es2 — 东方故事Ⅱ (git clone from mudchina/es2)

Source: git clone https://github.com/mudchina/es2 (commit ae8adf6, cloned 2026-08-25). Port: 40214. Status: done (boots clean, full registration verified end-to-end with a real Chinese name, look/ score/quit/whoami/update all confirmed working post-login).

Note: the sibling repo mudchina/es2-utf8 (pre-converted) was deliberately NOT used — per this project's established methodology (AGENTS.md §4), doing the GBK→UTF-8 restoration itself, on the original GBK snapshot, is the point.

What this is

A first-generation, from-scratch Chinese LPC mudlib, not a reskin of an existing English codebase. bin/es2.info names it MUDNAME=ES2; adm/obj/master.lpc's header credits "[email protected]" (1994-11-07) with feature/treemap.lpc crediting "Buddha@TMI" and noting it is "part of the TMI mudlib distribution" — i.e. this archive genuinely is (part of) the historical root of the "ES2/TMI-2" lineage name this project's AGENTS.md catalog uses for dozens of unrelated LATER Chinese mudlibs. mudlib directory is the repo root itself (no nested subdirectory); master file : /adm/obj/master, simulated efun file : /adm/obj/simul_efun.

Layout: adm/ (master/simul_efun/daemons), cmds/ (wiz/arch/imm/usr/ std command dirs, Lima-esque ADM_PATH/WIZ_PATH/... dispatch by wizhood()), feature/ (mixin classes inherited into std/char.lpc: action, attack, command, dbase/treemap, equip, food, message, move, name, save, skill, team, vendor, ...), d/ (game-world domains: city gates/streets, a snow-mountain sect, a temple, a bamboo canyon, a green forest estate, a graveyard/death area, a moon-pavilion zone, ...), u/ (individual wizard-workspace directories, e.g. u/cloud/), quest/ (experience-tier quest-list tables), daemon/class/* (character-class skill masters). Start room /d/snow/inn (饮风客栈), death room /d/death/gate.

Conversion (GBK→UTF-8)

scripts/convert_lib.sh libs/es2/raw libs/es2/work (default GB18030 encoding — confirmed correct via name : 东方故事Ⅱ decoding cleanly from the raw config.ES2's mojibake bytes). Results: 1933 files converted, 616 already UTF-8, 8 lossy (invalid-byte-dropped) files, 77 binaries skipped. Also stripped raw/'s .git/ from work/ (not part of the mudlib) and removed the prebuilt bin/driver/bin/addr_server binaries' influence (this project uses its own driver; left the files in place since nothing references them).

• 8 lossy-conversion files individually audited per AGENTS.md §4.1's warning that a dropped byte can silently corrupt adjacent syntax: d/choyin/npc/yamen_po.c, d/latemoon/upstar/upcenter.c (a @LONG/LONG heredoc room description — closing tag intact, one content character dropped), d/temple/{obj,npc/obj}/{spells_book, magic_book}.c (all 4: inside a // comment, harmless), u/cloud/obj/npc/flower_girl/guihua.c (see "Not fixed" below — independently corrupted, not a lossy-conversion artifact), cmds/std/exercise.c. All but guihua.c compile fine; the dropped bytes in yamen_po.c/upcenter.c/exercise.c land inside string literals or comments and only cost a character or two of decades-old, already-garbled NPC dialogue/room text (pre-existing raw-byte corruption confirmed present in raw/, not introduced by this conversion pass) — left as-is per AGENTS.md's "don't fabricate content" rule.
• Straggler extensionless GBK text files (file/extension-based sweep misses these): doc/help/{weapon,quest,guild,history,theme, taoist,story,myth} — converted with a direct iconv -f GB18030 -t UTF-8 pass.
• static→nosave string-literal collision (AGENTS.md §4.3): the blanket \bstatic\b→nosave sed rewrote log_file("static/...") path literals in adm/obj/master.lpc, adm/daemons/securityd.lpc, cmds/std/suicide.lpc, cmds/arch/purge.lpc, cmds/wiz/call.lpc (8 hits/5 files) — reverted "nosave back to "static in all of them. No #define nosave static/#define protected static compatibility shim found (the other known collision class).

Config / driver-compat fixes

libs/es2/config.fluffos modeled on the raw adm/etc/config.ES2 (decoded name/messages), pointed at libs/es2/work, port 40214; dropped 5 directives the driver flags obsolete (address server ip/ port, reserved size, binary directory, swap file).

1. adm/obj/master.lpc's valid_override() was 2-arg (AGENTS.md §10.4) — simul_efun.lpc #includes 7 fragment files, needing the 3-arg (file, name, main_file) signature for efun:: calls inside them to resolve cleanly. Widened the signature and OR'd in main_file == SIMUL_EFUN_OB. 2. Missing runtime directories, §7.11: /log/ and /data/ were never shipped (gitignored in the upstream repo) — mkdir -p log data. More importantly, the shared log_file() simul_efun (adm/simul_efun/file.lpc) had NO assure_file() guard at all (unlike feature/save.lpc's save(), which already calls assure_file()) — added one, with a forward declaration (this driver doesn't resolve forward references within a file without one — hit "Undefined function assure_file" until added). Also guarded master.lpc's own bypass call (efun::write_file(home + "log", ...) in log_error()) the same way. Without this, the FIRST diagnostic/crash log write of the whole boot threw before Accepting telnet connections. 3. feature/treemap.lpc's _query/_delete/_set were private (AGENTS.md §7.48) — feature/dbase.lpc inherits F_TREEMAP and calls them directly, illegal on this driver (private is file-local only here, not "protected-like"). Changed to protected in feature/treemap.lpc. Same class found and fixed in feature/action.lpc's continue_action()/eval_function() and feature/attack.lpc's attack() — all three called cross-file from std/char.lpc, which inherits F_ACTION/F_ATTACK. 4. origin() returns a string on this driver, not an int bitmask (AGENTS.md §7.47) — feature/team.lpc:20's origin()==ORIGIN_LOCAL || origin()==ORIGIN_CALL_OUT is always false because of incompatible types. This one line, inherited into every std/char.lpc-based object, was the single largest contributor to the initial compile-sweep failure count (1126 of 1610 raw failure lines — every NPC/player-body compile unit hit it once). Fixed: origin()=="local" || origin()=="internal" (the §7.47 mapping table: ORIGIN_CALL_OUT (0x10) → "internal"). 5. adm/simul_efun/object.lpc overrode destruct() (AGENTS.md §7.38, "Invalid simulated efunction override" — this driver hard-rejects it). Deleted the wrapper (lost its ob->remove(euid) pre-cleanup step, same documented tradeoff as every other lib that hit this). 6. std/room.lpc's make_inventory() had no guard around new()/move() (AGENTS.md §7.25, textbook match) — a room's "objects" mapping referencing a missing/broken content path crashed that room's first-ever population (*Bad argument 1 to EFUN call_other() ... Got: int(0), /std/room.lpc in the trace). Wrapped new() in catch(), added objectp() checks at both call sites in reset(). Root-caused (not just guarded) 3 of the crashing rooms: d/latemoon/{obj,npc/obj}/latebook.lpc's "objects" entry pointed at __DIR__"npc/girl", but girl.lpc only exists under the sibling d/latemoon/npc/ directory, not d/latemoon/{obj,npc/obj}/npc/ — a pre-existing content-tree drift (multiple parallel copies of the same zone), not a conversion artifact. 7. Same content-tree-drift class, but via carry_object(...)->wear()/ wield() chained straight off a possibly-0 return (no std/room.lpc helper involved, so #6's fix didn't cover these): d/village/ flowergirl.lpc (__DIR__"obj/flower" missing, only exists at /obj/flower.lpc or under a sibling npc/obj/), d/latemoon/room/ npc/aaa.lpc (__DIR__"houndbane", only exists under an unrelated d/class/fighter/), d/latemoon/room/npc/fong.lpc and .../jane.lpc (__DIR__"obj/{deer_boot,blue_dress,redbelt}", only exist under sibling d/latemoon/{obj,npc/obj}/). Guarded each call site (object ob = carry_object(...); if (ob) ob->wear();) rather than guessing which sibling copy was "the" intended path. 8. adm/simul_efun/message.lpc's tell_room() passed a defaulted int 0 as message()'s 4th (exclude) argument when called with only 2 args (tell_room(room, str), the overwhelmingly common call shape — 71 call sites across the tree) — this driver's message() requires that argument be void | object | object *, never a bare int, and throws *Bad argument 4 to EFUN message(). First surfaced via d/choyin/stove.lpc's create() crashing, but this was a shared root cause affecting essentially every tell_room() call in the game, not a per-file bug. Fixed once in the simul_efun: omit the 4th argument to message() entirely when no exclude list was passed, instead of passing a defaulted 0. 9. cmds/std/to.lpc used an old-MudOS call_other-shaped functional literal: (: this_object(), ({ "do_to", me, arg }) :), illegal on this driver ("Illegal to use local variable in functional"). Every sibling ->edit() call site in this archive (std/{bboard,jboard} .lpc, obj/mailbox.lpc) already uses the modern (: func, bound_args... :) shape — rewrote to match: (: do_to, me, arg :), plus a forward declaration for do_to (defined later in the same file; same forward-reference requirement as fix #2 above). 10. cmds/wiz/tail.lpc called a bare tail() efun that was never real on MudOS/FluffOS (not in this driver's .spec files at all). Reimplemented with read_file()+explode()+implode() (last ~15 lines) since this driver's read_file() also has no "negative start = from end" mode. 11. d/city/{south,east1,east2,west1,west2}door.lpc (all 5 city gates) call create_door("out", ...) unconditionally, but each one's own "exits" mapping has the "out" entry commented out (its target "outskirts road" room was never shipped in this snapshot) — create_door() hard-error()s without a matching exits[dir] entry, crashing all 5 gate rooms' first load. Disabled the now-inconsistent create_door() call to match the already- disabled exit, rather than fabricating the missing target room. 12. adm/daemons/securityd.lpc's valid_write() save-file check forgot the driver appends the save extension (AGENTS.md §7.49, textbook match) — compared the driver's file argument (full filename, .o included) against the bare query_save_file() return value, so it NEVER matched and every player save threw *Denied write permission in save_object(). This is the reason a freshly-registered character landed with "你的四周灰蒙蒙地一片， 什麽也没有" (blank void) instead of the start room: enter_world()'s user->save() call errors out midway through the function, before the user->move(startroom) line ever runs — boots clean, registers clean, only breaks on the FIRST real save (exactly the "survives a smoke test, only shows up on the full flow" signature §7.49 warns about). Fixed: file == qsf || file == qsf + __SAVE_EXTENSION__. Re-verified with a full fresh registration after the fix: character correctly lands in 饮风客栈 (start room), look/score/quit all produce correct output, and the save files (data/{login,user}/f/fluffos.o) are written successfully.

Author-typo fixes (pre-existing in raw/, AGENTS.md §6.6)

Confirmed against the raw GBK bytes for each before fixing (none are conversion artifacts):

• u/cloud/npc/goddd.lpc: a message_vision(...) call missing its closing quote before string concatenation ("...说:" 就凭你..., the classic §6.6 shape) -- fixed. Separately, later in the same file, an unbalanced-paren typo (num += ((... opens one extra paren versus its parallel else-branch sibling 3 lines below, which has the correct single-paren shape) plus sizeof(level) (undeclared) where every other use in the function says sizeof(levels) -- both fixed to match the parallel/sibling code exactly, not guessed.
• u/cloud/obj/sword_book.lpc: set("long",\net("long", ...) — a stray duplicate et( fragment (apparent leftover from a botched set("long", paste) breaking the call in two. Removed the stray fragment.
• u/cloud/npc/chess_player.lpc: name(this_player()) called name() as if it were a global "get an object's name" function; it's actually feature/name.lpc's own per-instance varargs string name(int raw) method. Fixed to this_player()->name() (2 call sites, also affects u/cloud/tearoom2.lpc which inherits this NPC).
• d/snow/npc/guard.lpc: is_killing(who) passed an object where feature/attack.lpc's is_killing(string id) expects the string id — every other call site in the tree correctly passes ->query("id"). Fixed to match.
• d/green/npc/master.lpc: map_skill("spells",magic-array) — missing quotes around the "magic-array" skill-name string literal (parsed as subtracting two undefined variables magic/array instead). Fixed to map_skill("spells","magic-array").
• d/village/lordhouse3.lpc: a commented-out if(...){ block left its closing } NOT commented out, producing a dangling brace before return 1;. Commented out the stray } too.
• u/cloud/duchang2.lpc: do_bet()'s source is genuinely truncated mid-statement in the raw archive (this_player->add(" with nothing after it, wc -l confirms the .c file itself ends there) — per AGENTS.md §6.6, closed the function body as a safe no-op (notify_fail()+return) rather than fabricating the missing gambling-payout logic. Also fixed the adjacent return "没钱就别赌！" from a bare string return inside an int-typed function to the notify_fail()+return 0 idiom every other action-command handler in this archive uses.
• d/npc/oldman.lpc: kill_ob()'s set("nickname", ...) was immediately followed by a bare string literal with no enclosing function call — a message("vision", ...) header (confirmed via the identical 4-arg message("vision", <text>, environment(), this_object()) shape used by 3 sibling files) had been dropped. Restored it. Separately in the same function, the parameter object ob (the target being killed) was reassigned to an ARRAY (all_inventory(environment()), a "reveal boss, aggro the whole room" trigger this override repurposes kill_ob for) and then indexed — a real type collision, not a missing-quote typo. Renamed the array-holding local to obs throughout rather than changing the parameter, preserving the original reveal-and-attack-all behavior exactly (this is a deliberate, if unusually-named, override — a sibling d/green/npc/oldman.lpc shows the "normal" kill_ob override shape, single-target delegation to ::kill_ob(); this file's author chose different semantics for this NPC and that choice was left alone, only the compile-blocking type bug was fixed).

eat_func/eat_tonic — dead-but-broken data, not fixed as content

d/snow/{obj,npc/obj}/{agaric,ginseng,ice_lotus,ebony_lotus}.lpc (all 4 tonic herbs, both obj/+npc/obj/ copies) and d/goathill/{obj,npc/obj} /dead_leech.lpc all set("eat_func", (: eat_tonic :)) — eat_tonic is referenced NOWHERE else in the archive (grepped the whole tree), a hard compile error for an undefined function pointer. Even if it existed, feature/food.lpc (the ONLY food class anything in this snapshot actually inherits — feature/food{old,bak}.lpc are unused duplicates, see "Not fixed" below) never reads "eat_func" at all (the one line that would have — //if (query("eat_func")) return 1; — is itself commented out). Dropped the dead set() call in all 10 files rather than inventing tonic-healing logic to give eat_tonic a body.

Not fixed (pre-existing content gaps / dead code, confirmed via raw/)

• adm/daemons/network/ (12 files: ping_q, mudlist_a, ms, userid, pingtcp, name_server, mail_serv, netmail, telnetd, pingd, inetd, plus services/) — an incomplete/never-finished Intermud-mail-and-services subsystem. References headers that don't exist anywhere in this archive at any path (uid.h, mailer.h, priv.h, post.h; config.h/daemons.h exist only under include/net/, not the bare paths these files #include), macros never defined (DAEMON, MAILDIR, POSTAL_D, THIS_MUD), and one file (mudlist_a.lpc) uses new/old as variable names, both driver-reserved keywords now. All of it is reachable only via dns_master.lpc's own create(), and dns_master is excluded from adm/etc/preload in the raw archive already (AGENTS.md §7.6's own standard exclusion) — confirmed genuinely dead, not a conversion regression.
• u/cloud/obj/npc/flower_girl/guihua.lpc: independently corrupted in the raw archive (confirmed via direct byte inspection of raw/, unrelated to this session's conversion pass) — the file's real // guihua.c source is preceded by ~128 bytes of binary junk, and later in the same file the content abruptly switches into what looks like a SECOND, different NPC's set() calls (a "杂货贩"/peddler) mid- statement, including embedded non-UTF8-representable binary bytes in a set("gender", ...) call. Not a simple dropped-character case — genuinely unrecoverable without fabricating content. u/cloud/ (a builder's personal workspace, containing dozens of .c.<unix- timestamp> backup/auto-save artifacts alongside the live files, e.g. d/snow/npc/rat.c.832813889) already shows other signs of interrupted/abandoned editing sessions from the original archive.
• d/green.n/npc/womana.lpc: genuinely truncated at the START in the raw archive too (wc -l on raw/'s copy: 4 lines, opening with a bare }) ); and no inherit/create() header at all). Not referenced by any other file in the tree (grepped) — dead/orphaned, left as-is per AGENTS.md §6.6 ("don't fabricate content").
• d/snow/test.lpc: literally plain English prose ("test again / just a test."), not LPC source at all — a builder's leftover scratch note, not referenced anywhere.
• F_UNIQUE and F_MERCENARY feature macros are referenced but never defined anywhere in include/globals.h (or any other header) — affects d/snow/npc/obj/old_book.lpc, d/canyon/bamboo/obj/ {parry_book,slipcase}.lpc, d/green/obj/jade.lpc (F_UNIQUE) and d/snow/npc/mercenary.lpc (F_MERCENARY). A genuine archive content gap (these inherits were presumably a shared feature-file that either predates this snapshot or was pulled from a different site branch) — not something to invent a base class for.
• feature/foodbak.lpc, obj/example/book.lpc: confirmed unreachable dead scaffolding. F_FOOD (the only macro anything inherits) points at feature/food.lpc; foodbak.lpc/foodold.lpc are never inherited by anything and additionally contain their own independent typos (foodbak.lpc's If( capitalization). obj/ example/ is explicitly example/documentation scaffolding (// An example book), and its one compile failure is an ancient 80-column hard-line-wrap landing mid-comment (// name of \nthe skill) — cosmetic, not live content.

Interactive test result — full registration flow

Registration shape: English id (3-12 letters) → confirm new-character (y/n) → real Chinese name (2-6 characters; check_legal_name()/ is_chinese() both needed the AGENTS.md §8.1 byte-range→codepoint-range fix, see below) → password (≥5 chars, entered twice) → email (free text) → gender (m/f) → lands in /d/snow/inn (饮风客栈). Verified live with a real Chinese name (秦风): look showed the full room description, exits, board, and NPCs (店小二/旅客×3/孙悟空); score showed a fully populated character sheet (age/gender/ganzhi birthdate, 精/气/神 bars, combat attack/defense, food/water, kill counts, potential %, overall score); quit produced a clean "你丢下一件布衣。...欢迎下次再来！" and correctly wrote both save files (data/login/f/fluffos.o, data/user/ f/fluffos.o) with zero runtime errors in debug.log.

§8.1 GBK byte-range Chinese-detection fix

adm/simul_efun/chinese.lpc's is_chinese() was the length-gate variant (strlen(str)>=2 && str[0] > 160 && str[0] < 255) and adm/ daemons/logind.lpc's check_legal_name() had the matching byte- calibrated length bound (strlen(name) < 2 || strlen(name) > 12 for a message that says "1 到 6 个中文字") plus an i%2==0 GBK-lead-byte loop gate. Both fixed per the AGENTS.md §8.1 playbook: is_chinese() now checks only the first character's CJK-Unified codepoint range (`str[0]

= 0x4e00 && str[0] <= 0x9fff`, guarded against an empty string);

check_legal_name()'s bound is now the literal 1..6 characters the message states, and the loop checks every character (no %2 gate). Live-verified: 秦风 (2 characters) registered cleanly on the first try.

Admin account

fluffos / Mud@2026, Chinese display name 秦风. Granted via this lineage's own mechanism: adm/daemons/securityd.lpc's create() reads adm/etc/wizlist (a plain <euid> <status> text file) into the in-memory wiz_status mapping every boot — added a fluffos (admin) line alongside the pre-existing lin (admin) entry, then registered fluffos through the normal flow. wizhood()/get_status() both read this mapping; (admin) status makes feature/command.lpc's enable_player() call set_path(ADM_PATH) + the real enable_wizard() efun. Verified: login banner shows "目前权限：(admin)", whoami shows User ID = fluffos, and update /adm/obj/master succeeds ("master: loaded successfully. 重新编译 /adm/obj/master.lpc：成功！") — the canonical read+compile ACL check per AGENTS.md §1.5.

WASM status update (2026-08-25, another session)

Promoted wasm_status from "" to playable — booted clean on the first try, zero fixes needed (unlike several sibling libs onboarded the same day, e.g. ds386/discworld/deadsouls_fluffos, which all hit an eager-simul_efun socket_status() compile failure specific to the WASM build's missing sockets package). Verified with a real scripted WASM session: login as fluffos/Mud@2026, arrival at the documented 饮风客栈 with the 孙悟空 crossover NPC present and active, and score producing the full character sheet (stats, 精/气/神 bars, combat ratings) matching the native-tested output exactly. quit wasn't distinctly captured in this transcript (孙悟空's own ambient action lines kept the client's idle-detector from settling, the same issue seen testing discworld's womble NPC) but is already verified clean under native testing above and untouched by anything WASM-specific.

Deep functional test (round two, AGENTS.md §10.7, 2026-08-27)

Prior work on this lib had only ever run a basic boot/registration smoke test (see "Interactive test result" above) plus the narrow §7.129 sibling-sweep fix (item 8 in "Config / driver-compat fixes"). This pass is the first FULL §10.7 playthrough: registration → look/score/i at every state change → movement → sect join (apprentice) → skill learning (learn) → safe combat (fight) → a full death/revival cycle → quit + debug.log check → reconnect after a real wall-clock gap, plus an explicit systematic grep for all thirteen standing cross-cutting bug patterns (§7.121/§8.3a/§7.122/§7.123/§7.124/§7.126/§7.129/§7.130/ §7.131/§7.132/§7.133/§7.134/§7.135). Test character: qinfeng/秦风二 (Test@2026), joined 封山剑派 (Fengshan sword sect) under 柳淳风, learned sword to skill level 2-3, went through a full smash-induced death → 白无常 ghost dialogue → reincarnation → revive-room cycle, then quit cleanly — save files removed post-test, only the seeded fluffos/秦风 admin account remains. Four bugs found and fixed:

1. §8.3a, feature/command.lpc's command_hook — declared private nomask int command_hook(string arg), registered via add_action("command_hook", "", 1) in the same file's enable_player(), but feature/command.lpc is inherited into std/char.lpc (the base of obj/user.lpc AND every NPC) — the textbook §8.3a shape, already confirmed broken the same way on 4 sibling ES2-lineage libs (demonangel, xuanjianlu, etc.) but never checked on es2 itself. Fixed: dropped private, kept nomask. 2. §8.3a variant, std/item/combined.lpc's destruct_me — also declared private, and inherited by std/money.lpc (every coin in the game), std/medicine/{pill,powder}.lpc, std/weapon/throwing.lpc, and several obj//d/ content items. Unlike the sibling demonangel instance of this exact file (where the call_out("destruct_me", ...) call site was dead/commented-out), here it is live: set_amount() does if( v==0 ) call_out("destruct_me", 1); — reached whenever a stacked item (most commonly money) is reduced to zero (spending your last coin, dropping/combining stacks down to 0). Fixed identically: dropped private, kept nomask. 3. New bug, not a prior AGENTS.md class — dbase key mismatch silently disables the wimpy auto-flee safety net for every player, forever: cmds/usr/wimpy.lpc (the player-facing wimpy [<percentage>] command) reads and writes the dbase key "wimpy" — but the ONLY consumer of the auto-flee threshold, std/char.lpc's heart_beat() (wimpy_ratio = (int)query("env/wimpy")), and every NPC's own create() (set("env/wimpy", N), ~30 files), all use the key "env/wimpy" instead. These are two entirely separate dbase entries on this driver's feature/dbase.lpc/treemap — running wimpy 30 as a player set "wimpy"=30, which nothing ever reads, while "env/wimpy" stayed permanently unset (0) for every player character ever created, so heart_beat()'s intp(wimpy_ratio = query("env/wimpy")) && wimpy_ratio > 0 guard was always false and the auto-flee branch never ran — the player-facing command was a complete, silent no-op from the very first player who ever typed it. Fixed by changing both the read and write in wimpy.lpc to use "env/wimpy", matching every other consumer. Live-verified: after wimpy 30, a fight against a training-hall NPC (武馆弟子) correctly triggered an auto-flee/retreat line ("你向後一纵，躬身做揖说道...") once damage pushed a stat below the 30% threshold — this did NOT happen pre-fix in code inspection terms (the key was never read) — and the save file correctly persisted "env":(["wimpy":30,]). Flagged for a sibling sweep: since this is the ES2 codebase's own feature/dbase.lpc/env/ naming convention (not something es1 shares — es1 uses a different stat system entirely), check haiyang2/demonangel/xkx2001/rzrmud/ xo/zhyx/naruto (the later ES2-derived family) for the same cmds/usr/wimpy.lpc vs. env/wimpy key mismatch — grep each for query("wimpy")/set("wimpy" (no env/ prefix) outside of std/char.lpc's own correctly-prefixed reads. 4. AGENTS.md §7.112 class, d/death/npc/{wgargoyle,bgargoyle}.lpc — both NPCs' init() unconditionally schedules a 5-stage call_out("death_stage", ...) narrative chain (ghost dialogue → reincarnate() → move to REVIVE_ROOM) with no re-entry guard. This is the exact, previously corpus-swept (150+ libs) §7.112 pattern under its most common filenames, but es2 was never part of either sweep wave. Fixed with the same death_stage_active set_temp()/ delete_temp() guard already applied to sibling haiyang2's copy of this same file (see AGENTS.md §7.112's reference implementation), adapted to each file's own exit points (bgargoyle.lpc has an extra "not actually a ghost yet, attack them" branch that also needed the flag cleared). Live-verified via a full smash→ghost→白无常 dialogue→reincarnate()→REVIVE_ROOM(城隍庙) cycle with zero debug.log/driver-log errors throughout.

Confirmed clean (checked, no fix needed):

• §7.121/§7.124 (float-in-int / fraction-vs-percentage): no int-typed economy/threshold function found doing unconverted float arithmetic; the two float count/count_total locals in feature/attack.lpc's biwu_ob() and cmds/std/biwu.lpc's start_biwu() are a deliberate fractional extra-turn accumulator inside a void function, not a narrow-return-type bug.
• §7.122 (autoload duplication): es2 has its own from-scratch feature/autoload.lpc (save_autoload()/restore_autoload(), NOT the TMI-2 compute_autoload_array()/destroy_autoload_obj()/ load_autoload_obj() lineage es1 shares with TMI-2-descended libs) — this driver's plain save_object() never serializes physical inventory contents at all, so there is no "captured twice" pathway: the autoload list is the ONLY mechanism that ever recreates a marked item, called exactly once per setup(), and ordinary reconnect (obj/user.lpc's reconnect(), adm/daemons/logind.lpc's reconnect()) never calls setup()/enable_commands() again — traced the full login/reconnect code path to confirm no double-restore_autoload() call site exists.
• §7.123 (bare file-scope IDENT = (...)): none found via manual inspection of every nosave TYPE ident;-then-bare-assignment candidate.
• §7.126 (stale .c extension in .o save data): es2 has no AREA/coordinate-grid engine and only 3 .o files total outside player saves (no corpus of exit-storing dbase files to be affected).
• §7.130 (unconditional query_idle() after non-interactive branch): std/char.lpc's heart_beat() correctly guards its query_idle() call with if( !interactive(this_object()) ) return; one line earlier — no unconditional post-branch call found.
• §7.131 (find_living/find_player without set_living_name()): feature/command.lpc's enable_player() calls set_living_name() on every login, and adm/daemons/logind.lpc's find_body() uses find_player() with a children(USER_OB)/getuid() fallback — a modern master-object-era design, not the classic per-call-scan archetype this bug class targets.
• §7.132 (map()-over-mapping wrong-arg binding): no map() call over a mapping with a single-parameter callback found in the tree.
• §7.133 (net_dead() never defined): both obj/login.lpc and obj/user.lpc define net_dead() (the latter private, but obj/user.lpc is a leaf nothing else inherits, so the §8.3a demotion doesn't apply — confirmed via a live abrupt-disconnect reconnect test, "重新连线完毕" printed correctly every time).
• §7.134/§7.135 (uninitialized-default accumulator/accessor): no new instances found beyond the std/room.lpc make_inventory() guard already fixed at onboarding time (item 6 in "Config / driver-compat fixes" above).

Verified via scripts/lpcc_check.sh (same 21 pre-existing FAILs as before this pass — all previously-documented dead/gapped content, zero new regressions) and a full native driver boot + the playthrough above, with zero debug.log/driver-stdout errors across the whole session.

深度功能测试（2026-09-04，round three，paid shop）

新角度：补上 2026-08-27 round two 没做过的真实购买。那次已经 apprentice 封山剑派柳淳风，本轮不再拜师。

修复的 programming bug

1. log_error() 把编译警告当「编译时段错误」广播给玩家（§15w）： adm/obj/master.lpc 的 log_error() 对 this_player(1) 无条件 write("编译时段错误：" + message)。本驱动把 warning: Unused local variable / Unknown #pragma 也送进 log_error()，第一次 登录刷了 34 条「编译时段错误」。闸门改成 this_player(1) && strsrch(message, "arning:") == -1，警告仍写入 work/log/log。修复后同路径 0 条玩家可见「编译时段错误」。 2. can_afford() 在没有白银时把任何非整两黄金的价格都判成「零钱」： feature/finance.lpc 的 1996-10-14 改写在 silver 缺失时直接 else if (amount%10000) return 2。身上已有足够铜钱支付 15 文包子 （一百文钱 + 一两黄金）仍报「你没有足够的零钱，而对方也找不开」。 补上「无银、有铜、铜钱覆盖 amount%10000」分支。修复后 buy 包子 from waiter 成功，找零 100−15=85 文。

黄金不能拆成铜钱仍是设计（纯黄金买 15 文会走同一句零钱提示）； vendor_goods 的键是中文「包子」不是 dumpling，buy dumpling from waiter 会「对方好像不愿意跟你交易」——list 列出的就是中文键，不是 bug。

实测过程

管理员 fluffos / Mud@2026。第一输入是「您的英文名字」，没有内码 菜单。端口 40214。落地饮风客栈 /d/snow/inn，店小二 id waiter。 孙悟空在场，ambient 不算 live clock，idle 0.45 够用。clone /obj/money/coin + call coin->set_amount(100)，list 包子十五文 / 鸡腿三十文 / 酒二十文 / 匕首五十文。修复后 buy 包子 from waiter： 「你向店小二买下一个包子。」i 包子 + 一两黄金 + 八十五文钱。

live debug.log 是 libs/es2/log/debug.log（Boot Time Fri Sep 4 03:48:47 2026），无 error: / Too deep recursion / F_FINANCE。 管理员存档未提交。

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

scripts/lpc_fix_negative_ranges.py: 5 line(s) in 2 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（§10.7，2026-10-07）— 兵器的 actions 函数指针

set("actions", (: WEAPON_D, "query_action" :));：旧 MudOS 把 (: 字符串, "函数" :) 当作 call_other 函数指针，本驱动把它当作表达式函数指针，求值只得到字符串 "query_action"。没有对应兵器技能时 feature/attack.lpc 直接用兵器的 actions，COMBAT_D 拿到的不是 mapping，reset_action() 后仍不是，记一条 bad action 就不出招——拿着这些兵器打不出攻击。改成别的兵器早已用的 (: call_other, WEAPON_D, "query_action" :)（KB 04 §6.10）：std/weapon/throwing.lpc、std/weapon/fork.lpc、std/weapon/whip.lpc、std/weapon/sword.lpc、std/weapon/axe.lpc、std/weapon/blade.lpc、等共 8 个。改动的文件在 HEAD 与工作树分别加载（新进程），无回退。

深度功能测试（§10.7，2026-10-08）— 编译警告整理

scripts/lpc_warnings.py es2 --fix 加 scripts/lpc_fix_no_effect.py --apply，共改动 180 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）：175 -> 175 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 21 -> 0；新进程登录（输入名字）正常。 手改 include/net/ftpdsupp.h：check_access() 的局部 file 只在 #ifdef GUEST_WIZARD_FTP 分支里用到，声明挪进同一个 #ifdef。

• 2026-10-10：编译警告第二轮（按驱动报告逐条修）：按位 & 改逻辑运算 0 处，返回类型/原型 1 处，NPC+F_UNIQUE 的 init() 0 个，带初始化的未用局部变量 0 个，只在 #if 分支用的声明移进分支 0 个文件；改动 1 个文件，HEAD 与工作树各载入一次无回退（1 objects; HEAD PASS 1, tree PASS 1; regress 0; newly loading 0），这些文件的警告 1 -> 0。

深度功能测试（§10.7，2026-10-12）：编译警告收尾

能编译的文件里的警告 4 -> 0（toolkit_fix.sh es2 重扫：1841 个文件可编译，剩下的诊断都在本来就编译不过的文件里）。四条都是 Function pointer returning string constant is NOT a function call：d/city/npc/walker.lpc 的 inquiry「买官」、d/city/shangshu/kefang.lpc、d/city/street3.lpc、daemon/class/scholar/taolin.lpc 的 item_desc 写成 (: "一段文字" :)。这不是函数名而是要显示的文字：feature/dbase.lpc 的 query() 会 evaluate() 函数指针、cmds/std/look.lpc 对函数型 item_desc 也是求值后 write()，结果都等于那段文字，所以直接写成字符串，行为不变。

另外 adm/daemons/network/mail_serv.lpc 的 tmp[i] == EOT+".";（上一行是 =，TMI-2 原有的笔误，语料里已普遍修过）改为赋值；该文件缺 daemons.h/uid.h/post.h，本来就编译不过，只是顺手改掉。

验证：5 个改动对象 HEAD 与工作树各用新进程载入：4 -> 4 通过（mail_serv 两边都载不进），regress 0；删除声明审计 0 项待看。5 个对象载入无回退。

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）：include/net/ftpdsupp.h。

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）

武功名拼错的 NPC（2026-10-19，KB 06 §7.258）

set_skill("kunlun-shnfa", …)、map_skill("dodge", "shaolin-shnfa") 这样的调用把武功名拼错了：本库没有这个武功文件，有的是拼对的那个。以前这种 NPC 在 create() 里报 No such skill 载入不了；武功缺失时跳过之后它们能载入，可是少了那门武功（set_skill 被跳过，map_skill 映射不到，门派弟子用普通轻功/拳脚）。按核对过的对照表改正，共 1 处、1 个文件：chaos-steos→chaos-steps（1 处）。（scripts/lpc_fix_skill_names.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/latemoon/in_room/flower1.lpc 的 north 出口：corridor7 → backroom

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。

call_out_info() 的每一项只有 3 个元素（2026-10-20，KB 06 §7.287）

驱动返回的是 ({ 物件, 函数, 延迟 })，不带 call_out 的参数，旧代码读第 4 项 info[i][3]（callouts 指令、calloutd、cod 工具）一读就报 Array index out of bounds，指令从第一行起什么也打印不出来。改为 (sizeof(info[i]) > 3 ? info[i][3] : 0)（scripts/lpc_fix_callout_info.py）。改动：/cmds/arch/callouts.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/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）。

sefun cat()：缺文件时 write(0) 报错（2026-10-20，KB 06 §7.297）

void cat(string file) { write(read_file(file)); }：文件不存在（压缩包没带的欢迎/公告文件，或只有运行中的 mud 才会写的日志）时 read_file() 回 0，write(0) 在本驱动里报 Bad argument 1 to receive()，调用者（登录对话、cat 指令）被中断；老驱动什么也不打印。改成 write(read_file(file) || "");（scripts/lpc_fix_cat_unchecked.py）。改动：/adm/simul_efun/file.lpc。

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。

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。本库有 1 个 NPC 文件写成 set("chat_msg", ({ }))（或 chat_msg_combat）并带正的 chat_chance（例如 /daemon/class/ronin/master.lpc），每次心跳有 chat_chance% 的概率 random(0) 为 0、msg[0] 报 Array index out of bounds（驱动捕获，心跳继续，但每次一条运行时错误），这个 NPC 在内存里就一直发生。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）。

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 1 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/village/lordhouse3.lpc（do_push）。最多的函数名：do_push 1 处。

(: 函数 :) 指针指向对象自己没有定义的函数（2026-10-22，KB 06 §7.355）

item_desc / inquiry / chat_msg_combat 里的 (: 函数名 :) 指向本对象（含包含文件、继承的父类）没有函数体的函数：编译不报错，用到时驱动报 *Undefined lfun pointer called: 函数名（look 某物、ask NPC about 话题 直接出错，战斗 / 闲聊里的指针在 NPC 的 chat() 里出错，这一拍心跳的后半段被跳过）。scripts/lpc_scan_callout_targets.py --kinds=lfun。 还剩 1 个函数没有补（chat_msg 1）：原型还在函数体丢了、或者根本没写过，没有可以直接补回的捐赠（别的库同一路径的文件里没有这个函数，或者它还要用本库没有的辅助函数、签名不同），不编造：/d/green.n/npc/oldman2.lpc:do_sigh。 另有 2 处函数体被作者整段注释掉（指针没跟着删），是有意关掉的，没动。

(: 函数名 :) 指针指向本库没有的函数：只含这个指针的行加了注释（2026-10-22，KB 06 §7.355 补充）

inquiry / item_desc / chat_msg* 里的 (: 函数名 :) 指向本对象（含包含文件、继承的父类）没有函数体的函数：用到时（ask、look、NPC 闲聊）驱动抛 *Undefined lfun pointer called: 函数名，这一拍心跳的后半段被跳过。这些函数没有可以补回的来源（别的库同一路径的文件里没有，或者它还要用本库没有的东西），所以 3 个文件里 4 行只含这个指针的行（(: do_sigh :), (: follow_player :), "剃度" : (: ask_for_join :),）加了注释，原文保留并写明原因，这一项当作没有这个话题 / 物品 / 闲聊条目（scripts/lpc_fix_dangling_lfun.py；#include 的文件在库里缺失的对象、指针不在整行上的写法没动）。 前提（function_exists()）和加载检查（HEAD 与工作区对比，本库改动的 .lpc 文件）的结果见 KB 06 §7.355：没有一个回归。

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 1 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：拜入少林 / 武当 / 峨嵋等派的玩家被改成错误的法名（二字名 张三 变成一个字的 澄，更长的名字取到第三个字）。

本库 1 处（1 个文件）：拼名字的赋值 1 处。例：name[0..1] → name[0..0]（/daemon/class/bonze/master.lpc:104）。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。

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 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。

example/book 注释折行无法加载（2026-10-27，KB 04 §6.25 Also）

obj/example/book.lpc 的 // name of the skill 等注释被折行，下一行 the skill 成了裸标识符，驱动报 unexpected L_IDENTIFIER。用手足 sjsh 完好副本覆盖。证明：HEAD 加载失败；改后加载成功。

深度功能测试（2026-10-27，§6.25 one-arg message_vision）

message_vision("…") 补第二参（me / TP / this_player()）。

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。