xiakexing100 — 侠客行一百

Archive: 侠客行100.rar. Port: 40037. Status: done (boots clean and quick, full registration flow verified end-to-end including a real Chinese name).

What this is

"侠客行一百", credited to "Coollizard & xkx100 2003-2004", http://yoyo.xm.fj.cn/xkx100, up since 2001-01-14 per its own connection banner. Same "Century-family" adm/single/{master,simul_efun} layout as shiji/zhonghua2/shujian2008/etc. ~14,227 raw files, 12,209 after .c→.lpc rename — the largest lib processed in this session run.

Fixes applied

1. AGENTS.md §15h, standard shape: is_chinese()'s GBK lead-byte check → CJK codepoint check; check_legal_name()'s byte-count bound < 4 || > 10 → character-count bound < 2 || > 5, removed the i%2==0 even-byte-offset gate. 2. Confirmed NOT needed, via source reading: no §4 fix (master.c's valid_write gates via find_object only, valid_read is a bare return 1;); no deep named.lpc fix (named.c exists but lacks the PATH()/sliding-window pattern seen in the nitan-family shape — a simpler, different implementation here). 3. AGENTS.md §15p: /adm/daemons/network/dns_master was in adm/etc/preload — removed proactively before the first boot attempt. Booted clean in under 20 seconds, zero compile errors, no hang.

Interactive test result — full registration flow

No hidden pre-id prompts. Verified the complete registration path in one continuous connection: id xkxbtest → confirm y → real Chinese name 秦风 (avoiding "韦小宝"/Wei Xiaobao — a famous Jin Yong character found in this lib's banned_name list) → accepted, proceeds straight to "请设定您的密码：".

lpcc sweep

12,209 files, 12,118 pass / 91 fail (99.3%). Failure tail dominated by an accept_hit type-mismatch cluster (120 of the reported instances across files, string vs object argument), plus the usual handful of missing globals/syntax typos — not triaged individually per AGENTS.md §6b/§13. Memory stayed comfortably healthy throughout the sweep despite this being the largest lib in the current session run (~13GB free consistently, lpcc RSS peaked well within normal range, no pressure requiring an early kill).

Re-verification pass (driver rebuild + LPC formatter + WASM build)

• Reformatted all 12209 .lpc files under work/ with tools/lpc-syntax/format-corpus.mjs: 10795 written, 1414 already idempotent-clean, 0 refused. Verified the earlier §15ae fix (feature/command.lpc's command_hook staying nomask with private still commented out) survived unchanged.
• Native retest against the freshly-rebuilt driver (~/src/fluffos/build-debug/src/driver, rebuilt from latest upstream master): clean boot, zero fatal errors in log/debug.log. Full registration flow re-verified with a fresh real Chinese name (秦墨/id xkxbrew, then 秦徽/id xkxbrev), reaching an actual starting room, look/score/quit all producing correct output — also separately re-confirmed post-login command dispatch (the §15ae fix) via a reconnect + look on the saved character. No regressions from either the driver rebuild or the reformat.
• WASM build test found and fixed a real, pre-existing mudlib bug (scripts/wasm_client.js against ~/src/fluffos/build-wasm/src): boot itself was clean, but every connection attempt silently stalled right after the login banner, never reaching the id prompt at all — no fatal error printed anywhere. Root-caused by reproducing the exact same stall natively (temporarily moving aside work/log/MUDVISITOR, the untracked runtime-state visitor-counter file adm/daemons/logind.lpc's howmany_visitor() reads on every connection): read_file(VISITORS) returns a non-string (0) when the file doesn't exist, and the un-guarded sscanf(content, "%s %d", ...) that follows throws Bad argument 1 to sscanf — uncaught, this aborts the rest of logon() silently (confirmed via debug.log's 执行时段错误：*Bad argument 1 to sscanf), before the player ever sees an id prompt. This is a genuine pre-existing mudlib bug, not a WASM-specific driver gap — it just never manifested in earlier native testing passes because MUDVISITOR already existed on disk from prior test runs (it's untracked/gitignored runtime state, not shipped in the original archive's git history), whereas scripts/wasm_client.js's in-memory filesystem deliberately never copies log/'s *contents* (only creates the empty directory, by design, to keep the harness fast), so a WASM run is effectively always a "genuinely fresh archive, first boot ever" scenario for this file — the exact condition that triggers the bug. Fixed: added a stringp(content) guard (return 0/treat as "no visitors yet" if the file doesn't exist) to both howmany_visitor() and the identically-shaped howmany_card() (GIFTCARD file, same pattern, fixed proactively). Re-verified: (a) natively, with MUDVISITOR moved aside, connection now proceeds straight to the id prompt showing "第一位访问者" and a full registration completes cleanly, zero debug.log errors; (b) under WASM, full registration now also completes end-to-end (id xkxwasmz → real Chinese name 秦岫 → stat roll → gift/email/gender → landed in an actual starting room, look/score/quit all producing correct output). This lib has no IP-format-dependent login gate, so once this fix was in place it was fully playable under WASM too.

WASM-enablement pass (loopback-allow / admin seeding)

Applied the four standard WASM-first changes (AGENTS.md §1.3b/§1.3e/§1.5).

1. Loopback always allowed — ban list + per-IP throttles (adm/daemons/logind.lpc:73-124)

logon() ran BAN_D->is_banned(), an in_login > 10 per-IP in-progress-login throttle, and BAN_D->is_multi() (per-IP character cap) unconditionally. Added an is_local local computed once and gated all three on !is_local.

Written fail-closed from the start (2026-07-24): is_local = (stringp(lip) && (lip == "127.0.0.1" || strsrch(lip, "127.") == 0)) — only a genuine loopback-shaped string is treated as local; an unparseable/non-string ip is NOT local and still goes through the ban list and both throttles. (An earlier partial pass on this file had used the fail-open shape seen on this batch's other four libs — !stringp(lip) || lip == "127.0.0.1" || strsrch(lip, "127.") == 0 || sscanf(lip, "%*d.%*d.%*d.%*d") != 4, which let ANY malformed ip bypass every gate — but that was corrected to the fail-closed form above before this pass was considered complete; see the sibling libs' NOTES.md entries for the full rationale: the old WASM driver bug that produced garbage query_ip_number() output is now fixed, so there is no reason left to treat "can't parse it" as "must be loopback".)

2. Uptime startup gates — none present

No uptime() < N connection-rejection gate exists in this lib. All uptime() hits are in-game content: gambling rooms (d/*/duchang*.lpc, npc/douji, npc/saigui), NPC/room respawn timers, cooldowns, and adm/daemons/autosaved.lpc's once-per-boot data-file initialization (if (uptime() < 400) { ...write_file... } — first-boot bootstrap of some /data/* flags, not a connection gate). Nothing to bypass.

3. Anti-flood throttles — covered by item 1

No separate registration-throttle daemon exists; the per-IP protection in this lib IS the in_login/BAN_D->is_multi() checks patched above, already loopback-exempt.

4. Admin account seeded — fluffos / Mud@2026 / 浮浮

Registered through the real flow (id fluffos → y → 浮浮 → test1234/confirm → talent 1/20/20/20/y → personality 0 → email → m), landed in a random starting room. Granted (admin) by appending fluffos (admin) to adm/etc/wizlist (securityd.lpc's get_status() reads this file into wiz_status). Verified: login as fluffos shows 目前权限：(admin), update /adm/daemons/logind → 重新编译 /adm/daemons/logind.lpc：成功！.

Save files for the orchestrator to force-add (untracked, not gitignored):

• libs/xiakexing100/work/data/user/f/fluffos.o
• libs/xiakexing100/work/data/login/f/fluffos.o

Bonus fix: quit/exit was silently broken for EVERY player (pre-existing, not WASM-related)

While verifying the admin account's quit, discovered cmds/usr/exit.lpc's savequit() never actually let anyone quit: int a = 1; is reassigned inside the if (link_ob) {...} branch to a = (link_ob->save()), and link_ob (me->query_temp("link_ob")) is the /clone/user/login connection object set by enter_world()/reconnect() on every login, not just a rare possession/reconnect edge case. FluffOS's save_object() efun returns the byte-size of the text written on success (confirmed by reading ~/src/fluffos/src/vm/internal/base/object.cc's save_object_recurse(), which accumulates and returns textsize), not a 0/1 boolean — so a ends up as some multi-hundred-byte count, and the final check if (a == 1 && me->save()) was false for virtually every quit, for every character, admin or not. The player saw all the normal quit text ("你共停留了...欢迎下次再来！") followed immediately by a scary "警告：档案保存失败，无法退出游戏，请马上联系巫师解决。" and was never actually disconnected (destruct(me) never reached). Confirmed present verbatim in the original raw archive (raw/xkx100/cmds/usr/exit.c), so this is a decades-old author typo/driver-assumption mismatch, not something introduced by this project's conversion or by the WASM pass. Fixed: if (a == 1 && me->save()) → if (a && me->save()) (test truthiness, matching the efun's actual 0=fail/nonzero=success contract). Verified with a fresh registration and with the fluffos admin account: quit now completes cleanly with no warning and no lingering interactive body.

Retest (2026-07-24)

Fresh boot, fresh registration (id qinlanx/qintestb, real Chinese names 秦岚新/秦测三) through look/score/quit — landed in a starting room, correct score output, clean quit with no warning. fluffos/Mud@2026 admin login verified: look then update /adm/daemons/logind → 重新编译 ... 成功！, then clean quit (the bonus fix above). Zero 执行时段错误 lines in debug.log across the whole session (the one hit from an unrelated missing-log/nosave/ directory surfaced only while probing the quit bug with the call wizard command directly on a userp target — log/ is fully gitignored/regenerated at boot, not a real defect worth chasing further here). Test characters removed afterward; fluffos kept. Reverted an incidental data/topten.o leaderboard diff caused by the test registrations (pre-existing tracked file, not meant to churn from QA runs).

深度功能测试 / Deep functional test (AGENTS.md §10.7, 2026-07-24)

One continuous native session (~/src/fluffos/build-debug/src/driver, port 40037), following the §10.7 checklist end to end: read help intro/help newbie/help newbie-basic/help newbie-shaolin1 first, registered a real Chinese name (qftester / 秦风, password test1234, born 嘉兴/group 3, talents 20/20/20/20, personality 0), landed at /d/jiaxing/yanyu ("烟雨楼"), look/score/i/hp all correct. Explored on foot toward 牛家村 reading room .lpc source (d/jiaxing/*.lpc) to navigate since exit lists intermittently hide under the lib's night-time darkness gate. Used the wizard spgo/ summon commands (via the seeded fluffos admin) purely to fast-travel the SAME test character to distant content (少林寺 练武场/广场) rather than trekking the whole map by foot — the actual sect-join, skill-learn, combat, and shop interactions below were all performed by the ordinary player character, not the admin.

Safe sparring: the newbie guide's suggested targets (a village 小孩/女孩) both declined fight — traced this to their attitude being genuinely "friendly" in their own create() (/d/village/npc/ kid.lpc, /d/village/npc/girl.lpc), which inherit/char/npc.lpc's accept_fight() is coded to always decline (case "friendly": ... return 0;) — working as designed, not a bug (a different, unrelated /d/jiaxing/npc/boy.lpc//d/village/npc/boy.lpc with attitude: "peaceful" momentarily looked like a mismatch until the actual live object path was confirmed via call obj->query(...) — red herring, not reused). Used the lib's own documented, code-confirmed-safe wooden-dummy mechanic instead (d/shaolin/obj/mu-ren.lpc, accept_fight() checks me->query("damaged")/a random(fight_times)>=10 break-chance, exactly the §10.7-item-3 shape): fight mu at /d/shaolin/wuchang1 exchanged several rounds of blows with correct combat text, halt ended it cleanly, hp/i afterward correct, zero debug.log errors.

Skill/sect join: followed help newbie-shaolin1's documented path — apprentice biqiu (清为比丘 at /d/shaolin/guangchang1w) immediately accepted as a 俗家弟子/少林派第四十一代弟子 (no manual ask ... about 出家/kneel needed for this particular monk); score showed 师父/称谓 updated correctly; skills biqiu listed the master's own skill roster; xue biqiu dodge 1 (organic NPC-teacher path) advanced 基本躲闪 from unlearned to 1/0, cha confirmed it. Both the sect-join and skill-learn steps worked correctly, no separate admin-command shortcut was tested (none is needed/expected in this lib — apprentice IS the join mechanism).

quit → debug.log → reconnect (§10.7 items 5, 8, 9): clean quit produced no warning (confirms the pre-existing savequit() fix from the "Bonus fix" section above still holds) and left zero new debug.log lines. Net-dead (dropped the connection without quit) and reconnected PROMPTLY (a few seconds later): worked correctly, position/state intact. Net-dead and NOT reconnecting until the real timeout elapsed found a new, live-reproduced bug — see below. Separately, clean quit followed by a genuine ~75-real-second wall-clock gap, then a FULL fresh login (not a net-dead reconnect) via qftester/test1234: correctly restored sect membership, skill, and all score fields from the on-disk save file (data/user/q/qftester.o — family mapping, skills/learned mappings, dbase.qi/jing etc. all intact).

Shop purchase (§10.7 item 6): character had no starting funds ("你没有存款"), so used the admin account to clone /clone/money/coin, call coin->set_amount(500), spgive qftester coin (funding only — the purchase itself was performed by the ordinary player character). list at 烟雨楼's 店小二 showed the menu correctly; buy jitui (30 文) succeeded, correctly made change into higher denominations (500 铜钱 → 1 烤鸡腿 + 4 两白银 + 70 文铜钱); eat jitui consumed it. No errors.

Not verified live: progression to a real death/respawn cycle — the session budget went to the net-dead bug investigation below instead; code review only (d/city/ghost*.lpc/鬼门关 revival flow referenced by help newbie-basic's〖死亡〗section was read but not exercised). A sect promotion (拜达摩/升辈) is a multi-hour undertaking per the newbie guide's own admission and was not attempted.

Bug found and fixed: tell_room() 2-arg wrapper crashes the net-dead force-quit handler — matches AGENTS.md §7.12 exactly (severity-escalated instance)

adm/simul_efun/message.lpc's tell_room() wrapper:

varargs void tell_room(mixed ob, string str, object *exclude) {
  if (ob) message("tell_room", str, ob, exclude);   // BEFORE
}

Every 2-argument call site leaves exclude as LPC's default int 0 (not an empty array), and the real message() efun's 4th argument rejects a bare 0. clone/user/user.lpc's user_dump() — the NET_DEAD_TIMEOUT-driven (900s / 15 real minutes, include/user.h) force-quit handler that runs when a net-dead player never reconnects — calls tell_room(environment(), ..., "\n") with only 2 args at its very first line (the DUMP_NET_DEAD case). This is byte-for-byte the same lib/function/case/timeout shape AGENTS.md §7.12 documents as "severity-escalated" from dtsl's deep functional test: the crash aborts user_dump() at that exact statement, so the two lines after it ("/cmds/usr/exit"->savequit(this_object())) never run — silently disabling the entire net-dead force-quit safety net: a player who net-deads and never manually reconnects would stay alive in server memory forever (never actually saved-and-removed via this path) until a driver restart.

Live-reproduced (not just pattern-matched): temporarily lowered NET_DEAD_TIMEOUT from 900 to 12 (include/user.h), ran the wizard update /clone/user/user command to recompile it, connected as qftester, dropped the connection without quit, waited past the shortened timeout:

执行时段错误：*Bad argument 4 to EFUN message()
Expected: object, array,  Got: int(0).
程式：/adm/single/simul_efun.lpc 第 77 行
呼叫来自：/clone/user/user.lpc 的 user_dump() 第 75 行，物件： /clone/user/user#105 ("秦风")
呼叫来自：/adm/single/simul_efun.lpc 的 tell_room() 第 77 行

Fixed (the standard AGENTS.md §7.12 wrapper fix, protects every 2-arg call site in the lib at once, not just this one):

varargs void tell_room(mixed ob, string str, object *exclude) {
  if (ob) message("tell_room", str, ob, exclude || ({}));   // AFTER
}

Re-verified with the same shortened-timeout methodology: no crash, debug.log clean, and who afterward showed only the admin online — qftester was correctly force-quit-and-saved this time (confirms the two previously-dead lines, including savequit(), now execute). Restored NET_DEAD_TIMEOUT to 900 and did a full clean driver restart before finishing (the test-only timeout value was never left in place).

Note for the project: two other confirmed "Century/adm-single" lineage siblings (AGENTS.md §11) — shiji and xjcq2000 — still carry the exact same unfixed 2-arg tell_room() wrapper in their own adm/simul_efun/message.lpc (checked via grep during this pass, not fixed — out of scope for this task, which is xiakexing100-only). shujian2008, xkxz2, and zhonghua2 (same family) already carry the fix from earlier passes. Worth a follow-up pass on shiji/ xjcq2000 specifically for this bug given xjcq2000 is also the lib AGENTS.md §10.8 documents hitting an unrelated driver-fatal crash ~25 minutes into its own net-dead soak — the two are different bugs but both live in the same net-dead code path family.

Observation, not fixed (design ambiguity, see §10.7's scope note): two independent day/night clocks can visibly contradict each other

cmds/std/look.lpc's exit-listing darkness gate (天色太黑了，你看不清明显的出路) keys off NATURE_D->game_time(), which derives an in-game shichen (時辰) hour from a heavily-compressed calendar clock (adm/daemons/natured.lpc: #define TIME_TICK ((time()%(60*24*3600))*365) — 60 mud-years pass every 60 real days, i.e. roughly 6 mud-hours per real minute, confirmed live: the session's in-game date advanced across several 时辰 within a couple of real minutes). The SEPARATE sun-position flavor line (NATURE_D->outdoor_room_description()) keys off current_day_phase, a nosave variable advanced by its own independent real-time call_out("update_day_phase", ...) cycle. These two clocks are not synchronized and were observed live disagreeing outright in the same room printout — "一轮火红的夕阳正徘徊在西方的地平线上"（sunset flavor text) immediately followed by "天色太黑了，你看不清明显的出路" (darkness gate, only fires for 亥/子/丑/寅时) moments apart, and at one point "太阳正高挂在西方的天空中"（afternoon flavor text) paired with the SAME darkness message. Purely cosmetic/exit-listing (movement in a known direction still works even when the list is hidden — confirmed by navigating through several "dark" rooms via directions read from source), and both clocks' formulas are pure LPC business logic with no encoding/efun/grammar involvement, so this is very plausibly authentic original-game behavior rather than anything this project's conversion introduced. Documented per §10.7's scope note rather than "fixed" by guessing which clock should govern the gate.

§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 70 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。

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。

发现并修复的 PROGRAMMING bug

1. log_file()（adm/simul_efun/file.lpc）完全没有 assure_file() 保护（AGENTS.md §7.11-class）：已加上前向声明 + assure_file(LOG_DIR + file);。

Proactive checks（无需改动）

• 实际生效的 master file 是 adm/single/master.lpc（config.fluffos 指定），其 log_error() 已经在更早一轮修复过（strsrch(message, "arning:") == -1 判断）。仓库里还有一份未使用的 adm/obj/master.lpc（没有任何档案 #include/引用它，也不是 config.fluffos 指定的 master file），其中的 log_error() 仍是 未修复的旧形状——确认它是死代码后未改动。
• win_times 的 %-operator 修复确认存在且正确： d/city/npc/jinyong.lpc:197、d/beijing/npc/refereew.lpc:177、 d/city/npc/gulong.lpc:246 均已用 to_int(query("win_times")) % 5。
• feature/dbase.lpc 未发现 tybxjh/wlhd 那种密码写保护，不适用。

实测过程

管理员 fluffos/Mud@2026（此前 WASM 阶段已注册并提交存档）用真 实密码重新连线，落地在此前保存的地点（车马店），score 正常显示 天赋/属性，quit 前的游戏内状态与存档一致。全程 debug.log 只有 正常的编译期提示（Unknown #pragma 等价的 In file included from 追踪行、未使用局部变量），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，未提 交无关变化。

AGENTS.md §7.100 修复（2026-08-19，批次三）

ROOM 基类冗余 replace_program(ROOM); 自崩溃地雷（详见 AGENTS.md §7.100）：4430 处命中，其中 d/huangshan/banshan.lpc 是本项目已见 过多次的"同一个 create() 里两处独立冗余调用"形状（setup() 前 后各一次），两处均删除。自带建房工具 clone/misc/roommaker.lpc 的 字符串拼接模板同样修复。

git diff --stat：4428 files changed, 1 insertion(+), 4430 deletions(-)， 与预期精确吻合。

验证：build-debug 驱动真实冷启动，端口 40037 正常监听。既有管理 员账号 fluffos（登录密码为 Mud@2026，与原始注册记录的 test1234 不同，二者都在 NOTES 中留档过）登录，目前权限：(admin) 确认，车马店→南大街→车马店多房间走访无误，quit 干净退出，全程 无新增 "cannot replace"/"cannot bind" 日志行。

§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): 4 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.