Journey to the West 2006 (AKAI Studio / Tomud)

WASM 修复摘要（迁移自 meta.json 的 group_note）

西游记2006（AKAI Studio 出品的、西游记/ES II 家族的 Tomud 品牌分支）。需要在正常的 id/密码流程之前先完成一次自定义的客户端版本握手：服务器在横幅之后发送的第一行是裸的 'version ' 提示，期待字面回传 '2060'（Tomud/笑傲江湖客户端的自动应答）；任何其它第一行回复都会得到"your client isn't Tomud..."并立刻断线——这不是 bug，任何测试客户端只需要把 '2060' 当作第一行发送即可。WASM 修复了 5 个 bug：（1）四处绝对路径尖括号 #include 在这个驱动的 include 目录规则下永远解析不了，尽管目标档案确实存在（d/obj/misc/jitan.h 的 </d/city/misc/banned.h>，d/obj/books-nonskill/book-qujing.lpc 的 </d/qujing/obstacle.h>，加上 d/qujing/wudidong/monk1.lpc 和 u/vikee/dntg10/sky/tongmingdian.lpc）——已改成引号写法。（2）标准的 §6.6 convertd.lpc 希腊字母/符号表反斜杠错字（'a\",' 应为 'a",'），共 45 处，在真正生效的 adm/daemons/convertd.lpc 和一份未被引用的死代码副本 daemons/convertd.lpc 里都有（两份都做了修复，保持树内一致）。（3）adm/simul_efun/message.lpc 的 tell_room() 把一个未设置的裸 varargs 参数 'exclude' 直接转发进 message() 第 4 个参数（§7.12），导致第一次只带 2 个参数呼叫 tell_room()（某个预加载房间自己的心跳/sounding()）就报"Bad argument 4 to EFUN message()"崩溃——已用 'exclude || ({})' 修复。（4）两个原始批量 GBK→UTF8 转换那一遍漏掉的残留原始 GBK 字节内容档案（d/sea/npc/beast1.lpc、d/obj/quest/shuijingqiu.lpc）——已用 iconv -c -f GB18030 -t UTF-8 重新转换（用 -c 丢弃了少量真正损坏的尾部字节，只是外观内容的轻微损失）。管理员账号播种：fluffos (admin) 加入 adm/etc/wizlist（securityd.lpc 真的会在开机时读取 WIZLIST）。注册流程在一次连续的 WASM 客户端会话里完整验证过：版本握手（2060）→new→英文 id→中文名字→管理员密码+确认→登录密码+确认→电子邮件→性别（m/f）→属性分配菜单（9 接受默认值，y 确认）→在南城客栈进入游戏世界，带完整房间描述，look/score 都干净，quit 被游戏自身"新账号首次存档前需游玩满 1 小时"的规则正确拦下（不是 bug）。管理员权限已直接通过登录时"目前权限：(admin)"确认。LPC 格式化工具对全部 9264 多个档案运行（写入 9118 个，9 个转档之前就存在的未结束字符串/文本块内容错误未做格式化，137 个未改动）。没有 :: 父类呼叫拆分命中，没有 CJK 重新加空格命中；case 标签带尾随注释的盲点找到了几处命中（combatd.lpc 3 份副本、ftpd.lpc 3 份副本），但逐一 diff 复核确认格式化工具正确保留了后面的每一条语句。格式化后用同样的完整注册流程重新验证过——干净，管理员权限依然是 (admin)。追加修复：scripts/scan_known_bugs.py（新写的静态扫描工具）标记出真正生效的 adm/daemons/logind.lpc 的 check_legal_name() 仍带有标准 §8.1 的 i%2 奇偶门槛加 [i..<0] 后缀切片写法（这里之所以无害，只是因为这个家族的 is_chinese() 只检查第一个字符，但还是改成了标准的逐码点写法）以及 6 处 is_killing(me) 呼叫点（§7.50），分布在 daemon/class/*.lpc 的 kungfu 技能档案和 lestat/old/roar.lpc 里——已改成 is_killing(me->query("id"))。logind.lpc 有 5 份死代码副本（www/relative/、u/canoe/、u/canoe/program/、u/vikee/、daemons/）依然显示奇偶门槛模式，但从未被 LOGIN_D 加载，保持原样。httpd.lpc/ftpd.lpc/dns_master.lpc 不在真正生效的预加载列表里（休眠状态，只能通过巫师手动加载触达）——没有掏空，标记为已知的潜伏风险。修复后重新验证干净。

§7.86 跨库扫描修复（留言板 post 崩溃）

• BBS_BOARD、BULLETIN_BOARD inherit + 多余 replace_program() 致命形状（AGENTS.md §7.86，post 命令崩溃）：全档案 36 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 sweep (2026-08-19)

Fixed the corpus-wide inherit ROOM; ... replace_program(ROOM); redundant-replace bug (AGENTS.md §7.100). 235 live occurrences deleted: 234 via scripted sweep (fix_710_room.py), plus 1 hand-fixed roommaker-tool template (obj/roommaker.lpc, simple string-builder variant, same shape as near-identical sibling xyj2006zzzhx). 6 already-commented-out instances left untouched. No real .lpc source found under work/data/. Verified via build-debug driver boot: clean compile, zero new "cannot replace"/"cannot bind" debug.log lines; confirmed serving via raw-socket connect showing the real login banner on port 40157.

§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.

Deep functional test (round two, 2026-08-27)

First full §10.7 round-two live playthrough of this lib (previous work on it was WASM-onboarding + mechanical corpus sweeps only, never a continuous played session). Booted ~/src/fluffos/build-debug/src/driver config.fluffos natively on port 40157 and drove it with raw Python socket scripts (client sends the literal 2060 version-handshake line first, per this lib's own Tomud-client-detection gate — confirmed not a bug, see the WASM summary above).

Admin account had to be re-seeded live. adm/etc/wizlist already listed fluffos (admin), but no corresponding save file (data/{login,user}/f/fluffos.o) existed on disk — the account was never actually registered through the login flow, only added to the wizlist text file, so a login attempt as fluffos got "没有这个玩家" (no such player). Registered it live: English id fluffos, Chinese name 管理员, admin/management password Mud@2026 (the project's standard password IS accepted for this field), login password Play2026x (the login password is REQUIRED to differ from the admin password — the registration flow rejects a login password equal to the admin password with "您的登陆密码不能和管理密码一样"). Confirmed (admin) privilege via the login banner. This is now a real, working seeded admin account (previously only a wizlist-only phantom entry).

Bugs found and fixed (2, both genuine programming bugs — missing guards causing real crashes, not content/design):

1. adm/simul_efun/file.lpc's log_file() writes into /log/nosave/, a directory this archive never shipped, with no assure_file() guard — crashing every wizard call/clone/purge/ purgehouse/xpass/superpass command AND the player-facing suicide -f account-deletion command the welcome banner itself advertises (AGENTS.md §7.11, same shape as the xajhxo/ nitan_ceshi instances). Reproduced live: admin call ceshiwyz->add("money",1000) (a normal wizard debug command) threw an uncaught 执行时段错误：*Wrong permissions for opening file /log/nosave/CALL_PLAYER for append. "No such file or directory" straight to the admin's screen, aborting the whole call before the requested side effect ran. Since cmds/usr/suicide.lpc's real account-deletion path (slow_suicide()) calls the identical log_file("nosave/SUICIDE", ...) line BEFORE any of the actual deletion logic, every real suicide -f on this lib would have hit the same uncaught error and aborted before deleting anything — despite the rules banner telling players to use exactly this command to clean up unused accounts. This file already carries its own assure_file(string file) helper (mkdir-p's every path component), just never wired into log_file(), and defined textually AFTER log_file() in the same file (needs a forward declaration on this compiler). Fix: ```lpc void assure_file(string file); // added forward decl

void log_file(string file, string text) { assure_file(LOG_DIR + file); // added guard write_file(LOG_DIR + file, text); } `` Verified live post-fix: the identical call command now completes cleanly (/obj/user#206 ("云游生")->add("money", 1000) = 1000, no error), and /log/nosave/CALL_PLAYER now exists with the expected audit-log line. This is the project-wide simul_efun log_file(), so the fix transparently covers every one of the ~20 call sites across cmds/wiz/call.lpc, cmds/wiz/clone.lpc, cmds/arch/purge.lpc, cmds/arch/purgehouse.lpc, cmds/adm/xpass.lpc, cmds/adm/superpass.lpc, cmds/usr/suicide.lpc, and others that write into nosave/...` or other never-shipped log subdirectories.

2. adm/obj/master.lpc's log_error() broadcasts every plain compile WARNING (not just genuine errors) straight to any connected player's screen (AGENTS.md §7.103, exact same shape/fix as zzfy3's own instance — this lib's own master.lpc still had the unguarded original). Reproduced live: registering a fresh character and typing ordinary early commands (look/score/i) each triggered several 编译时段错误：...warning: Unused local variable 'x'-style dumps as the corresponding command file (cmds/std/look.lpc, cmds/std/score.lpc, cmds/usr/inventory.lpc) got lazily compiled for the first time that boot. Fix: if (this_player(1)) ... → if (this_player(1) && strsrch(message, "warning:") == -1) ..., identical to the established corpus fix. Verified live: post-fix, the same register→look→score→i sequence produced zero warning dumps.

Cross-cutting pattern sweep (all the standing AGENTS.md patterns listed in the round-two task brief, checked via targeted grep against work/, excluding the already-documented dead-code trees daemons/, www/relative/, u/canoe/, u/vikee/): §4.3 (no log_file() call site has "static" in its literal path argument), §4.4 (no remaining non-UTF-8 byte sequences in any .lpc file — the two known post-WASM GBK stragglers were already reconverted per the summary above), §7.86/ §7.100 (already swept corpus-wide, re-confirmed no live residual replace_program()-on-top-of-inherit hits), §7.121 (no int-typed field/parameter assigned a bare 0.xx float literal), §7.123 (no column-zero bare IDENT = (\[|\{ file-scope statement outside a function), §7.126 (no load_object()/new() call fed a string pulled from a dbase mapping — this lib's saved data uses literal, already- .lpc-safe paths, not a macro-placeholder area engine), §7.129 (already fixed at WASM-onboarding time — exclude || ({}) guard confirmed still present in the live adm/simul_efun/message.lpc), §7.131/§7.152 (find_living/find_player work correctly — this lib DOES call set_living_name(), confirmed live via a cross-connection tell reaching the test character both before and after a reconnect), §7.133 (net_dead() is defined and reachable — obj/user.lpc/ obj/user1.lpc/obj/login.lpc all define it directly, not a classic- driver remove_interactive()-on-master shape), §7.148 (no parameter named a reserved keyword like nosave), §7.153 (feature/more.lpc's pagination switch correctly covers "b"/"q"/default with no missing else; confirmed live via a real multi-page help menpai — q correctly aborted mid-page). No new instances found for any of these.

Live playthrough (real Chinese-named test character 云游生 /Ceshiwyz, deleted before commit — only the seeded admin account remains): version handshake → new → English id (letters only, 3-8 chars, no digits — confirmed by trial) → Chinese name (2-6 Chinese characters) → admin password → confirm → login password (must differ from admin password) → confirm → email → gender → stat-allocation menu (9/y for defaults) → arrived at 南城客栈 with a full room description; look/score/i all clean. Skill acquisition: the starting-room NPC 阿凯哥哥 (Akaigege) grants one full free skill set via ask akaigege about <menpai-keyword> (e.g. fangcun), gated by a yudian/skills == "got" flag that correctly blocks a second grant ("你不是已经要过功夫了吗？") — skills correctly showed all 13 granted skills afterward, score showed the resulting attribute increases. Safe-sparring mechanism: fight <npc> (documented in help fight as "点到为止，不会真的受伤") — used against 疥顶小僧 in 朱雀大街, fought to a clean loss (这场比试算我输了，佩服，佩服！), left "受伤" (wounded, a recoverable condition) but alive and not ejected from the room. Shop: list/buy at 南城客栈 dispatch correctly — list shows the real priced inventory, buy huasheng from xiao er on a broke fresh character correctly returns "你的钱不够" rather than silently no-op'ing or crashing (no §7.143/§7.151 shape found); currency here is physical money-objects (gold_money/silver_money/coin_money carried in inventory, not a simple int stat), so a full paid purchase wasn't exercised live. Board: read board on the (empty) 南城客栈留言板 correctly reports no messages, no crash.

Quit/reconnect: the "new account must be connected 1+ hour before it can be saved" gate (cmds/usr/quit.lpc, time() - birthday < 3600) is confirmed intentional design (also documented at WASM-onboarding time) — bypassed for testing purposes only by editing the test character's own birthday field backward in its .o save file while the driver was stopped (never done to the real admin account). With that bypassed, did TWO full quit→reconnect cycles: quit produced a clean "欢迎下次再来！" both times with zero new log/log error lines either time; a second connection (the seeded admin) could tell ceshiwyz ... and have it delivered correctly both before and after each reconnect (confirms find_player() + §7.152-style post-reconnect set_living_name() re-registration both work), and the room listing never showed more than one copy of either character at a time (no §7.150-style leaked duplicate). Non-autoload- flagged carried items (the starting 粗布衣 clothing) are dropped/ destroyed on every quit and freshly reissued by restore_autoload() at next login — traced this to feature/autoload.lpc's intentional "only autoload-flagged items persist across a session, everything else is graceful-dropped rather than silently vanishing on destruct()" design (the same TMI-2-lineage save_autoload()/restore_autoload() idiom documented corpus-wide in AGENTS.md §7.122's family) — NOT a bug, left untouched.

Not reached this session (budget, not correctness concerns): a full paid shop purchase (no free starting currency reachable without a wizard-granted add(), and the currency is object-based rather than a simple stat), sect/menpai formal apprentice/bai enrollment beyond the one-shot NPC skill grant, real PK/death-and-respawn (would require finding a genuinely lethal opponent and burning significant real time), and the bang/economy-heavy bank/pawnshop flows. None of these showed any red flag during code reading; flagging honestly as unexercised rather than guessing them clean.

Verified grep -h '"port"' libs/*/meta.json | grep -oE '[0-9]{5}' | sort -n | uniq -c | awk '$1>1' prints nothing before committing. Driver killed by exact PID after testing; test character's save files deleted, only the seeded fluffos admin account save files remain under work/data/{login,user}/f/.

Targeted follow-up (AGENTS.md §7.154 sibling check): xyj2006zzzhx's NOTES.md (round-two deep test) had explicitly flagged xyj2006n's cmds/std/look.lpc:410 as byte-for-byte the same unfixed present("fire", this_player())-with-no-null-check line documented in AGENTS.md §7.154. Confirmed identical here: feature/move.lpc::remove() (called on every destruct()) calls LOOK_CMD->look_room(me, ob, query("env/brief")) whenever an object leaves a non-living, non- character environment (line 283), and look_room()'s outdoor exit- listing block unconditionally evaluated present("fire", this_player()) before checking whether this_player() was even an object. Reproduced live on a fresh boot two independent ways: (1) the real production path — waited for adm/daemons/natured.lpc's boot-persistent update_day_phase() call_out chain to reach its event_dawn() phase, which destruct()s the outdoor 钟馗(Zhong kui)/ 青霞仙子(Qingxia xianzi) NPCs spawned at midnight; debug.log recorded *Bad argument 2 to present() Expected: object Got: 0 at cmds/std/look.lpc:410 via natured.lpc:event_dawn()→destruct()→move.lpc:remove()→ look_room(), exactly matching the zzzhx trace shape (and a bonus third instance via /std/room.lpc::reset() destructing a decayed corpse the same way); (2) a synthetic set_heart_beat(1) harness object (heart_beat calls are a reliable way to get this_player()==0 outside a live command, confirmed via log_file()) that cloned a corpse into an outdoor room and called decay(3) from heart_beat() — crashed identically. Applied the identical fix from xyj2006zzzhx: added an objectp(this_player()) short-circuit guard before the present()/wizardp() calls. Rebooted fresh and reran the same set_heart_beat() harness (confirmed this_player()==0 at the decay(3) call site via the same log_file() check) — zero present() crashes in debug.log afterward; the torch/wizard check is now cleanly skipped instead of throwing when there's no player in context. Test harness file and its log were removed before commit; only cmds/std/look.lpc changed.

AGENTS.md §7.19 fix: enable_player() reentrancy from init()

feature/command.lpc's enable_player() (wrapper around enable_commands()) was reachable from an NPC's init() (via the shared std/char.lpc setup() chain), and enable_commands() is only safe to call from create() -- calling it again on an object already living() makes the driver re-invoke that same object's init() as a side effect, which recursed back into enable_player() on the same call stack until "Too deep recursion" aborted the boot on a room's first-ever visit. Fixed with a true reentrancy flag (in_enable_player_now, set at entry, cleared before every return), NOT a living()-gated guard -- disable_player() in the same file legitimately re-calls enable_commands() while already living() (sleep/wakeup via cmds/std/sleep.lpc, revive via feature/damage.lpc), which a living() guard would silently break. Verified via lpcc --batch single-file compile check (PASS). Part of the corpus-wide §7.19 sweep (Batch C).

商店付费购买 + 将军府拜师（2026-09-04 librarian shop slice）

2026-08-27 深度测试只验证了穷号 buy huasheng from xiao er 回「你的钱不够」， 正式 bai 未走到。本轮补完付费购买和将军府拜师，并确认 quit→重连后师承仍在。

第一行必须发字面 2060（Tomud 握手，get_id 拒其它）。登录密码 Play2026x，管理密码 Mud@2026。if_young("no") 自动走，没有 Y/N 提示。 落地南城客栈 /d/city/kezhan，list 显示花生豆 二十文钱。

Admin fluffos：clone /obj/money/gold 得到一两黄金，然后 buy huasheng from xiao er live 回「你向店小二买下一碟花生豆」。 i 为九十九两银子 + 八十文钱 + 花生豆。没有崩溃，没有「钱不够」误报。

新手拜师：goto /d/jjf/jjf_bingqi，bai qin ping（秦平，管家，无经验门槛） → 将军府第四代弟子，师承秦平。秦琼要求 combat_exp >= 100000： call me->set("combat_exp", 100000)（本库 eval 只 force_me 一条命令， 不是 LPC；用 call），goto /d/jjf/keting，bai qin qiong → 将军府第三代弟子，师承秦琼。quit 后重连，score 仍是师承秦琼 / 第三代弟子。

本机临时建了空 /u/f/fluffos/（不入库）。驱动 PID 已杀，端口 40157 已空。