山海战神.rar → shzs

• Archive: archives/山海战神.rar (1.4MB, small simple/ES-II-derived MudOS 0.9.20 lib — "Final Frontier" themed base, reskinned).
• Mudlib root in archive: simple/ (extracted directly at archive top level, no extra nesting).
• Port: 40001.

Status: DONE — boots clean, playable over telnet

Login banner, Chinese prompts, and character-creation flow all render correctly. This lib was the pilot used to build scripts/convert_lib.sh and the fix catalog in AGENTS.md — see that file's "Common driver-compatibility issues" catalog (§1-§14) for the general write-up of every issue found here. Lib-specific specifics only, below.

What was fixed

1. Encoding: all .c/.h files GB18030→UTF-8 (774 converted, 124 already ASCII/UTF-8, 0 hard failures). Also converted data/*.o save files (dict/emote/daemon/ballot_box/stars — these are plain-text save_object() format, not binary) — data/emoted.o needed the lossy -c fallback (one stray invalid byte at position 25324, pre-existing data corruption in the original archive, unrelated to us). Also adm/etc/welcome (login banner) and config.fluffos itself (see AGENTS.md §5 for the double-corruption trap hit here on first attempt — fixed by re-copying fresh from raw/ and converting before any edits). 2. .c → .lpc rename: 855 files. Fixed 117 literal ".c" string references across .lpc files, plus 36 in include/globals.h (the F_DBASE macro — AGENTS.md §2, this is the one that cost the most time to find since it surfaced as a runtime "Inherited file does not exist" rather than a compile error). One orphaned non-LPC file caught by the blind rename (d/shenmin/shenminmap.c, ASCII-art map, unreferenced anywhere) — renamed to .txt instead (AGENTS.md §12). 3. static → nosave: 89 occurrences across the lib (AGENTS.md §3). 4. master.lpc's valid_read()/valid_write() had the lazy load_object(SECURITY_D) recursion bug (AGENTS.md §4) — added the re-entrancy-guard + catch() fix. 5. master.lpc's valid_override() was the old 2-arg signature, which wrongly rejects efun:: overrides written in #included simul_efun fragments (AGENTS.md §14) — added the 3rd main_file parameter. (Never actually manifested in a real boot, only in the lpcc sweep testing a fragment file as a standalone object — fixed anyway since it's free and more correct.) 6. master.lpc does NOT have get_root_uid() (driver printed a non-fatal warning and continued — did not need the AGENTS.md §7 stub fix here; this lib's exit(-1) path apparently doesn't fire the way lib #2's did. Left as-is since the lib boots and plays fine regardless). 7. Config: removed 5 obsolete keys (address server ip/port, binary directory, swap file, reserved size), set port number : 40001, mudlib directory to the absolute work/ path. 8. Full lpcc_check.sh sweep run (854 files after the shenminmap.txt rename; 832 pass / 22 fail). Fixed 6 pre-existing typo bugs found this way (all unrelated to our conversion — verified against raw archive bytes): - d/zuichen/{cdkedian3,cdkedian4,cdyaopu,chengzhong}.lpc: fullwidth comma ， used as an argument separator (AGENTS.md §9). - d/shiwang/qianqingon.lpc: fullwidth period in #include <ansi。h> (AGENTS.md §9). - std/room/shop.lpc, cmds/std/put.lpc: missing closing " before string concatenation (AGENTS.md §10). - adm/daemons/rankd.lpc: 8× switch(x) { default: ... } with no real case label (modern grammar requires at least one) — removed the pointless switch wrapper, kept the body (indentation not perfectly preserved, cosmetic only). - cmds/wiz/invis.lpc: declared void main(...) but returned 1/0 (command-dispatch status codes) — changed to int main(...). - obj/weapon/axe.lpc: copy-paste bug, inherit AXE/init_axe(...) but /std/weapon/axe.lpc was never implemented in this lib and the file's own header comment/id/name all say "dagger" — redirected to inherit DAGGER/init_dagger(...) (AGENTS.md §11). - obj/cloth.lpc: same pattern, inherit CLOTH with no /std/armor/cloth.lpc to back it — redirected to inherit U_ARMOR (matches its setup() call and "basic starting equipment" role).

Known remaining issues (not fixed — documented, low priority)

• /adm/daemons/network/{ms,userid,pingtcp,mail_serv,netmail,pingd,inetd}.lpc — a cluster of optional network-service daemons (internal mail server, ping, telnet redirect, inetd) failing to compile due to missing headers (config.h, mailer.h, daemons.h, uid.h, priv.h, post.h — not present anywhere in include/) and a few undefined functions/variables cascading from that. None of these are on the preload list or required for normal play (login/rooms/combat/items all work); left broken.
• adm/simul_efun/object.lpc: harmless lpcc-sweep-only false positive (AGENTS.md §6b/§14) — real boot never hits it.
• A number of call_other()/Bad argument 1 to EFUN call_other() sweep failures are the same false-positive category as lib #2 (AGENTS.md §6b) — not individually triaged since the pattern is now understood; spot-checked 2, both were "target genuinely not compiled yet in this isolated single-file run", not real bugs.

How to run

cd libs/shzs
~/src/fluffos/build-debug/src/driver config.fluffos
# separately:
python3 ../../scripts/mudclient.py 127.0.0.1 40001 --timeout 10 --send "" --send "look" --send "quit"

Post-hoc fix: UTF8-native is_chinese/registration (AGENTS.md §15h)

Applied in a later batch pass across the whole project: is_chinese/is_chinese2 in the shared chinese.lpc simul_efun fragment used GBK byte-range checks that silently never match real Chinese text once strings are UTF-8 (this driver's str[i] returns a Unicode codepoint, not a GBK byte). This broke character registration specifically -- any real Chinese name was rejected. Fixed the range check to test the CJK Unicode block instead, and halved the GBK-byte-calibrated length bounds in check_legal_name to match. See AGENTS.md §15h for the full writeup; confirmed via a real interactive registration test (Chinese surname + given name reaching the next prompt).

Re-verification pass: driver rebuild + formatter + WASM (2026-07-23)

• LPC formatter: ran format-corpus.mjs across all 854 .lpc files under work/ — all reformatted cleanly except 2 refused on the token-mismatch safety gate (adm/obj/simul_efun.lpc, adm/simul_efun/system.lpc), expected/fine per the formatter's own self-check contract, not investigated further.
• Native retest against the freshly-rebuilt driver: booted clean, zero fatal errors. Ran a full interactive registration with a real English id (qflibtest, this lib's login uses an English id + a separate Chinese display name 秦风) through name/password/email/ attribute-allocation/gender, reached the actual game world (梦旅馆大厅), and confirmed look/score/quit all produced correct, real output (score sheet shows the chosen 10/10/10/10/10/10 attribute split; quit exits cleanly). No regressions from the reformat or the driver rebuild.
• WASM build test (scripts/wasm_client.js): fully playable. Boots cleanly under WASM, and the complete registration flow (id → confirm → Chinese name → password ×2 → email → attributes → gender) works end-to-end exactly like native, reaching the same starting room with look producing correct output and quit exiting cleanly. This lib does not gate login on query_ip_number() or on any other WASM-restricted efun (its two resolve() call sites are in ftpd.lpc/dns_master.lpc, both optional network daemons never invoked during normal login) — no WASM-specific limitation applies here. Status: fully playable under WASM.

WASM-enablement pass (2026-07-24): admin seeding only (no gates found)

Checked for all four standard gate classes (AGENTS.md §1.3b/e): no band.lpc/BAN_D-style ban daemon, no site-restriction daemon, no uptime() check anywhere in the login path, and no per-IP anti-flood/ multi-login throttle (logind.lpc/logind2.lpc only use query_ip_number/query_ip_name cosmetically, in log lines and "who connected from where" messages — confirmed by grep, not just absence of symptoms). So items 1-3 of the standard pass (loopback-allow, uptime-bypass, throttle-exemption) are not applicable to this lib — nothing to patch.

Admin account: registered fluffos / Mud@2026 through the real registration flow (id fluffos accepted directly — no length/reserved- word conflict), Chinese name 浮浮, gender m, attributes 10/10/10/10/10/10. Granted (admin) via /adm/etc/wizlist (this file already had a fluffos (admin) line staged from an earlier interrupted pass; verified /adm/daemons/securityd.lpc reads this file for wizard-level decisions). Verified in ONE continuous session: login as fluffos, look, then update /adm/daemons/securityd.lpc recompiled successfully (ACL grants write/compile access), then clean quit. Re-verified a fresh normal registration end-to-end afterwards (id qinfeng, name 秦风, same attribute split) reaching 梦旅馆大厅 with working look/score/quit; its test save (data/login/q/qinfeng.o, data/user/q/qinfeng.o) was removed afterwards. debug.log clean (compiler warnings only, no errors) across both sessions. Save files for the orchestrator to force-add: data/login/f/fluffos.o, data/user/f/fluffos.o (untracked, not gitignored).

Note: an earlier same-batch agent's driver process for this lib (PID 4045698, started 00:37, before this session began) was found still bound to port 40001 with no admin work done yet; it was killed (exact PID) and the driver relaunched fresh before doing the registration above.

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

First genuine hands-on *playthrough* pass on this lib (all prior passes verified registration + look/score/quit/update, never a real walk through content). Native driver (build-debug), one continuous session per check, scripts/mudclient.py. Test character: id linfeng, Chinese name 林风, password Abc12345, attributes 15 力量/ 10 敏捷/10 智力/6 注意/17 体质/2 技术 (the game's own example split from help's attribute-allocation screen) — kept as a representative playthrough character, currently standing on a d/lwe/jie* street (state below). Save files: work/data/user/l/linfeng.o, work/data/login/l/linfeng.o.

This lib's help newbie/help intro don't exist as such; its actual newbie doc is help (topics: story, cmds) — read first, per the checklist, and immediately found broken (see bug 1).

Bug 1 — help/help cmds/help story and the day/night broadcast

text were still raw GBK bytes, never converted (AGENTS.md §4.1 class, just missed on this lib's original pass)

• Symptom: the *very first thing* a new player is told to try (help, and help motd — quoted verbatim in the MOTD bulletin as "本公告可以用 help motd 再次查看") rendered as mojibake, and every outdoor day/night phase transition (adm/daemons/natured.lpc, preloaded, cycles on a call_out the whole time the driver is up) broadcast mojibake to every player standing outdoors.
• Root cause: doc/help/topics, doc/help/cmds, doc/help/story, adm/etc/nature/day_phase, adm/etc/motd were never put through the GBK→UTF-8 conversion in any prior pass on this lib — confirmed by file/byte-decode check (still literal GBK bytes, not even mis-decoded, genuinely untouched) and by diffing against raw/simple/. This is exactly the AGENTS.md §4.1 "convert EVERY text file, not just source" rule; this lib's original conversion pass (documented above as "774 converted, 124 already ASCII/UTF-8, 0 hard failures") simply never walked doc/help/ or adm/etc/nature/.
• Fix: iconv -f GB18030 -t UTF-8 on all five files, byte-for-byte (verified: same line count, same non-Chinese content, only the Chinese text changed from garbage to real sentences — e.g. day_phase's dawn/noon/dusk broadcast lines, previously unreadable, now read correctly, e.g. "已经是中午了，太阳从你的上方照耀着大地。"). adm/etc/motd likewise (its "help motd" bulletin, previously garbage, now renders and was independently re-confirmed live at every login in this session).
• Verified live: help (topics listing), help story (the intro backstory, "山海战神之简单武力的故事背景..."), help cmds (the full command chart with Chinese glosses) and the MOTD bulletin at login all render correct Chinese now; re-tested after the driver restart that picked up the fix.

Bug 2 — per-command 指令格式： (syntax) help lines mis-decoded as

GBK when the source byte run was actually BIG5 (same §4.1 class, a narrower/more surgical instance)

• Found while checking help study, help load, etc. (the individual command help topics help.lpc falls through to in its default search path): the "指令格式：" prefix decoded correctly (it's genuinely GBK, shared boilerplate the Chinese-reskin author typed) but the bracketed argument placeholder immediately after it was garbage in cmds/std/{pilot,steal,load,study,wear,vl,sl}.lpc and cmds/wiz/edstar.lpc — a single source line mixing two different encodings, GBK boilerplate concatenated with the ORIGINAL Final-Frontier-base BIG5 English-to-Chinese translation of the argument name (this codebase's ES2/FF heritage — see study.lpc's own header comment "Original ES II command" / "Last modified by Spock @ FF" — was shipped in Traditional/BIG5, and the reskin author added Simplified/GBK text around it without re-encoding the file). Confirmed by decoding the exact raw byte ranges: the "指令格式：" bytes are valid GBK, the following <...> bytes are valid BIG5 and garbage under GBK (e.g. study.lpc's raw bytes for <物品名稱> decode to <珇嘿> under GB18030 — silently wrong, not an error, so the original conversion's lossy-conversion log never flagged it).
• Fix: byte-precise per-line re-decode against raw/simple/ (GBK prefix up to and including the fullwidth "：", BIG5 for the rest), applied as a single-line patch preserving the file's original CRLF line endings and every other line untouched. Examples: study.lpc "指令格式： study <珇嘿>" → "指令格式： study <物品名稱>"; edstar.lpc "...edstar -a <琍╰畒夹 x,y,z> <琍╰郎>" → "...edstar -a <星系座標 x,y,z> <星系檔案>".
• Caught and reverted a self-inflicted mistake while building this fix, worth recording: my first attempt regenerated whole files fresh from raw/ bytes (auto-picking GBK vs. BIG5 per line by scanning for stray bopomofo/hiragana/katakana characters — a decent first-pass heuristic, it correctly re-decoded 65 files' worth of lines) — but that wholesale regeneration silently reverted earlier, unrelated fixes already baked into the committed source (static→nosave in 5 files, i.e. AGENTS.md §4.3) and, separately, flattened CRLF to LF across every touched file (Python's default text-mode write), producing a 90+-line diff for what should have been a 1-line change. Caught both via git diff --stat before committing anything (static reappearing was caught by re-grepping for it post-fix; the CRLF flattening by the diff being suspiciously large for a single-line semantic change). Reverted with git checkout --, redid the fix as a precise byte-level single-line patch per file instead. Lesson for future encoding-repair work on already-converted files: never regenerate a whole file from raw/ once other fixes have landed on top of the conversion — patch only the specific corrupted byte range, and read/write in binary mode (or explicit newline='') to preserve the original line endings.
• The bopomofo/kana-based automated detector this project has used elsewhere (e.g. AGENTS.md §4.1's "human skim" heuristic) has a real false-negative gap: when BIG5 bytes get mis-decoded as GBK and land on *other valid-looking (but wrong) CJK ideographs* rather than bopomofo/kana, it produces no signal at all — confirmed directly: after fixing the 9 command-syntax lines above, the *body* text of those same help blocks (e.g. study.lpc's "丁: ︓ぶ 100 虫︗, 珇のㄤず甧肂." / load.lpc's "丁: 10 虫︗.") is still mis-decoded BIG5-as-GBK and the automated scan (re-run clean afterward) does not flag it, because the wrong decode happens to land on ordinary CJK characters. Manually spot-checking ~15 more 指令格式 lines across the corpus (d/lwe/npc/zhahuo.lpc's dialogue, several others) turned up the same shape repeatedly. This is scoped far beyond what's reasonable to hand-fix in one pass — the automated detector alone already found 65 files with at least one bopomofo/kana artifact, and that's a lower bound, not the true count. Fixed only the 9 highest-confidence, most player-visible lines (the help <cmd> syntax headers, since those are what a genuinely new player types first) and stopped there rather than risk more half-finished wholesale edits under time pressure; the wider cleanup is flagged below as explicitly NOT done. This is the existing AGENTS.md §4.1 "mixed encodings within one file" class, just present at a much larger scale in this lib's cmds//feature//adm/daemons//d/ trees than the single cataloged example (huoying's config.cfg) suggested — worth a dedicated future pass, not a quick fix.

Bug 3 — a player who disconnects uncleanly is stranded in the void,

via TWO independent code paths (new bug class — see report for the AGENTS.md draft)

Found by literally doing what §10.7 step 5 asks: quit, grep debug.log, reconnect after a real gap, confirm state — except the very first "reconnect after real gap" (a session that ended by the mudclient.py connection simply going away, i.e. exactly what a real player's dropped wifi/closed laptop lid looks like, not a clean quit) landed linfeng in "影子世界" (obj/void.lpc, the VOID_OB net-dead holding room — zero exits) instead of back where they'd been.

• Path A — force-quit-from-void persists /obj/void as startroom forever. obj/user.lpc's net_dead() (driver-invoked automatically on a dropped connection) immediately parks the disconnected player in VOID_OB (remembering the real room in a nosave/temp "last_location" var) and schedules user_dump(DUMP_NET_DEAD) for NET_DEAD_TIMEOUT (900s) later, which force-runs command("quit") on the still-netdead, still-in-void object if they never reconnect. cmds/usr/quit.lpc:47-59 (pre-fix) unconditionally persisted base_name(environment(me)) as "startroom" — with no special case for VOID_OB — so this force-quit-from-the-void wrote "startroom":"/obj/void" straight into the save file. Every future login (any number of days later, driver restarts, doesn't matter) then calls enter_world() → user->move("/obj/void") → succeeds (void is a perfectly valid loadable room) → player is now permanently starting every session in a room with no exits. Reproduced live: an early test session in this pass ended without quit (idle-timeout close), and by the time I came back the disk save already read "startroom":"/obj/void" — confirmed with grep -o '"startroom":"[^"]*"' data/user/l/linfeng.o.
• **Path B — LOGIN_D's own reconnect() never restores the void-parked location, and obj/user.lpc's reconnect() apply that *would* is dead code. obj/user.lpc defines a reconnect() apply, doc-commented "called by the LOGIN_D when a netdead player reconnects", whose whole job is move(query_temp("last_location")) — exactly the fix Path A needed. Nothing calls it.** grep -rn "->reconnect(" work/ across the whole lib returns zero hits. adm/daemons/logind.lpc has its *own*, differently-scoped reconnect(object ob, object user, int silent) function (re-links the connection, sends a "reconnected" room message, clears the netdead flag) that get_passwd() actually calls on a netdead reconnect — and it never moves the player out of VOID_OB at all. So even a player who reconnects promptly, well inside the 900s window, stays stuck in the void; obj/user.lpc's own reconnect() was seemingly meant to be the fix but was simply never wired up. Reproduced live with debug write()/tell_object() instrumentation (removed before the final fix): a fresh login after an unclean disconnect showed find_body=found netdead=1 last_location=/d/lwe/jie11 env=/obj/void, entered the reconnect(ob, user) branch, and the room remained /obj/void the entire time — last_location was captured correctly by net_dead() but never consumed by anything in the live call path.
• Fix (both paths, independent, both needed — matching the two distinct trigger conditions): 1. cmds/usr/quit.lpc: rewrote the duplicate/overlapping startroom-setting logic into one clean if/else chain; when environment(me) is VOID_OB at quit time, pull the real room from me->query_temp("last_location") instead of persisting the void itself (falls back to leaving startroom untouched if even that's unavailable, rather than ever writing /obj/void). 2. adm/daemons/logind.lpc's reconnect(): before re-linking, if environment(user) is VOID_OB and user->query_temp("last_location") is set, user->move() there first — doing, in the code path that's actually invoked, what obj/user.lpc's dead-code reconnect() apply was clearly meant to provide.
• Verified live, both paths, after rebuilding nothing (LPC-only change) and restarting the native driver: (a) logged in fresh, moved to /d/lwe/jie2, let the connection drop uncleanly (no quit), reconnected within seconds — landed correctly back in 街道 (jie2), not the void, and a subsequent clean quit persisted "startroom":"/d/lwe/jie2" (not void) to disk; (b) separately confirmed the disk-corruption half by hand-editing "startroom":"/obj/void" into the save file (simulating what an UN-reconnected, timed-out net-dead session would have written pre-fix) and logging in fresh — enter_world()'s own startroom path is unaffected by this fix (by design — the fix stops void from ever being *written* again; recovering a save that's already been corrupted by the old bug needs a one-time manual startroom reset, which is what was done for linfeng's own save after the pre-fix reproduction above). debug.log stayed empty (no error:/fatal lines) across every session in this entire pass, fix included.
• This is a new bug class, not a rediscovery of an AGENTS.md §7.16/§7.17-style crash — nothing aborts or errors, debug.log is and stays completely clean, which is exactly why boot-log watching and registration smoke tests can never catch it. Full draft writeup for AGENTS.md is in the task's final report (not added here — the orchestrating session owns AGENTS.md edits).

What else was tested and confirmed working

• Registration: real Chinese name (林风), full flow (English id → confirm y/n → Chinese name → password → confirm → email → attribute allocation with the in-game example split 15/10/10/6/17/2 → gender) — reached 梦旅馆大厅 (the Dream Inn hall) correctly. look/score/i all correct at every stage (fresh, post-move, post-shop-visit, post-reconnect).
• Exploration: walked the d/lwe town zone (夜/清晨/正午/傍晚/入夜 time-of-day description variants all render correctly and change live as in-game time passes — confirmed by re-looking the same room at different points in the session), reached both the general store (d/lwe/shop.lpc, NPC 杂货老板/zhahuo) and the weapon store (d/lwe/shop1.lpc, NPC 段楚沛/duan chu pei, id list includes wuqi's directory-name alias lwe wuqi but *not* bare wuqi — the in-room [duan chu pei] bracket is the actual matchable id, a minor navigation trap for a human player but not a bug).
• Shop list: list at the weapon store correctly printed real priced stock (锯齿屠刀 24 银币, 长剑 18 银币, 短棍 1银20铜, 匕首 7银20铜) — confirms feature/dealer.lpc's do_list() and the underlying MONEY_D->price_str() formatting work correctly. list at the general store correctly printed the "nothing for sale" path (zhahuo only carries worn/wielded equipment, which do_list() correctly excludes — a content-completeness gap in the shipped NPC, not a bug: it has zero unequipped stock and no vendor_goods array).
• kill/combat command surface: kill <target not present> correctly returned "这里没有这个人。" (no crash, no silent failure).
• Session persistence: confirmed across three separate reconnect/relogin cycles in this pass (once accidentally exercising the net-dead path pre-fix, twice deliberately post-fix) that attributes, inventory, and (after the Bug 3 fix) location all persist correctly across a clean quit+relogin.
• debug.log: grepped after every single quit in this entire pass (a dozen-plus sessions) — stayed completely empty (no file even gets created) throughout, both before and after every fix.

Explicitly NOT verified live (with reason)

• Safe-sparring / training-dummy mechanism: none exists in this lib. Grepped accept_fight across the whole tree — zero hits. No 木人/训练/陪练/dummy-pattern NPC is reachable from the starting zone (d/mochen/dongwuchang.lpc's "东练武场" training ground, with wooden dummies and sandbags in its room description, is real content but lives in a completely disconnected zone — d/mochen/d/mohao/ d/shiwang/d/zuichen/d/tian/d/shenmin have zero cross-links to or from d/lwe, confirmed by grepping every d/lwe/*.lpc for those path prefixes and vice versa). This lib genuinely ships without a reachable safe-combat mechanism.
• Organic skill-teacher path: only one NPC in the entire lib implements the teacher/apprentice pattern (d/mohao/ouyangxiu.lpc, inherit F_MASTER, attempt_apprentice()/recruit_apprentice()) and it's in the same disconnected d/mohao zone above — unreachable from the actual starting area. Also checked the study command's item-based path (cmds/std/study.lpc, reads a "skill" property off a carried book/item): grep -rn 'set("skill"' across the whole work/ tree returns zero hits — no such item exists anywhere in the shipped content, reachable or not. This lib ships with no functional organic skill-acquisition path at all; not a conversion bug (nothing here is broken code, there's just no content wired up to use it), documented here rather than silently skipped.
• Sect/faction-join mechanism: same finding — feature/apprentice.lpc (the family/create_family/recruit_apprentice sect-equivalent infrastructure) is only ever invoked by the same single unreachable ouyangxiu.lpc NPC. No join command, newbie-gift shortcut, or other faction-entry path exists in the reachable zone.
• Live combat exchange, death/respawn: not reached. No safe sparring target exists (above) and deliberately picking a real fight with an unknown wild NPC's stats, without a safe first data point, was judged not worth the risk this late in an already long session; the kill command's basic dispatch (invalid-target case) was confirmed clean, but a full turn-by-turn combat log, a kill, and the death/respawn cycle were not exercised live. Reason: time budget — most of this pass's time went to the Bug 3 investigation (instrumented debugging across ~15 test sessions to isolate a two-path bug that only reproduces via specific disconnect timing), which was judged the higher-value use of the remaining budget over finishing the shop-purchase/combat checklist items.
• A completed shop purchase (buy): list was verified live (above) and the underlying feature/dealer.lpc:do_buy() code path was read and looks correct (affordability check, MONEY_D->player_pay, ob->move(this_player())), but a live buy <item> transaction wasn't completed in this pass — repeated navigation attempts to reach the weapon-shop NPC in the same room as a buy command kept landing one room off (this lib's street grid has more live exits than its set("exits", ...) blocks show via static grep, e.g. jie2 live-shows a south exit no .lpc source search found — not investigated further, out of scope for this pass) — code-reviewed, not live-verified; flagging explicitly rather than silently presenting it as tested.

深度功能测试第二轮 / Deep functional test round two (2026-08-15, post driver-upgrade re-test)

驱动于 2026-08-12 升级后的重测。标准检查清单发现并修复三处问题：

1. config.fluffos：maximum evaluation cost 从 700000（已知 风险区间）提升到 5000000。 2. adm/simul_efun/file.lpc：log_file() 没有 assure_file() 目录预建保护，补上调用及前向声明；cat() 补上 read_file() || "" 空值防护。 3. obj/user.lpc::reconnect()（AGENTS.md §7.108，第五条独立确认 的血统——本档案与此前命中过的 shenzhou/shiji/shujian2008/shujian3 均无关联，是完全不同的 obj/user.lpc 架构，同一 bug 在这条独立 血统上再次出现）：adm/daemons/logind.lpc 有同款 exec(old_link, user); 踢掉重复登录写法，reconnect() 缺少 enable_commands()。 按 §7.108 记录的写法预防性修复，现场用两个真实连线复现"保持第一 个连线不断开→第二个连线登录→答 y 踢掉旧连线"验证：score 修复 后立即正常显示完整角色档案。

cmds/wiz/update.lpc（§7.106）与 master.lpc::log_error()（§7.10 的 "arning:" 大小写无关写法）均已是正确写法，无需改动；本档案无 adm/daemons/closed.lpc，不受 §7.107 影响。

现场验证摘要

驱动干净启动，管理员 fluffos/Mud@2026 登录确认 目前权限: (admin)，update /adm/daemons/logind 成功验证真实写入权 限。踢掉重复登录重连路径现场验证通过（见上）。debug.log 全程干净 （448 行，无真实错误）。

本轮修改的文件

• config.fluffos
• work/adm/simul_efun/file.lpc
• work/obj/user.lpc

深度功能测试第三轮 / Deep functional test round three (2026-08-18)

Standard §7.111/§7.112/logind-save/case-mismatch checklist (all clean, no action needed):

• §7.111 (master.lpc::standard_trace() unguarded file_name(error["object"])): already fixed here — objectp(error["object"]) ? file_name(...) : "(driver)" at adm/obj/master.lpc:200.
• §7.112 (init() unconditionally scheduling a call_out() chain): grepped every init() with a call_out( in it (~47 hits, all NPCs). All of them either fire a single one-shot call_out("welcome", 0, ...) from the shared feature/attack.lpc (no chain, harmless if duplicated), or already guard their "greeting" call_out with remove_call_out("greeting"); call_out("greeting", 1, ob); — the correct anti-duplication idiom. No plain ROOM objects schedule call_outs from init() at all. Nothing to fix.
• logind.lpc::enter_world() missing ob->save(): present and correct (user->save(); ob->save(); right after setup(), before the MOTD).
• u/<wizid> case-mismatch: not applicable — this lib has no u/ wizard-home tree at all.

Bug found and fixed — adm/simul_efun/gender.lpc was BIG5 bytes

mis-decoded as GBK during the original conversion pass (same class as round two's Bug 2, much wider blast radius)

• Symptom: found while doing a from-scratch PK-combat playthrough (round one/two never exercised PK). Every first-person combat/action message that should read "你" (you) — e.g. "你对着测试四号喝道...", "忽然，狂声大笑，纵身跃起，向...发动攻击" — instead showed a blank where "你" belongs (leading space, no pronoun). Also affected score's "你是一...的男性人类" line, tell/reply, emote, and any other message using gender_self()/gender_pronoun().
• Root cause: adm/simul_efun/gender.lpc (the shared gender_self() / gender_pronoun() simul_efuns) was raw BIG5 bytes in the archive (raw/simple/adm/simul_efun/gender.c) but got run through the project's GB18030 conversion pass like everything else. Because these particular BIG5 byte sequences (case "女性"/"男性"/"雄性"/ "雌性" etc., and their return values "你"/"妳"/"他"/"她"/ "牠"/"它"/"祂") happen to be *unmappable* under GB18030, iconv's lossy fallback silently remapped them into Unicode Private-Use-Area codepoints instead of erroring — so the file "converted" cleanly with no warning, but every case label became garbage that could never match a real "男性"/"女性" gender string, and gender_self() always fell through to its own corrupted default: value (also garbage, effectively empty when rendered). Confirmed byte-for-byte: decoding raw/simple/adm/simul_efun/gender.c as BIG5 recovers perfectly formed, grammatical Chinese matching this exact function's known contract.
• Fix: precise byte-level patch (Python binary mode, no whole-file regeneration) replacing only the 8 corrupted string literals with their correctly-BIG5-decoded UTF-8 equivalents. git diff --stat confirmed exactly 8 lines changed, CRLF preserved.
• Verified live: rebuilt driver, ran the same PK-combat scenario — first-person messages now correctly show "你" throughout (e.g. "你对 着测试四号喝道小王八蛋！你死定了！"), third-person pronoun refs also correct. debug.log stayed empty across the whole session.
• This mis-decode class (§4.1/round-two's Bug 2) is proven to recur independently at least twice in this one lib now (help-topic syntax lines in round two, this shared simul_efun in round three) — flagging for the orchestrator: any lib sharing this "Final Frontier"/ES2 ancestry (BIG5-authored base + GBK-authored Chinese reskin layered on top) should have adm/simul_efun/gender.lpc (or equivalent) spot checked, since a silently-empty gender_self() has no crash signature and is very easy to miss without an actual PK/combat playthrough.

Bug found and fixed — netdead reconnect never restores heart_beat,

permanently soft-locking anyone who dies and ever has one dropped connection (new bug class, not previously in AGENTS.md)

• Found while trying to complete the death → reincarnation loop (kill → die → /d/death/gate → /d/death/hall → say return to 史巴克). d/death/npc/toilet.lpc's relay_say() requires every limb's limbs_eff_hit to be at least 50% of limbs_max_hit before return is allowed ("您目前的状况还无法回到人世，请多休息一下吧" otherwise) — reasonable design, not a bug. But after admin-forced death (call hbtestb->die(), matching real PK-death numerically) the character's limbs sat frozen at 1/1/max (重伤) for over a minute of real wall time with zero movement, never healing even while standing — the one path (feature/damage.lpc::heal_up()'s standing branch) that's supposed to add +1/tick to a ghost's limbs_eff_hit.
• Root cause, confirmed live via the info <target> wizard command's "心跳:N" field (the reliable way to read real heart-beat state — the call obj->query_heart_beat() idiom used earlier in this pass is *not* reliable for this efun, since query_heart_beat has no user-defined LPC wrapper and call_other to a nonexistent function silently returns 0 rather than erroring, which cost real time here chasing a false "always 0" signal before switching to info): obj/user.lpc::net_dead() correctly calls set_heart_beat(0) on an unclean disconnect (self-context, correct). But the *actually invoked* reconnect path, adm/daemons/logind.lpc::reconnect(object ob, object user, int silent) (called from get_passwd() when user->query_temp("netdead"), and from confirm_relogin()'s duplicate-login-kick path), never re-enabled it — so heal_up() (and aging, idle-timeout, hunger/thirst decay — everything gated on heart-beat) silently stopped running for the rest of that session, any time a player's connection ever dropped and they reconnected. Combined with the death mechanism's healing requirement, this means: any player who dies and has ever had even one dropped connection can never heal enough limb health to reincarnate again, with zero error/crash signature (debug.log stays completely clean the whole time). obj/user.lpc already has its own reconnect() *apply* with the correct shape (enable_commands(); set_heart_beat(1); ...), doc- commented "called by the LOGIN_D when a netdead player reconnects" — but nothing calls it (same "intended but never wired up" dead-code pattern round one's Bug 3 found in this same function name).
• Fix: set_heart_beat()/enable_commands() are efuns with no cross-object form (bare calls always target this_object(), which inside logind.lpc::reconnect() is the daemon itself, not the player — this is exactly why a first attempt at this fix, calling the efuns directly from logind.lpc, silently did nothing). Added a thin resume_heart_beat() wrapper to obj/user.lpc (right after the existing dead-code reconnect() apply) that does enable_commands(); set_heart_beat(1); in the user object's own context, and call it as user->resume_heart_beat(); from logind.lpc's actually-invoked reconnect().
• Verified live with an A/B test (git-stashed the fix, rebuilt, re-ran the same scenario, then restored and rebuilt again): pre-fix, info <player> showed 心跳:1 on fresh login but the "心跳:" field disappeared entirely after a real drop-and-reconnect cycle (heart_beat = 0, confirmed reproducible). Post-fix, 心跳:1 is present both before and after the same reconnect. Further verified the actual downstream effect: killed a test character, dropped+reconnected them, then sampled body every 20s — limb values climbed steadily (1→3→5→8 over 60s, previously frozen forever at 1) confirming heal_up() is running again and the reincarnation path is no longer permanently soft-locked. debug.log stayed empty throughout.
• Flagging for the orchestrator: this is a new, previously uncataloged bug class — likely recurs in any lib whose net_dead()/reconnect() split mirrors this shape (a set_heart_beat call in the disconnect handler with no corresponding restore in the actually-invoked reconnect path). Worth a grep sweep across the corpus for set_heart_beat(0) in net_dead()-style applies paired with a reconnect() that doesn't call set_heart_beat(1) (directly or via a wrapper) — not fixed elsewhere in this pass, this lib only.

What else was tested this round

• PK combat (two live characters, cetesta vs cetestb): full exchange of blows, damage accumulation, skill-up messages, auto-flee trigger at low spirit/stamina, and the loser fleeing to an adjacent room — all worked correctly end-to-end (this is what surfaced the gender.lpc bug above).
• Attacking an essential shop NPC (kill li meng zhe): confirmed this is *intentional* content, not a bug — d/lwe/npc/bartender.lpc and every other vendor NPC in d/lwe override kill_ob() to show a dramatic "transforms into a beast" flavor message and immediately remove_killer() both ways, cancelling the fight before any damage lands. Confirmed via code (identical pattern across zhahuo/zhanggui/xiaoer2/jiubao/wuqi/faju), not a gap. Matches this lib's existing documented finding that no reachable wild monster exists in the connected d/lwe zone.
• Death → hell zone: /d/death/gate and /d/death/hall (with NPC 史巴克/toilet.lpc) *are* reachable (unlike the disconnected d/mochen/d/mohao/etc. zones noted in round one) — death correctly routes here via feature/damage.lpc::die(). The DEATH_ROOM-> start_death(this_object()) call in die() targets a function that doesn't exist anywhere in the codebase (gate.lpc is a plain ROOM) — confirmed this is harmless, not a bug: call_other to an undefined function silently returns 0 in this driver config, no error, no debug.log entry, and death still proceeds correctly.
• A completed shop purchase (buy): still not completed live this round either (admin-cloned gold onto a test character, but navigating on-foot to the weapon shop repeatedly landed one room off from wherever the last list/buy attempt needed — same live-exit- vs-static-grep mismatch round one already flagged for this lib's street grid). Not investigated further; list's working output and do_buy()'s code path were already reviewed as correct in round one.
• Registration flow, look/score/i/body, and the netdead/void reconnect fix from round two all re-confirmed still working correctly throughout this pass's ~15 test sessions.

Files modified this round

• work/adm/simul_efun/gender.lpc (BIG5-as-GBK mis-decode fix)
• work/adm/daemons/logind.lpc (call user->resume_heart_beat() in the real reconnect path)
• work/obj/user.lpc (new resume_heart_beat() wrapper)

§7.114 fix: private input_line() broke multi-line to/board/mail input (2026-08-19)

Same ES2/FF-ancestry bug as zzhj/dfgs2 (see AGENTS.md §7.114): work/feature/edit.lpc (this lib's F_EDIT mixin, inherited by std/char.lpc) declared its input_to() re-arm target private: private void input_line(string line, string text, function callback). The recursive re-arm from inside input_line() silently failed to register once reached through the inherited mixin, so any multi-line .-terminated input session (board post, to, mail) dropped everything after the first line. Fix: dropped private.

Live-verified on a fresh driver boot (port 40001), fresh character (秦捌捌/qinbaba): to say → two lines of text → . → resulting say output showed both lines (你 说道: hello line one\nhello line two), not truncated after line 1. This lib has no chfn.lpc and its board object isn't placed in any reachable room in this archive, so cmds/usr/to.lpc (which drives the identical edit()/input_line() flow) was used as the representative live test instead of post. Killed the driver by exact PID when done.

Files modified this pass:

• libs/shzs/work/feature/edit.lpc — dropped private from input_line().

§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). 10 live

replace_program(ROOM); occurrences across 10 room files (obj/void.lpc, d/zuichen/jinshedong.lpc, d/shiwang/yongdingmen.lpc, d/mohao/*.lpc x4, d/death/*.lpc x3) — same fatal redundant-replace-after-inherit shape as the rest of the §7.100 family. No room-building tool exists in this lib. Fixed by deleting the redundant lines. 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): 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.

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

scripts/lpc_fix_negative_ranges.py: 1 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（§10.7，2026-10-05）— tail() simul_efun

scripts/lpc_add_tail.py --apply: cmds/wiz/tail.lpc calls tail(file), an efun of the MudOS this archive ran on that FluffOS does not have (Undefined function tail: the command did not compile). adm/simul_efun/file.lpc now defines void tail(string file): the last 10 lines of the file, written to the player. Load check (HEAD against the working tree): 1 of the 1 tail command file(s) failed to load at HEAD, all load now, and the simul_efun file loads. KB 04 §6.2.

深度功能测试（§10.7，2026-10-05）— 房间列出的物件缺档时房间照常加载

房间基类 std/room.lpc：make_inventory() 用 catch() 包住 new(file)，做不出来就返回 0；reset() 的 case 1: 拿到 0 就跳过这一项（if (!objectp(ob[list[i]])) break;）。档案里没有的 NPC 或物品（或 create() 出错的物件）原先会让列出它的房间整个加载失败（Bad argument 1 to environment()，或对 0 的 call_other()），连带这个房间后面的区域；现在房间照常加载，只是少了那个物件，下次 reset() 再试（KB 05 §7.25；scripts/lpc_resilient_room.py、scripts/lpc_room_reset_guard.py）。 验证：探针（scripts/lpc_room_reset_probe.py：房间基类列出一个不存在的物件，再调用 reset()）HEAD 单件 崩溃、多件 崩溃，工作树 单件 OK、多件 OK。

深度功能测试（§10.7，2026-10-07）— 空的 if (current_water == 0) {}

2 个房间的 valid_leave() 里有一行什么也不做的 if (current_water == 0) {}（编译警告 Expression has no side effects），删去，行为不变：d/yishi/samo2.lpc、d/yishi/samo3.lpc。

深度功能测试（§10.7，2026-10-07）— 编译警告整理

scripts/lpc_warnings.py shzs --fix 加 scripts/lpc_fix_no_effect.py --apply，共改动 496 个 .lpc 和 1 个 .h（多为删去未用的局部变量）。HEAD 与工作树分别加载（新进程）：490 -> 490 通过，无回退；改过的头文件的 1 个包含者 1 -> 1 通过。启动时警告 17 -> 0；新进程登录（输入名字）正常。

编译警告收尾（2026-10-13）

scripts/toolkit_fix.sh 一轮把 feature/container.lpc、std/equip.lpc、std/money.lpc 的 query_autoload() 改成与 F_AUTOLOAD 一致的 mixed；其余手工，能编译的文件警告清零；改动的 6 个对象在 HEAD 与工作树各载入一次，regress 0。

• 知府欧阳修 d/mohao/ouyangxiu.lpc 把在大堂上 chat 的人打昏，处理函数叫 do_chat，正好盖住 std/char/npc.lpc 设定 NPC 下一句闲聊的 do_chat(mixed)；目前没有别处对他调用 do_chat，但名字冲突留着就是隐患。处理函数改名 do_noisy（两条 add_action 一起改），行为不变。
• obj/user.lpc 没用的 last_cmd（与 feature/alias.lpc 的同名）、std/space_ship/space_ship_room.lpc 没用的 doors（与 F_DOOR 的同名）删去。

family 映射（再补）（2026-10-15，KB 05 §7.30）

mapping fam; fam = (mapping)ob->query("family");（先声明后赋值）和泥潭式的 query("family", me) 同样直接取下标：无门派的新人拜师（attempt_apprentice/recognize_apprentice）、NPC 的 greeting、房间的 valid_leave、ask_me 等报 Value being indexed is zero。2 处补上 if (!mapp(fam)) fam = ([ ]);（刚 recruit_apprentice() 之后的除外）。

去掉的预定义宏 SAVE_EXTENSION（2026-10-18，KB 04 §6.11）

老版 MudOS 给每个 LPC 对象预先定义了 SAVE_EXTENSION（".o"）；FluffOS 的 ChangeLog（COMPAT_BUSTER 一条）把它删了，改用 __SAVE_EXTENSION__。本库没有自己 #define 这个宏，用到它的文件一律 Undefined variable 'SAVE_EXTENSION'，整个对象载入不了。改成 __SAVE_EXTENSION__（scripts/lpc_fix_predefined_macro.py）：adm/daemons/network/netmail.lpc。netmail 是网络邮件守护进程（本库缺 mailer.h 等头文件，改后仍载入不了）。改后用 lpcc 逐个载入这些对象：0 个通过，1 个还有别的错误（adm/daemons/network/netmail：/adm/daemons/network/netmail.lpc:17: error: Cannot #include mailer.h）。

for (int i ...) 的循环变量在循环外还在用（2026-10-18，KB 04 §6.12）

当铺／商店的 do_buy() 写成 for (int i = 0; ...) if (...) break;，循环一结束又读 if (i == sizeof(inv))。FluffOS 里 for 头里声明的变量只在循环内有效，所以 Undefined variable 'i'，整个房间（NPC）载入不了，当铺根本进不去。把 i 声明挪到函数开头（int i, index;），循环改成 for (i = 0; ...)：std/room/shop.lpc。改后用 lpcc 逐个载入这些对象：1 个通过。

同一个 init() 写了两遍、内容不同（2026-10-19，KB 06 §7.265）

d/zuichen/huli 里 init() 定义了两次（一个加 buy/work 之类的指令，另一个再加 list/block_cmd），驱动报 Redeclaration of function 'init'，整个物件载入不了。两处作者本来都要：把第二个 init() 里的 add_action(...) 并到第一个 init() 末尾，删掉第二个。（scripts/lpc_merge_dup_init.py）

出口指向不存在的房间（2026-10-19，KB 06 §7.257）

下列房间的出口写的文件不存在，而对面恰好只有一个房间的出口指回本房间，所以按对面的房间改正（笔误、大小写、路径少了一级）：

• /d/shenmin/wroad2.lpc 的 eastup 出口：jiaolou → jiaojou
• /d/shenmin/zhuque.lpc 的 westup 出口：jiaolou → jiaojou

改后房间照常载入（与 HEAD 逐个对比，没有退步），出口现在指向存在的房间。

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。

socket / dumpsocket 指令：dump_socket_status() 驱动里没有（2026-10-20，KB 06 §7.291）

这些巫师指令 write(dump_socket_status());，MudOS 的 efun 在 FluffOS 里不存在，指令编译不过。在模拟 efun 文件里补了 dump_socket_status()：用 socket_status() 排出同样的表（没有 sockets 包的构建，比如 WASM，只有表头；scripts/lpc_add_sefun.py dump_socket_status）。

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/obj/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）。

翻页器 start_more(read_file(...))：文件缺失时 explode(0)（2026-10-20，KB 06 §7.300）

东方故事系的翻页器 void start_more(string msg) { ... explode(msg, "\n") ... } 被这样调用：start_more(read_file(HELP_DIR + "help/topics"))、start_more(read_file(NEWS))。压缩包没带的文件使 read_file() 回 0，explode(0, ...) 在本驱动里报 Bad argument 1 to explode()（不带参数的 help 指令就是这样出错的），老驱动把 0 当空文本。在 start_more 开头加了 if (!stringp(msg)) return;（scripts/lpc_fix_start_more.py）。改动：/feature/more.lpc。

disembark：在船舰以外的地方对 0 调用函数（2026-10-21，KB 06 §7.332）

cmds/std/disembark.lpc 只有人在船舱房间里才给 env（船舰对象）赋值，否则 env 是 0，接着 if (!env->is_space_ship() && !env->is_shuttle()) return notify_fail("你不在任何船舰中。\n"); 对 0 调用函数：原驱动答 0，于是走到提示；这个驱动抛 Bad argument 1 to EFUN call_other() Expected: object, string, array, Got: int(0)，玩家在任何普通房间敲 disembark 都是运行时错误。条件前面加 !objectp(env) ||，提示文字不变。加载检查通过；扫描时是 disembark 的裸命令触发的。shzs 和 zzhj 同一处（前者简体、后者繁体）。

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：/std/char/npc.lpc（共 1 个文件）。

巫师命令不带参数时对 0 调用 efun（2026-10-21，KB 06 §7.336）

巫师（管理员）把命令敲成不带参数时，驱动把 0 传给字符串参数；下面的命令第一句就把这个 0 交给 efun，运行时报 Bad argument 1 to ...，这条指令从来没有用过。用 scripts/lpcc_bare.py 不登录、在一次 lpcc 求值里以 0 调用 main() 逐个确认后，由 scripts/lpc_fix_bare_arg.py 在第一处使用参数的语句前加守卫：文件自带 int help(object me) 的，守卫是 if (!arg) return help(me);（help() 定义在 main() 之后的 1 个文件补了原型 int help(object me);，否则编译报 Undefined function help）；没有 help() 的是 return notify_fail("…\n");。改动（2 个文件，括号里是接住 0 的 efun）：/cmds/arch/grep.lpc（sscanf）、/cmds/wiz/test.lpc（present）。加载检查 HEAD 与工作树对照，没有退步。

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 2 条注册（1 个文件，scripts/lpc_fix_dead_add_action.py），动词留给别的处理函数：/d/mochen/dongwuchang.lpc（do_ci、do_da）。最多的函数名：do_ci 1 处、do_da 1 处。

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 汉字循环 / is_chinese 区间 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 3 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。

无参数时命令对 0 的调用（玩家/--all-dirs 残留，2026-10-27，KB 06 §7.336 Also）

无参数命令把 0 交给 present / sscanf / file_exists 等会运行时报错。lpc_fix_bare_arg.py --all-dirs 在首次使用参数前加守卫。多行 if (arg && !sscanf)（吐纳默认消耗）与已有 can_learn/get_object 守卫的误报已回退；本库本批改动见该提交。

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。