in_pname（KB 04 §6.10）。
• cmds/std/team/swear.lpc：结义失败提示被多余的 ; 截成两句，fail 只剩"……可惜现在有人不在，$N"，后半句"的提议只好作罢。"丢了；接回去。

验证：改动的文件和包含 npc/indian 的守护进程在 HEAD 与工作树分别加载（scripts/lpc_listcheck.sh，新进程），无回退。

深度功能测试（§10.7，2026-10-07）— 编译警告整理

scripts/lpc_warnings.py zhongjidiyu --fix 加 scripts/lpc_fix_no_effect.py --apply，再用 scripts/lpc_nitan_family_fixes.py zhongjidiyu（adm/daemons/natured.lpc 的空语句 lt[LT_MON];、adm/daemons/combatd.lpc 的空语句 your_temp["guarding"];），共改动 216 个 .lpc 和若干 .h，都不改变行为。HEAD 与工作树分别加载（新进程）：199 -> 199 通过，无回退。 未动：adm/daemons/cruised.lpc（跨站漫游守护进程）是编译不过的草稿（set、socket_info 未定义，CAHNNEL_D 拼错），没人加载。

启动时警告 85 -> 0；新进程登录正常。

• 2026-10-12：clone/misc/swmuding.lpc（神木王鼎）自己声明了一张 mapping insects 虫表，与继承的 clone/misc/muding.lpc 的 insects 同名；读表的是木鼎的代码，读到的是木鼎自己的表，神木王鼎的虫表从来没用上（Redeclaration of global variable 'insects'）。改名 sw_insects，create() 里 insects = sw_insects; 交给木鼎（nitan170911、wxddym、zjdy 系同修）。HEAD 与工作树各载入一次，regress 0。

编译警告收尾（2026-10-13）

scripts/lpc_twin_port.py adopt revive zhongjidiyu 搬来 revive 同日整理（cfd7e6a9897）的改动，与本库逐 token 相同，能编译的文件警告清零；改动的对象在 HEAD 与工作树各载入一次，regress 0。

• clone/megazine/obj/duka_3b.lpc（赌球卡 B）赔率变量声明两次，代码用后一份；删去前一行，赔率不变。

find_player() 只认在世界里的身体（2026-10-15，KB 07 §8.26）

simul_efun 的 filter_player() 加上 environment(ob)：注册中途断线留下的、不在任何房间的身体不再被当成玩家，两个以上时 find_player() 不会再报 The player(<id>) has existed in the world.（wllz3 实测过的问题）。本库若注册流程不会留下带玩家 uid 的身体，这一改动不改变行为。

NPC 带着本库没有的物品，载入失败（2026-10-15，KB 06 §7.73）

2 个文件、3 处在 create() 里 carry_object("<本库没有的文件>")->wear()（或 wield()、set_amount()）：carry_object() 得 0，接着的调用报 Bad argument 1 to EFUN call_other()，NPC 根本载入不了，该有它的房间里就少了这个人。多是从别的系统搬来的 NPC 用了原系统的物品路径（最常见：/d/city/npc/obj/qunzi、/d/shaolin/obj/changjian、/d/city/obj/cloth）。每处改为 { object carried_ob = carry_object(...); if (carried_ob) carried_ob->wear(); }（scripts/lpc_guard_carry_object.py）：NPC 照常出现，只是不带这件东西；文件补回来（或是虚拟对象）照样会带上。载入检查：改动的文件改前能载入 0 个，改后 1 个（1 个原来载入不了），没有退步；其余 1 个另有原因（多为 set_skill() 了本库没有的武功）。

set_skill() 遇到本库没有文件的武功不再抛错（2026-10-18，KB 06 §7.256）

feature/skill.lpc 里的 set_skill()、map_skill()、prepare_skill() 碰到没有武功文件的名字就 error("F_SKILL: No such skill (…)")，调用它们的 NPC create() 整个中断：NPC 载入不了，刷出它的房间也少一个人。本库目前没有对象设了缺失的武功；三处都改成 return：这一门武功不给、不映射，对象照常载入，只是少了它（缺的武功本身仍是内容缺口）。（scripts/lpc_guard_set_skill.py）

RANK_D 对 0 的称呼（2026-10-19，KB 06 §7.273）

adm/daemons/rankd.lpc 的 query_respect()、query_rude()、query_self()、query_self_rude() 里 4 个函数开头补上 if (!objectp(ob)) ob = this_object();。NPC 在 create() 里写 "…" + RANK_D->query_respect(this_player()) + "…"（inquiry 对话、chat_msg），驱动自己跑 reset() 时没有人在场，this_player() 是 0。旧驱动里 0->query("age") 返回 0，称呼走默认分支；新驱动直接报 Bad argument 1 to EFUN call_other()，房间的 make_inventory() 吞掉错误，被打死的 NPC 再也刷新不回来。用 scripts/lpc_reach_census.py --walk 找到（xkx100 的武当弟子、一灯、汴梁书生…）；修在守护程序里，不动 1,800 个 NPC 文件。

member_array('/', str) 改为 strsrch(str, "/")：字符字面量的 member_array 按字节搜索（2026-10-20，KB 06 §7.283）

member_array('/', str) 里的 '/' 是整数，驱动按 UTF-8 字节搜索，返回字节偏移（带第三个参数时起点也是字节）；最新的 FluffOS（923cdbc0）在字符串里有非 ASCII 文字时，对每个调用点警告一次（Warning: member_array(int, string) searches BYTES）。而 str[0..idx-1] 按字符切，中文字符串会在错误的位置截断。改为 strsrch(str, "/")（找不到同样返回 -1，按字符计数；scripts/lpc_fix_member_array_char.py，JSON 编码器等少数几处手改）。改动：/adm/daemons/vrm_server.lpc（1 处）、/cmds/arch/sameip.lpc（1 处）。

登录密码超过 255 字节：crypt() 抛错（2026-10-20，KB 06 §7.294）

驱动的 crypt() 对超过 255 字节的密钥直接报错（Error in crypt(), check your salt），老驱动则按任意长度散列。在登录提示符下输入 256 字节以上的密码，密码检查那一行就抛错，连线停在一个裸的 > 提示符上，登录走不完（password 指令的旧密码检查同样）；用 300 个 x 当参数的扫描发现的。/adm/single/simul_efun.lpc 里补了一个同名的 simul_efun crypt(key, salt)：只有超长的密钥才被截短（先截到 255 个字符，再逐字符截到 sizeof(to_buffer(key)) <= 255，限制按字节算，一个汉字占三个字节），再交给 efun::crypt()；普通密码原样散列，已存的密码散列不受影响（scripts/lpc_add_sefun.py crypt）。

get_config(__MUD_PORT__) 经旧版 runtime_config.h 得到字符串（2026-10-20，KB 06 §7.302）

本库的 include/runtime_config.h 是老 MudOS 的编号（整数设置从 15 起，本驱动从 256 起），get_config(__MUD_PORT__) 读到的是另一项设置（字符串），(int) 又只是编译期标注，于是 socket_bind(fd, LOCAL_PORT()) 报 Bad argument 2 to socket_bind()，版本守护进程 port = get_config(__MUD_PORT__) + VERSION_PORT 是字符串相加。头文件自己的注释写明 get_config(__MUD_PORT__) == __PORT__，所以代码里的这个调用（连同外面的 (int) / to_int()）改成了驱动的预定义 __PORT__（scripts/lpc_fix_mud_port.py）。改动：/adm/daemons/cruised.lpc（1 处）、/adm/daemons/v2.lpc（1 处）、/adm/daemons/versiond.lpc（1 处）、/include/getconfig.h（1 处）。

NPC 的 chat() 在 chat_msg 为空数组时下标越界（2026-10-21，KB 06 §7.334）

NPC 基类的 chat() 是 if (arrayp(msg = query(...))) { if (random(100) < chance) { rnd = random(sizeof(msg)); if (stringp(msg[rnd])) ...，不检查列表是否为空。本库现有的 NPC 文件里没有字面写成空 chat_msg 的（扫描只认 set("chat_msg", ({ }))）；守卫挡的是运行时才为空的列表和以后新增的 NPC。在 random(100) < chance 前面加 sizeof(msg) &&（scripts/lpc_fix_chat_empty.py；if (random(100) > chance) return 0; 的写法是 !sizeof(msg) ||）。改动：/inherit/char/npc.lpc（共 1 个文件）。

存档里的 .c 文件名改成 .lpc（2026-10-21，KB 03 §4.2 第 9 项）

存档里写着 "/d/city/npc/wei.c" 这样的文件名；extract 时源文件已经全部改成 .lpc，而驱动对带扩展名的路径只认字面，new() / load_object() / move() 都找不到它，调用方多半把失败当成「没有这个东西」悄悄跳过。本库 1 个存档共 1 处按 scripts/lpc_fix_save_dotc.py 改成 .lpc（只改 .lpc 存在而 .c 不存在的字符串；玩家存档、备份、留言板、巫师目录和文件索引缓存不动）。改动：/data/dbased.o（1 处）（共 1 个存档）。

注册到不存在函数的 add_action（2026-10-21，KB 06 §7.349）

add_action("函数", "动词") 绑定的函数，在对象本身、它包含的头文件、它继承的父类和继承它的子类里都没有定义（scripts/lpc_scan_callout_targets.py 静态解析，子类用传递闭包复核，scripts/lpc_verify_handlers.py 运行时 function_exists() 复核）。驱动从玩家最新的一条动作往回找，遇到这样的句子就报 *Function for verb 'drop' not found.，同一个动词的其他处理函数不再尝试（不是最新的一条就悄悄跳过）。删除了 4 条注册（3 个文件，scripts/lpc_fix_dead_add_action.py），动词留给别的处理函数：/adm/npc/referee.lpc（do_halt）、/clone/megazine/obj/maikaren.lpc（do_drop）、/inherit/misc/fboard.lpc（do_list）。最多的函数名：do_halt 2 处、do_drop 1 处、do_list 1 处。

不带前导斜杠的物品路径：缺失物品的 new() / carry_object() 调用链加了保护（2026-10-22，KB 06 §7.73 补充）

carry_object("d/city/obj/jiasha")->wear(); 这类路径不带前导斜杠的写法在这个驱动里按根目录解析（new("d/city/obj/changjian") 能加载 /d/city/obj/changjian），所以和带斜杠的一样：文件不在压缩包里时 new() 返回 0，->wear() 在 0 上报 Bad argument 1 to EFUN call_other()，整个 NPC 加载不了。此前的保护工具把这种写法当作「相对当前目录，没法判断」跳过了。1 个文件里 1 处（d/city/obj/cloth）改成带保护的块（scripts/lpc_guard_carry_object.py）。 对这 1 个改动文件做加载检查（HEAD 与工作区对比）：HEAD 能加载 0 个，现在 0 个（新增可加载 0 个，回归 0 个）。

汉字按字节切片的老写法没有换成按字符：改法名 "澄" + name[2..3]、辈分判断 name[0..1] == "澄" 等 4 处（2026-10-22，KB 06 §7.360）

原档按字节算字符串区间（一个汉字 2 字节）：name[2..3] 是名字的第二个字，name[0..1] == "澄" 问名字是不是以「澄」开头。转成 UTF-8 之后，这个驱动的区间和 strlen() 都按字符算（strlen("澄观") 是 2，"澄观"[0..1] 是两个字），老算术原样留着、意思全变：各派掌门的辈分判断（name[0..1] == "澄"）永远不成立；NPC 的称呼和提示里取到两个字而不是一个字。

本库 4 处（2 个文件）：与汉字常量比较 3 处、to_chinese()[..] 1 处。例：chk[0..1] == "你" → chk[0..0] == "你"（/adm/npc/ganjiang.lpc:542）；to_chinese(martial)[4..7] → to_chinese(martial)[2..3]（/cmds/skill/perform.lpc:119）。scripts/lpc_fix_gbk_slice.py 把下标按字符重算（社区的 UTF-8 版 xkx100utf8 里就是同样的写法，作对照）；没有改的：截断（name = name[0..7]，现在最多 8 个字，只是放宽）、ASCII 的区间（fname[0..3] == "room"、ctime(time())[4..19]）和上下文看不出是汉字的区间。证明（scripts/lpcc_eval.py，bixiecanyang；其余各库是同一份文本）：让澄字辈掌门 attempt_apprentice() 一个名叫「张三」的访客，HEAD 的法名是 慧，改后是 慧三；抽样 28 个库的 HEAD / 工作树加载对比 0 退步。工具不幂等（[0..3] 折半得 [0..1]，又是对齐区间），改过的文件记在 scripts/gbk_slice_converted.txt，不要重跑。

长度检查按字节写，旁边的提示按汉字写：2 处（2026-10-22，KB 06 §7.362）

老档的 strlen() 数字节（一个汉字 2 字节），转成 UTF-8 后它数字符：提示写「必须是 2 到 6 个中文字」的检查还是 strlen(name) < 4 || strlen(name) > 12，于是两三个字的房名、帮会名、群名、聊天室名、称号被提示自己说允许的那句话拒绝，上限却是提示的两倍；留言板的标题检查 i < 4 || i > 24 || i % 2 还会拒绝所有奇数个字的标题（post 你好吗）。本库 2 处（1 个文件）：if ((strlen(name) < 4) || (strlen(name) > 12)) {（/adm/npc/luban.lpc:1764）。scripts/lpc_fix_strlen_gate.py 以提示为准：条件里每个 strlen 门槛都恰是提示所说数字的两倍才折半，存长度的变量（i = strlen(arg);）的门槛和 || i % 2 奇偶项、其后循环的 i % 2 == 0 && 一并处理；提示里有英文 / 字母 / 字节的不动。证明（scripts/lpcc_eval.py，bxsj 的留言板，求值对象继承 NPC 并重写 edit() 记录调用，command("post 标题")）：HEAD 拒绝「你好」和「你好吗」、放行 23 个字的标题；改后接受前两个、拒绝第三个；shenmo 的 named.lpc invalid_new_name("张三")：HEAD 回 名字只能是2-4个中文字。，改后回 0。抽样 24 + 15 个库的 HEAD / 工作树加载对比 0 退步。

按字节写的汉字循环、茶馆猜诗和同盟频道标题（2026-10-23，KB 06 §7.363）

老档按字节做「一个汉字」：while (i--) if (i % 2 == 0 && !is_chinese(name[i..<0])) 只检查偶数下标，奇数位上的拉丁字母（张a）会过；茶馆猜诗 mixup() 用 strlen(str)/2 再 j=j+j、str[k..k+1] 交换两字节；泥潭系 channeld 的 switch (strlen(fname)) case 4/6/8/10 和 strlen(family)==4/6 按字节长度给同盟 / 门派名加空格。本驱动的 strlen() 和区间都按字符算。scripts/lpc_fix_byte_char.py：去掉奇偶项、把区间改成 name[i..i]，按字符重写 mixup() 并把 strlen(first)>=14 折成 7，频道标题的 case / 门派长度改成字符数（对照 aoxiangtianji）。本库 channeld 同盟 / 门派标题 1 个文件。证明（scripts/lpcc_eval.py）：sjplii 的 check_legal_name 在 HEAD 接受「张a」、改后拒绝，两边都接受「张三」；xkx100 的 clubpoem->mixup("离离原上草") 得到「离离原草上」（真交换汉字）；nitan_san 两字同盟名得到 【 兄 弟 】。关键对象加载通过。

血条 / 表情标题按字节算显示宽度（2026-10-23，KB 06 §7.365）

老档把汉字当两字节：bar_string[0..(val * N / max) * 2 - 1] 把「格数」换成字节下标，半格血条在本驱动（按字符算区间）上会整条画满；巫师表情 edemote 的 repeat_string("─", 13 - (strlen(emotion) / 2)) 也会少减汉字宽度。scripts/lpc_fix_display_pad.py：去掉 CJK 血条 / 空白条上的 * 2（jings/qis 赋值和 blank_string3[0..(23-a)] 的 23→11 一并改），表情标题去掉 / 2；ASCII 空格的 none_string / empty_string 仍保留 * 2（一格两个空格）。本库 2 处（1 个文件）。证明（lpcc_eval.py hhsj）：tribar_graph(6,9,12) 在 HEAD 画出整条 12 个■，改后为 6 个■+3 个□；麻将 strlen(tempB)/2 是两字符牌码，未改。

中文名 / 验证码仍按 GBK 取「两字节一字」（2026-10-23，KB 06 §7.373）

pname = pname[(random(sizeof(pname)) & 0xFFFE)..<1]; pname = pname[0..1]; 在按字符计下标时会取出两个汉字。scripts/lpc_fix_pname_slice.py 去掉掩码并把收尾改成 [0..0]（pname/pname2/sec_name/str/str1）。本库 4 处。证明：nt1 生成名长度 2–4；HEAD 常见 3–5。

深度功能测试（2026-10-27，§6.10 未知 pragma save_binary/optimize）

删除本库中 FluffOS 不认的 #pragma save_binary / #pragma optimize… 行。