pd — technical notes

Source: gh repo clone ehershey/pd (cloned 2026-08-28). Mudlib root at libraries/beta-ii/ inside that repo (a bundled FMud/ Flash client and bin/ driver binaries/scripts, both siblings of libraries/ at the repo root, were never copied into raw/ — excluded by construction, not by a later filter). Number 952, port 40254. Status: done — boots clean natively, full registration-to-gameplay playthrough AND the restore/reconnect path verified live and repeatedly.

What this lib is

A real, historically-operated English-language MUD ("Primal Darkness", 1999-2010, mud.primaldarkness.com/www.primaldarkness.com), Nightmare mudlib lineage (adm/obj/master.lpc credits "Descartes of Borg 03 august 1993, Nightmare Mudlib Version 3.0"). One of the largest libs onboarded this session: 32,436 .c/.h files (72,244 total files including data/docs/logs), all pure ASCII (already_utf8=71949, converted=51, lossy=78 per convert_lib.sh — the 78 lossy bytes were all a single garbled —/bullet character in a handful of // author comment headers, purely cosmetic, no functional impact). No GB18030/BIG5 conversion needed at all.

Real, extensive, genuinely-authored content confirmed across dozens of domains: aekari, citrin, distopia, durst, forest, frozen, guilds, helgrath, helves, inferi_island, lodos, marshes, metaisland, pastarlon, roston, saahr, tirun, and more — not stock/placeholder content. Real historical player and wizard save data ships with the archive (adm/save/users/, adm/save/boats/, wizards/<name>/), which is why several early test registrations collided with pre-existing real historical account names (brennik, ashlyn both turned out to be real accounts from the original site — picked less-common fantasy names for later tests to avoid this).

4 bin/*.mud configs (primal/tna/whit/builders) in the excluded bin/ directory confirm multiple named instances of the same codebase historically — primal.mud was used as the template for this lib's config.fluffos (port/paths/limits adapted, everything else preserved).

§2.2 on-sight checklist

• master.lpc's load_object/get_root_uid/get_bb_uid all read normally; no get_include_path() existed at all — added one (see "Compile-time fixes" below).
• No securityd/is_chinese/check_legal_name — not applicable (English lib, no Chinese-name gating anywhere in this lineage).
• adm/db/preload.db preloads a normal, small daemon set (/daemon/command, castle_d, race_d, localpost, remotepost, letter, economy_d, events_d, a network daemon, web) — no dns_master/intermud entries.
• No hardcoded MUD_PORT/PORTNO constant anywhere (grep came back empty) — this lib reads its port purely from config.fluffos, so none of the §5.3 "port baked into source" trap applies.
• No command_hook private/nomask overrides; no efun::set/query/delete usage; no bare switch...default grammar issue; no ed_start/ ed_cmd/query_ed_mode usage.
• Genuine, dormant, on-demand outbound socket_connect() calls exist (daemon/plains_d.lpc's connect_plainswalk() to a hardcoded 209.85.145.196:3000, adm/daemon/link_d.lpc's connect_server() to 63.84.93.196:2349, adm/simul_efun/mail.lpc's SMTP relay, and a full Intermud-Chat implementation in daemon/imc_socket.lpc) — but, unlike imud's always-on I3 connection, none of these fire automatically at boot or preload time: they're all gated behind a player/wizard command (a "plainswalk" cross-mud portal feature, a mud-to-mud link-server handshake, sending in-game mail, and an IMC chat network respectively), and none of the daemons that own them are in preload.db. Confirmed no automatic network activity across several full boot+playthrough cycles. Flagging here per the §2.2 standing discipline anyway, since a future wizard actually running the plainswalk or mud-link commands would attempt a real outbound connection to a long-dead historical IP (harmless — it'll just time out — but worth knowing before assuming this lib is fully network-sandboxed like the rest of the corpus).
• Inbound-listening sockets confirmed harmless: daemon/network.lpc (UDP), daemon/webservices/main.lpc (port 12333), daemon/web.lpc (port 8500), wizards/whit/network.lpc (UDP 6123) — these just open extra local listening ports that nothing external will ever connect to in this environment; not a security or stability concern.

Conversion (convert_lib.sh)

Ran scripts/convert_lib.sh libs/pd/raw/beta-ii libs/pd/work (default GB18030 encoding — a no-op given the pure-ASCII source, confirmed before running by sampling files with a Python decode round-trip). Full run took ~26 minutes given the file count. Mechanical fixups the script applied: 21,629 literal .c" reference rewrites, 3,713 local angle-bracket #include <x.lpc> → quoted conversions, 1,017 files' static → nosave. Two script-vs-directory-structure edge cases found and hand-fixed afterward (neither is a new bug class, both are consequences of the rename step encountering something odd mid-run):

• A directory literally named myfile.c (wizards/kabaal/room/myfile.c/{room1,room2,room3,room4}.c) — find's -name '*.c' glob (no -type f) matched the directory too, renaming it to myfile.lpc in the same pass that was still trying to rename its now-stale-pathed children, so those 4 files were left as .c inside the renamed parent and never got the encoding/literal-.c-ref fixup pass applied. Manually renamed the 4 files and re-ran the literal-.c-reference fixup on just them (8 references fixed). These 4 rooms' own exits reference a typo'd myfile/roomN.c path (missing the trailing .c/.lpc on the directory segment itself) that never resolved even in the original archive — a pre-existing, unreachable dead-end in what is clearly a wizard's personal test scratchpad (wizards/kabaal/room/), left as-is per scope (content, not a conversion regression).
• A genuinely lost keep.h header: 68 files under wizards/whit/keep/ (#include <keep.h>) failed with "Cannot #include keep.h" — the file doesn't exist anywhere in the raw archive, not a conversion artifact. Reconstructed include/domains/keep.h from surviving evidence: ROOMS, GARDEN, and WEP are unambiguous (they mirror the surrounding directory layout, exactly matching this codebase's own convention — compare include/domains/tundra.h's SERVER/ROOMS pair), and GARDEN_SERVER was fully verifiable since wizards/whit/keep/rooms/garden/server.lpc already exists on disk — a mechanical restoration, not invented content.

Compile-time fixes (lpcc_check.sh)

Given the scale (~32K files), ran the memory-capped sweep in chunks (fresh lpcc --batch VM per chunk, ulimit -v capped) rather than one continuous batch — a single continuous process accumulates enough resident object memory to hit std::bad_alloc well before reaching the end of a corpus this size (confirmed: ~750 files consumed 8GB in one continuous run). Final full-corpus pass: 31,923/31,935 files covered, 29,610 pass, 2,313 fail (down from ~9,988 failures before any fixes — a handful of memory-dense pockets, mostly large coordinate-grid room areas, needed a second smaller-chunk pass to get full coverage; 12 files in one chunk that hit a hard abort even at 50-file granularity remain untested, all inside the same already-known-dead wizards/amun/builders/ staging tree described below).

Dominant class: base-class setters declared narrower than the corpus uses them (AGENTS.md §7.127)

~8,800 of the original ~10,000 failures (88%) came from just 8 functions in std/object.lpc/std/room.lpc/std/monster.lpc/ std/lacky.lpc/std/living.lpc/std/vault.lpc (plus their bak/ test/wizard-personal sibling copies, since this driver compiles every .lpc in the tree standalone):

| Function | Declared as | Corpus actually calls it with | Fix | |---|---|---|---| | set_id | string | array, 7,350+ sites | mixed, coerce scalar→1-elem array | | set_swarm | int | int OR string (mixed real usage) | mixed, no coercion (only used as a boolean flag) | | set_languages | string * | array OR scalar | mixed, coerce scalar→1-elem array | | set_alignment | int | int OR quoted numeric string | mixed + to_int() before arithmetic | | set_long | string | string OR function (dynamic descriptions — query_long() already functionp()-checks) | mixed | | set_short | string | string OR function (same as above) | mixed | | set_door_func | string (3rd arg) | string OR function (query_door_func() already declared mixed) | mixed |

Each was confirmed against every other real definition/override of the same function in the corpus before widening — in every case (set_id most starkly: every non-base override already declares string *) the base declaration was the sole outlier, not the callers. See AGENTS.md §7.127 for the general "how to apply" writeup, now updated with this lib as its largest confirmed instance.

daemon/command.lpc's dispatch table (AGENTS.md §7.118 — see "Two severe bugs" below)

Missing/shadowed local headers (AGENTS.md §6.1 family)

• keep.h — see "Conversion" above (genuinely lost, reconstructed).
• drowpath.h (wizards/stormbringer/roston/drowpath/) and vplain.h (wizards/nulvect/builders/vplain/) both exist on disk but are #include <x.h>'d from a subdirectory one level below where the header actually lives (rooms//mobs/) — outside what this lib's new master::get_include_path() (added since this lib had none at all) searches (its own directory only, not ancestors). 17 files total; fixed by converting those specific includes to absolute quoted paths (#include "/wizards/.../drowpath.h") rather than widening get_include_path() to search ancestor directories corpus-wide, since only these two cases were found.
• adm/save/boats.old/ (133 files, a stray backup copy of old boat source sitting inside what's otherwise a save-data directory) failed with "Illegal to inherit after defining global variables" — their shared #include <ocean.h> (which, unlike every other header these files use, declares a real initialized global string *ocean_map) came before inherit PIER; in file order. Fixed by moving #include <std.h> (needed for the PIER macro itself) + the inherit line ahead of the other two #includes in all 133 files — purely a reordering, no logic changed. This directory is unreachable from the live game (adm/save/ is data, not loaded content) but was cheap and mechanical to fix regardless.

Remaining ~2,300 failures: concentrated in confirmed-unreachable backup/draft directories

A grep sweep of the remaining failure list shows this is NOT a new systemic class — it's a long tail of syntax errors, MudOS-only-efun gaps (dump_socket_status(), old-style (: val1, val2 :) functional literals), and one duplicate-function-declaration case (wizards/amun/builders/marshes/ — 158 files with their own local sp_sap() override AND #include <amun.h>, which ALSO defines sp_sap() with different logic; the "live" copy of this same content under d/inferi_island/marshes/ has no such conflict), overwhelmingly concentrated in directories whose own names say "not live content": guildbak/, /old/, wizards/*/builders/, cmds/oldskills/, *.nul/*.bak-suffixed files. Spot-checked wizards/amun/builders/ marshes/ specifically: zero references to that path exist anywhere outside itself in the whole corpus — confirmed genuinely unreachable from the live room network, not merely low-traffic. Logged here per this project's content-vs-programming-bug scope line rather than chased file-by-file; none of it affects the live, playable game.

Two severe, corpus-wide bugs found and fixed

Both discovered because the standard §10.1 playthrough script initially only exercised fresh registration — the project's own §8.20/§8.22-style lesson about the restore path needing its own explicit test applied literally here.

1. Command dispatch totally broken (AGENTS.md §7.118, largest confirmed instance) — daemon/command.lpc's rehash() builds its verb→file table with choses[j] = choses[j][1..strlen(choses[j])-3], arithmetic written for the original .c (2-char) extension; after this collection's standard .c→.lpc (4-char) rename it silently registered every verb one character too long ("_look.lpc" → "look.l" instead of "look"), so literally every command except quit (registered directly via add_action(), not through this table) returned the driver's generic "What?" fail message for every player, from the very first boot. Fixed with the established -5 arithmetic. Verified live: before the fix, a fresh character's look/score/inventory/say all returned "What?"; after, all four (plus read list/pick <race> during creation) produced correct output.

2. Existing-account restore/reconnect totally broken (new, AGENTS.md §8.22) — adm/obj/login.lpc's check_password() needs to temporarily read the connecting player's own save file (to compare the stored password hash), via a classic "borrow root through the master object" handoff: master()->load_player_from_file(name, ob) → export_uid(ob); ob->restore_player(name);. Two compounding misreadings of what this driver's export_uid() actually does (confirmed against the driver's own f_export_uid() C source): it's a silent no-op unless the target's OWN euid is currently unset (every object already has one, from create()'s own seteuid(getuid())), and even when it fires it only ever sets the target's UID, never its euid — it's a one-time ownership-transfer primitive, not a privilege grant. So restore_player()'s own restore_object() call always failed master::check_access()'s permission check on /adm/save/users/... with *restore_object: read permission denied, uncaught, silently aborting check_password() (and everything queued after it) with zero trace in debug.log — a brand-new registration worked perfectly (it never calls this function at all), but reconnecting to ANY existing account — including one that had just registered and cleanly quit moments earlier — landed the player in a session with literally zero working commands, quit included, no error visible anywhere. Diagnosed via message()/debug_message() instrumentation (plain write() produces no visible output during this lineage's pre-body login sequence, hence its own code always using message("logon", ..., this_object())). Fixed with two additions mirroring this lineage's own working SAVE-side handoff (save_player()/actually_save_player(), which happens to work by accident of a different code shape): a new self-only nomask void clear_euid() { seteuid(0); } on the shared std/object.lpc base class, called by master.lpc's load_player_from_file() right before export_uid() (so the export actually fires and sets the UID to Root as intended), plus an explicit seteuid(UID_ROOT); added to std/user/save.lpc's restore_player() right before restore_object() (succeeds because the UID export_uid() just set makes master::valid_seteuid()'s own uid==UID_ROOT special case pass). Applied identically to master2.lpc and the std/test//wizards/nulvect/std/ sibling copies. Verified live end-to-end on two independent accounts: register → quit → reconnect with the same password → correct restoration of room/inventory/equipment, look/score/inventory/ quit all producing correct output (all of which returned "What?" before the fix).

Admin account seeding (§1.5)

Registered fluffos through the normal registration flow (real driver, real prompts). This lineage's wizard-status check is two-part — wizardp() (a real efun, checks an in-memory O_IS_WIZARD flag set by setup() whenever the saved position field is anything outside ({"player","high mortal","ambassador"})) AND archp()/admp() (wizardp(ob) && member_group(name,"superuser")) — so BOTH halves needed seeding: added fluffos to adm/db/groups.db's (superuser), (assist), (allowed), (creator), (approval), and (mudlib) groups (matching every real historical admin account's own membership pattern), and set fluffos.o's saved position field from the default "player" to "Root Avatar" (a non-mortal title, matching this lineage's own convention of colorful custom titles like "Archon Avatar"/"Realms Avatar" for its real admins). Verified live: logging in shows the extra Immortal/Arch news pages wizards get, and update /std/object succeeds ("/std/object: updated and loaded."), exercising both the read and compile halves of the file ACL.

WASM status

Not yet attempted — out of scope for this onboarding pass per the assigned task (native boot/playthrough only). wasm_status: "pending" in meta.json.

WASM measurement (2026-09-03)

meta.json was already playable from the 2026-08-31 deploy-unblock; the README still said "not yet packaged." Cold-boot under the shared ~/src/fluffos/build-wasm failed: eager adm/simul_efun/mail.lpc calls socket_create/socket_bind/socket_error (SMTP outbound), so the whole simul_efun object was not loadable. Wrapped those bodies in #ifndef __PACKAGE_SOCKETS__ no-ops (native unchanged). Wizard network.lpc / daemon/web.lpc socket failures remain graceful preload skips.

Verified with scripts/wasm_client.js: new character wasmpd → name-policy enter → y → Test12345 twice → male → [email protected] → real name → identity new → read list (30 races including human) → pick human → Tirun Central Park → score ("Newbie Wasmpd the boy", 600 gold, 70 stat / 5 skill points) → quit ("Reality suspended. See you another time!"). The seeded fluffos admin rejected Mud@2026 ("Invalid password"; driver also logged the old-crypt 8-char warning) — not chased; new registration is enough for the gallery path. Shop/combat/death were not exercised this pass. The MEMFS copy does not write wasmpd back to the host.

LPC formatter (§9)

Run 2026-08-31 (catch-up pass) — the original onboarding couldn't run node (PATH issue, since fixed via ~/.zshenv). Ran format-corpus.mjs over all 31,936 .lpc files under work/: 31,709 reformatted, 84 already-clean, 143 refused (messy legacy code). Ran all three §9 blind-spot checks: :: split (0 hits), case-label/comment merge (86 candidate spots across 8 files, manually diff-reviewed, all intact), and pre-existing-unbalanced-quote garbling (1 hit, std/obj/deed.lpc — a genuine original-source typo, fixed by hand and reformatted). Re-booted and re-verified the full live playthrough (registration, look, score, quit, restore/reconnect) — all correct.

§10.7 deep functional test (2026-08-31, round two)

Full continuous playthrough on the native driver (build-debug), going beyond the onboarding pass's registration/restore verification above — specifically to independently re-confirm the restore-path fix (§8.22) holds for a SECOND, unrelated account, and to exercise gameplay systems (race pick, class/subclass acquisition, safe-sparring, domain travel, long-sit daemon watch) the onboarding pass didn't reach. Test character: id Vulmarrow, password Test12345, male human, class Mage/subclass Sorceror — kept (not cleaned up) as a representative playthrough character. Save file: work/adm/save/users/v/vulmarrow.o.

Bug found and fixed: two objects pass a bare int to a string-typed setter, so this driver's strict compile-time type check rejects them and they never load

/d/nopk/standard/cemetary/mon/monument.lpc:12 and /d/nopk/standard/rain_forest/obj/statue.lpc:8 both called set_prevent_get(1) — std/object.lpc's real declaration is void set_prevent_get(string str) (494 other call sites across the corpus correctly pass a message string, e.g. fighter/dummy.lpc's set_prevent_get("You can't take that out of the ground!")) — so both objects hit Bad type for argument 1 of set_prevent_get ( string vs int ) and never compiled at all (*No program in object '...'! every time anything tried to new() them). This is the mirror image of the already-catalogued AGENTS.md §7.127 pattern (there, a shared base setter was declared narrower than the corpus's real usage; here the base declaration is correct and these two call sites are the outliers) — narrow enough (2 files, not a systemic class) that it doesn't warrant its own new AGENTS.md entry, but worth noting as the same general "literal-argument-type mismatch on an otherwise-correct setter" shape. Found via the long-sit boot watch below: a wandering errand boy NPC's move_around() heart_beat wandered into /d/nopk/standard/cemetary/gravyar9, whose reset() tried to new() the broken monument.lpc, throwing a real (if silent-to-players) error every reset cycle — exactly the class of bug §10.0 exists to catch (invisible to registration/login smoke testing, only surfaces once a lazily-reset room with wandering NPC traffic is hit). Fixed by giving each object a real message string matching the established convention ("The monument/statue is far too heavy to carry."). Verified with a targeted lpcc --batch recompile of just these two paths — both compile clean, no more FAIL/error lines — and confirmed no recurrence of the No program in object error across a fresh driver boot plus the full long-sit idle window below.

What was tested and confirmed working

• Registration and race pick: full flow (name → name-policy confirm → password → gender → email → real name → identity-profile prompt) into the race-selection void room, read list (all 30 races render), pick human → landed correctly in Tirun Central Park with starting gear (boots, chainmail, helmet, newbie guide) and 600 gold.
• Movement/exploration into 3 distinct domains: Tirun (walked the hand-built street grid — market square, Honin Road, Ganthus Road, Wayward/Parthos Roads — using a small BFS script built from parsing every room's set_exits() in d/nopk/tirun/, since the town is large enough that blind wandering wasn't practical); the North Forest (d/nopk/standard/forest1-3, reached via the north gate, a genuinely calm/monster-free newbie corridor by design); and Lodos (a full second city, reached via the /d/coach/ inter-city coach system — coach to lodos from the western coach house, correct in-transit flavor text, arrival with newbie fare correctly free — money stayed at 600 gold throughout).
• Safe-sparring found and used before real combat: d/nopk/tirun/ fighter/dummy.lpc (reached via enter fighter hall → south from the Sparring Hall) is this lib's designated safe target — its attack dummy handler is pure narration keyed off random(this_player ()->query_skill("melee")), no real damage exchanged either direction; used repeatedly with zero HP loss before any real fight was attempted (none was attempted this pass — see Death/respawn below).
• Skill/guild (class/subclass) acquisition, organic path: walked to the mage hall (enter mage hall off East Honin Road), preview → become mage (class set, skill points 5→25), then BFS-navigated to the sorceror hall (enter sorceror hall off East Wayward Avenue), preview → become sorceror (subclass set, skill points →31, alignment flipped to Godly, title updated to "novice sorceror"). spells/skills both rendered correctly afterward (a missile attack spell newly available, full skill-point table). No separate admin-only shortcut command for class/subclass assignment was found — grepped cmds/creator/ for a set-style utility and found only a generic environment-variable set (unrelated); the only path to a class/subclass in this lineage is the become command inside the relevant hall, whether run by a mortal or (via a wizard's generic force) an admin. Documented honestly rather than guessing at a shortcut that doesn't exist.
• quit, debug.log grep, reconnect after a real gap — 3 independent passes, specifically to re-verify §8.22 (found during onboarding) holds for an account other than the one onboarding tested: (1) same-session quit+immediate reconnect — correct password accepted, full state (class/subclass/HP/money/skills) restored, zero new debug.log lines; (2) killed and rebooted the native driver entirely (picking up the monument.lpc/statue.lpc fix), reconnected — same clean result; (3) after that reboot, left one idle connection open for the full ~220-second long-sit window (below), THEN reconnected a real ~5 real-minutes-elapsed gap later — password accepted, full state restored again, zero new debug.log growth across the entire boot+idle+reconnect+quit cycle (debug.log line count identical before and after). This independently reconfirms §8.22's fix is not an artifact of the one account onboarding happened to test.
• Shop/economy: the weaponsmith shop (enter weaponsmith shop off Ganthus Road) correctly lists its repair-only service via read sign (100 gold/level of disrepair — no list/buy inventory verb, a repair-only shop by design, not a bug); the inter-city coach fare system (above) is a real, working economy transaction (newbie-tier fare correctly waived, gold unchanged) — a full paid transaction (mid/elite fare, or a buy-from-peddler transaction with Rejik, whose reset()-gated appearance in square1 didn't line up with this pass's timing) was not reached; flagging this as budgeted, not fully exercised, per the project's own explicit fallback for shop/economy coverage.
• Death/respawn: not verified live this pass — time-budgeted out in favor of the long-sit watch and the 3-way reconnect re-verification above, both judged higher-value given this lib's already-known restore-path history. Flagging honestly as unverified rather than silently skipping it.
• Long-sit boot watch (§10.0, native-driver equivalent): no WASM build exists for this lib yet (§ WASM status below), so used the native equivalent — rebooted fresh (with the monument.lpc/ statue.lpc fix applied), opened one idle raw-socket connection, and sat through a full ~220-second window without logging in, while a wandering errand boy NPC's heart_beat continued triggering reset() on the (now-fixed) cemetery room in the background. Zero new debug.log entries during the entire window — confirms the fix above actually closed the only lazily-triggered failure this pass found, and surfaced no others.

深度功能测试（§10.7，2026-10-05）— 主控 author_file()/domain_file()

驱动的 mudlib_stats 包在主控载入时各调用一次 author_file(<主控文件>) 和 domain_file("/")，要求返回字符串。本库这两个 apply 只为 domains / realms 目录下的文件返回名字，其余返回 0，所以每次启动都打印 author_file() in the master file does not work, using root_uid as fallback 和 domain_file() in the master file does not work, using bb_ui as fallback 两行，网页版每位访客的终端里都看得到。adm/obj/master.lpc 改成与 fluffos 组织维护的 Dead Souls 上游（deadsouls_fluffos）相同的写法：其余文件的域返回 "BACKBONE"，作者返回 "NONAME"。域等于 backbone 域时驱动让对象沿用创建者的域，与返回 0 时相同；作者统计（author_stats()）多出一行 NONAME，汇总 realms 目录以外的对象（KB 06 §7.222）。用 lpcc config.fluffos /nonexistent.lpc 前后对比，主控载入段只少了这两行。

深度功能测试（§10.7，2026-10-06）— Windows 文本被当作 GB18030 解码（KB 03）

本库原文是 Windows 西文（cp1252）：弯引号 ’、é、»、ß 这类字节在入库转码时被当成 GB18030，两个高位字节合成一个汉字（owner抯、world抯），单个高位字节则连同后面的 ASCII 字节一起被吃掉（Aimeé went 成了 Aimewent，存档里的字符串少了收尾引号）。scripts/latin1_misdecode_repair.py pd --prefix beta-ii --encoding cp1252 --apply 按 raw 逐行恢复，共 98 个文件（61 个存档、33 个 .lpc、几份帮助文档）。

• 四个玩家存档读不出来。 aximili、bahamut、popov、ramona 的别名里有 Ç、å 之类的键，被吃掉的引号让 restore_object() 报 Illegal mapping format while restoring alias，这四位无法登录。修复后 61 个改动的存档全部读得出（HEAD 下 57/61）。
• doc/lpc/intermed/chapter1（Descartes 的 LPC 教程）是 Mac Roman，0xD5 是 ’，单独改正（cp1252 会给出 Õ）。
• wizards/seeker/asciis.txt 的 A1A1 真是 GB2312 全角空格，转码原本是对的，不改。
• 入库后改过的行脚本不动（23 行，都在房间描述里），其中 抯 一类第二字节为 ASCII 的汉字按字符还原为 ’s（11 个文件）。旧信件里的 cp437 方块画（圹圹）不动。
• 改动的 33 个 .lpc 用 lpcc --batch 编译，32 个通过；wizards/hermes/workroom 在 HEAD 也载入失败，与本次无关。

深度功能测试（§10.7，2026-10-07）— 编译警告整理

scripts/lpc_warnings.py pd --fix（3 万多个文件，单独跑；第一次被一小时的超时打断，先改了 nosave 函数——static 留下的 nosave 改成 protected，与 private 并列的去掉——再完整跑一遍）：无用局部变量 1081 个、裸 return 21 处、varargs 16 处等，共改动 681 个 .lpc。HEAD 与工作树分别加载（新进程）：621 -> 620，少的那个 d/citrin/rooms/lake/20,4 只删了两个没用到的局部变量，单独连载三次 HEAD 与工作树都通过，是批量加载的假象（KB 08 §10.13）。 未动：wizards/danovae/extra/_catchip.lpc 有一行被 ; 截断的字符串，但那个变量之后没再用，接不接都一样。

启动时警告 68 -> 0；新进程登录（标题画面、新人物）正常。

深度功能测试（§10.7，2026-10-08）— .lpc 后缀的定宽切片

.c 改名 .lpc 时只改了字符串没改切片宽度：f[<2..] == ".lpc" 拿两个字符比四个字符，永远不成立（!= 永远成立），按后缀筛文件、去后缀的代码因此从不运行或截成 x.l（KB 06 §7.238）。scripts/lpc_lpc_suffix_slice.py --apply 改了 10 处比较、8 处去后缀（如 std/vault.lpc）；改动的文件 HEAD 与工作树分别加载（新进程），无回退。

深度功能测试（§10.7，2026-10-10）— 编译警告整理

scripts/lpc_warnings.py pd --fix 加 scripts/lpc_fix_no_effect.py --apply，共改动 755 个 .lpc（多为删去未用的局部变量）。HEAD 与工作树分别加载（新进程）：686 -> 686 通过，无回退。启动时警告 0 -> 0；新进程登录（输入名字）正常。 （这一轮用分块扫描重跑：上一轮单次 lpcc --batch 中途死掉，只扫到 31936 个文件里的 2241 个；pd 的房间多，一个 VM 加载几百个就把内存吃到十几 GB，改成每批 600 个。）手改 wizards/venenum/book.lpc set_size()：每个分支里的 int maxpages = 5; 等是新声明的局部变量，遮住了同名的全局 maxpages，书的大小从来没被设上；改成给全局赋值 maxpages = 5;（该文件 HEAD 起就编译不过，属休眠内容）。新进程登录到 New Character 画面正常。

• 2026-10-11：编译警告第二轮（按驱动报告逐条修）：按位 & 改逻辑运算 4 处，返回类型/原型 64 处，NPC+F_UNIQUE 的 init() 0 个，带初始化的未用局部变量 5 个，只在 #if 分支用的声明移进分支 1 个文件；改动 63 个文件，HEAD 与工作树各载入一次无回退（63 objects; HEAD PASS 56, tree PASS 56; regress 0; newly loading 0），这些文件的警告 85 -> 6。

深度功能测试（§10.7，2026-10-11）：set_items 逗号键

• d/old/citrin.old/rooms/ 三个房间 set_items 的键写成 ("grip", "hand grip")，逗号只留最后一个名字；改为数组键 ({ ... })（std/room/items.lpc 会展开数组键）。载入无回退。
• 2026-10-14: adm/save/users/rid/seeker.o held two Latin-1 bytes in real_name, so restore_object() stopped with Invalid utf8 string. scripts/lpc_save_encoding.py transcoded it from Latin-1 (KB 03 §4).

Save files the onboarding transcode broke (2026-10-15, KB 03 §4.4)

/adm/save/boards/spell_idea.o, /adm/save/letters/2/1249616752.o, /adm/save/users/rid/kenshiro.o: the archive's Windows-1252 bytes (a \x92 apostrophe, ¿, ¡) were run through iconv -f gb18030 -c, which dropped an undecodable byte together with the quote after it, so restore_object() stopped with Illegal mapping format and the object started empty. Re-decoded from raw/ as Windows-1252 (scripts/lpc_save_encoding.py --from-raw); the driver restores each one now (scripts/lpc_save_restore_check.py).

Object paths with two leading slashes (2026-10-19, KB 06 §7.276)

2 file(s), 2 site(s) wrote an object path as //… (like carry_object("//clone/cloth/hic_cloth")): /d/events/4thjuly2008/rooms/icystuff/final.lpc, /wizards/pyro/personal/fireball.lpc. The driver these games ran on did not mind (the path was normalised first); this one hands the string to the master's valid_read() as written, securityd looks up //clone, the NPC's euid has no entry and the answer is Read access denied.: the NPC does not load. One slash now.

x == "a" || "b"：后面的字符串常量永远为真（2026-10-19，KB 06 §7.281）

if (arg == "yuyou" || "yu you") 里的 "yu you" 是非空字符串，永远为真，整个条件对任何 arg 都成立；作者要的是每个字符串都和变量比较。在每个字符串前补上变量和运算符（scripts/lpc_fix_literal_or.py；forrest2.lpc 是 dir == "north" && "south"，手工改成 (dir == "north" || dir == "south") && ...）：/d/guildbak/newethereal/ethereal/mob/homer.lpc（3 处）、/d/guildbak/newethereal/mob/homer.lpc（3 处）。

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

member_array('/', str) 里的 '/' 是整数，驱动按 UTF-8 字节搜索，返回字节偏移（带第三个参数时起点也是字节）；最新的 FluffOS（923cdbc0）在字符串里有非 ASCII 文字时，对每个调用点警告一次（Warning: member_array(int, string) searches BYTES）。而 str[0..idx-1] 按字符切，中文字符串会在错误的位置截断。改为 strsrch(str, "/")（找不到同样返回 -1，按字符计数；scripts/lpc_fix_member_array_char.py，JSON 编码器等少数几处手改）。改动：/d/events/dragonaug2008/rooms/py/2.lpc（1 处）。

if (x = 常量)：条件里写成赋值的比较（2026-10-20，KB 06 §7.290）

if (arg = "button" || arg = "按钮")、if (!arg || arg = "") return 0;、else if (check = 0 && tempB != "")：把 == 写成了 =，条件的值就是那个常量，分支对所有非零数和任何字符串（连 "" 都算）永远成立、对 0 永远不成立，变量还被覆盖；驱动对此没有任何警告。每一处 = 改成 ==（scripts/lpc_fix_cond_assign.py，共 16 处）。改动：/cmds/creator/_people.lpc（2 处）、/cmds/oldskills/_mudthrow.lpc（1 处）、/cmds/oldskills/_paralyze.lpc（1 处）、/cmds/oldskills/_roar.lpc（1 处）、/cmds/oldskills/_stun.lpc（1 处）、/cmds/skills/_mudthrow.lpc（1 处）、/cmds/skills/_roar.lpc（1 处）、/d/distopia/item/journal.lpc（3 处）、/d/nopk/tirun/clergy/druid/druid_grove.lpc（1 处）、/d/nopk/tirun/clergy/druid/druid_tree.lpc（1 处）、/d/nopk/tirun/clergy/druid/waterfall.lpc（1 处）、/d/nopk/tirun/warehouse.lpc（1 处） 等 13 个文件。

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）。

tail 指令：tail(file) 驱动里没有（2026-10-20，KB 06 §7.291）

cmds/*/tail.lpc 调用 MudOS 的 tail() efun，FluffOS 没有，指令编译不过。模拟 efun 文件里补了 tail(file)（Discworld 家族的兼容版：文件最后约 20 行，从完整的一行开始，存在返回 1；scripts/lpc_add_sefun.py tail）。

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）。

guild.h 重新定义 ROOM（2026-10-20，KB 06 §7.306）

std.h 定义 ROOM "/std/room"，domains/guild.h 又把它定义成行会房间基类（586 个文件包含 <guild.h>），首次登录时加载行会房间就打印 45 次 warning: Macro 'ROOM' redefined。guild.h 里在 #define ROOM 前加了 #undef ROOM（取值不变）。

首次登录控制台与核心目录的警告清理（2026-10-20，KB 06 §7.306、KB 04 §6.10）

登录体检（scripts/play_survey.py）里 pd 的控制台先是 45 条 Macro 'ROOM' redefined（domains/guild.h 重新定义 std.h 的 ROOM，已在前一段处理），去掉后露出下面这些。scripts/toolkit_fix.sh pd（跳过 /d/、/wizards/、/tmp/，核心目录 1,751 个文件）：float 赋给 int 的 152 处（多数是 percent() 的结果）改成 to_int()（98 个文件，其中 std/living.lpc:604、cmds/mortal/_score.lpc:9 登录就会编译到），3 处继承顺序、4 个 void 返回、2 个覆盖函数的返回类型、2 个多余的继承；加载检查 104 个对象，HEAD 通过 95、改后通过 103，无回退，新能加载 8 个。lpc_fix_macro_redef.py 给 22 处重复的 #define（SKILLS、BALANCE3_D、TREASURY_D、STATIC_SHIPS）前加了 #undef，lpc_fix_extra_semicolon.py（新）去掉 9 个类定义/函数体后多余的 ;（adm/include/death_cl.h、combo_cl.h、faction_cl.h、funccheck_cl.h、recipe_cl.h、adm/daemon/identity_d.lpc、cmds/skills/obj/netherlash_ob.lpc）。

真错误一处：adm/daemon/identity_d.lpc:356 的过滤函数写成 if (n && c->name != c) return 0;（拿名字和整个 class ConnData 比较，应为 c->name != n），驱动报 != always true because of incompatible types ( string vs class ConnData )，身份守护进程根本编译不了，而 std/user.lpc 登录时就要用 IDENTITY_D。改正后能加载（本批 135 个对象的加载检查里新能加载 9 个）。

还留着的：std/user.lpc:1580 的 query_autoload() 同时来自 /std/user/autoload.lpc（mixed *）和 /std/object.lpc（int），返回类型不同，做不了无副作用的覆盖，保持原样；oldcrypt() 的弃用提示；std/backpack.lpc（ARMOUR 与 BAG 两条路径各带一份 nomask clear_euid，没有任何文件继承它）；adm/daemon/ 里几个老守护进程缺头文件或宏（finger.lpc 的 FOLDERS_D、alchemist_d.lpc 的 materials.h、link_d.lpc 的 link_info.h），/d/ 与 /wizards/ 的 3 万个区域文件没有扫。

吟游诗人的三首歌不能编译：function(object t, …) 里写了 $1（2026-10-20，KB 06 §7.317）

cmds/song/fools_ballad.lpc、noxious_chorus.lpc、wracking_reel.lpc 里有 LDMud 风格的内联闭包 function(object t, string p) { … }，函数体里却用 $1 取第一个参数：$N 只属于 (: … :) 简写，/cmds/song/wracking_reel.lpc:61:7: error: $var illegal inside anonymous function pointer. 一共四处，随后一条语句 Undefined variable 'tp' 和语法错误，三首歌从来没有编译过。现在函数体用声明的参数名 t（三个文件各 4 处）；加载检查 HEAD 21 通过 / 3 失败，工作树 24 通过。

query_autoload() 同时来自 autoload 模块和物件基类（2026-10-20，KB 06 §7.318）

/std/user/autoload.lpc 有 mixed *query_autoload()（要恢复的物品清单），/std/object.lpc 经 /std/living 带来 int query_autoload()（物品的 autoload 级别），玩家身体两个都继承：驱动取第一个并在每次编译玩家身体时警告 query_autoload() inherited from both … using the definition in /std/user/autoload.lpc（首次登录控制台就有）。驱动的选择本来就是想要的，所以在 std/user.lpc 末尾显式定义 mixed *query_autoload() { return autoload::query_autoload(); }，运行时行为不变，警告消失。

_backstab 的编译警告：dmg *= 0.85（2026-10-20，KB 04 §6.10）

cmds/skills/_backstab.lpc 第 282 行 dmg *= 0.85;，dmg 是 int：驱动警告浮点被截成整数。改写成 dmg = to_int(dmg * 0.85);（结果与原来一样）。加载检查：通过，没有警告。

三个编译警告（2026-10-21，KB 04 §6.10）

include/imc.h 的 imc_wizard()：if(!interactive(ob)) 0; 缺了 return（上面的 imc_archp() 写的是 return 0;），改成 return 0;（非连线对象不再落到 wizardp(ob)）；daemon/network.lpc 的 read_callback()：tmp + ""; 本意是 tmp = "";（sscanf 只匹配到第一段时 tmp 没被赋值，后面的 explode(tmp, "||") 会报错），改成赋值；daemon/imc_d.lpc:404：if (split[2] == "accept") split[3] == "accept";（注释写的是 “fixes autoconfigure”）把赋值写成了比较，改成 =。三条警告都消失（imc、mudlist 命令的控制台行）；daemon/tmp/network.lpc 是 tmp/ 里的副本，没有改。

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

存档里写着 "/d/city/npc/wei.c" 这样的文件名；extract 时源文件已经全部改成 .lpc，而驱动对带扩展名的路径只认字面，new() / load_object() / move() 都找不到它，调用方多半把失败当成「没有这个东西」悄悄跳过。本库 32 个存档共 61 处按 scripts/lpc_fix_save_dotc.py 改成 .lpc（只改 .lpc 存在而 .c 不存在的字符串；玩家存档、备份、留言板、巫师目录和文件索引缓存不动）。改动：/adm/save/boats/aerion.o（2 处）、/adm/save/boats/amria.o（2 处）、/adm/save/boats/amun.o（2 处）、/adm/save/boats/arial.o（2 处）（等共 32 个存档）。

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

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

注册到不存在函数的 add_action，第二轮：宏路径父类和静态解析不了的对象（2026-10-21，KB 06 §7.349）

第一轮的静态解析读不了经宏路径继承父类的对象（#define ROOM STD_DIR "room/room" 这类定义的正文以另一个宏开头）和它解析不了的继承，这些对象当作 UNSURE 没有扫；第二轮修了解析器，重扫并对这些对象在运行时问 function_exists()。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 10 条注册（10 个文件，scripts/lpc_fix_dead_add_action.py），动词留给别的处理函数：/d/events/dragonaug2008/rooms/py/5.lpc（jump_north）、/d/freepk/merkkerri/demonquest/dquest10.lpc（salvation）、/d/guilds/dw/hall.lpc（fstart）、/d/guilds/lastsaints/objects/spare_hall_rooms.lpc（homeroom）、/d/guilds/lastsaints/rooms/armour_room.lpc（go_west）、/d/nopk/standard/lorroad2.lpc（read）、/d/nopk/tirun/electionhall.lpc（read）、/d/nopk/tirun/monastery.lpc（confess）、/d/nopk/tirun/nranch.lpc（touch）、/d/nopk/tirun/obj/misc/deliv_warehouse.lpc（read）。最多的函数名：read 3 处、jump_north 1 处、salvation 1 处、fstart 1 处、homeroom 1 处。

守护进程里 int clean_up() 盖在基类 void clean_up() 上的警告（2026-10-21，KB 04 §6.10）

/d/events/dragonaug2008/rooms/instance_d.lpc、/adm/daemon/instance_d.lpc、/daemon/door_d.lpc、/wizards/pyro/sandbox/queue_d.lpc、/wizards/pyro/assgn/implemented/door_d.lpc 的 int clean_up() 覆盖 /std/daemon.lpc 的 void clean_up()，驱动编译时警告 Function clean_up inherited from '/std/daemon.lpc' does not match current function in return type ( void vs int )（共 7 处）。clean_up() 的返回值驱动会读（返回 0 就不再被询问，void 函数就是返回 0）：这几个守护进程靠 return 1 让驱动下次再来，数据清空后 return ::clean_up() 才销毁自己，所以不能改成 void；pd 里覆盖 std/daemon 的 clean_up() 是 int 和 void 各 6 个，基类也不好改。覆盖函数改声明为 mixed clean_up()：返回值不变，类型检查不再触发。重新加载这 5 个文件（连同本轮其他改动的文件共 15 个）：0 警告，无回退。

缺失物品的 new() / carry_object() 调用链加了保护（2026-10-22，KB 06 §7.73 补充）

new(路径)->move(...); / carry_object(宏("x"))->wear(); 这类语句，路径指向的文件在这个压缩包里没有：new() 返回 0，->move() 在 0 上报 Bad argument 1 to EFUN call_other()。在 create() 里这个对象永远加载不了，在 reset() 里第一个缺失物品之后的刷新每次 reset 都被跳过。95 个文件里 287 处（new）改成 { object new_ob = new(路径); if (new_ob) new_ob->move(...); }（scripts/lpc_guard_carry_object.py --new：路径按这个文件自己看到的宏解析，宏名冲突、虚拟对象路径一律跳过）。 对这 95 个改动文件做加载检查（HEAD 与工作区对比）：HEAD 能加载 15 个，现在 71 个（新增可加载 56 个，回归 0 个）。

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

carry_object("d/city/obj/jiasha")->wear(); 这类路径不带前导斜杠的写法在这个驱动里按根目录解析（new("d/city/obj/changjian") 能加载 /d/city/obj/changjian），所以和带斜杠的一样：文件不在压缩包里时 new() 返回 0，->wear() 在 0 上报 Bad argument 1 to EFUN call_other()，整个 NPC 加载不了。此前的保护工具把这种写法当作「相对当前目录，没法判断」跳过了。3 个文件里 3 处（d/standard/frealmz/mob/anathar、d/standard/frealmz/mob/roger、d/standard/frealmz/mob/talon）改成带保护的块（scripts/lpc_guard_carry_object.py）。 对这 3 个改动文件做加载检查（HEAD 与工作区对比）：HEAD 能加载 0 个，现在 3 个（新增可加载 3 个，回归 0 个）。

log/ 下的日志子目录没有人建：写日志的 write_file() / log_file() 报错（2026-10-22，KB 06 §7.358）

这个库的代码往 log/ 的 1 个子目录写日志：log/log（std/bak2/user_new.lpc）。压缩包、git clone 和站点包里都没有这些目录（本地启动没有写过它们，gen_keep_dirs.py 没记），write_file() 对不存在的目录抛 *Wrong permissions for opening file /log/dlog/money for append. "No such file or directory"，调用它的函数在写日志那一行中断，后面的奖励、标志、步骤都没有执行。测试对象里本库的 log_file() 什么也没写（探针对象没有写权限，log_file() 自己也不建目录），所以真实调用者会得到什么没有测出来。把 log/log 记进 scripts/wasm_keep_dirs.txt（scripts/lpc_keep_log_dirs.py；站点和 pristine 树按它建空目录）。证明（scripts/lpcc_eval.py，在 /adm/daemons 这类目录下的求值对象里写 /log/zzdirprobe/...，新进程）见 KB 06 §7.358 的实测表。

写死的 "/路径"->函数() 指向库里没有的文件：改成现存的同名文件（2026-10-22，KB 06 §7.359）

4 处语句级的 call_other 字面量路径，库里没有这个文件（区域改了名 / 大小写不对 / 差一个字母 / 文件挪了地方）："/d/standard/quest_room"->list_quests()（/d/freepk/merkkerri/jungle/gcity/gcity29.lpc、/d/nopk/arlon/room/fighter_hall.lpc）→ "/d/nopk/standard/quest_room"。call_other() 对不存在的对象抛 couldn't find object，所在函数当场中断。改成库里现存、并且定义了被调用函数的文件（scripts/lpc_fix_call_path.py；巫师目录、备份目录、名字不同的兄弟文件都不算候选）。

wizards volcano1 丢分号（2026-10-27，KB 04 §6.25 Also）

三份 wizards/**/room/volcano1.lpc（leviathas / nested leviathas / levi aciri）写成 inherit ROOM 后无 ;，且 ::create() / set_* 全无分号，驱动报 unexpected L_BASIC_TYPE。补回分号后可加载。同作者 deatblade 更残，未动。证明：HEAD catch 失败；改后加载并 query_short 正常。

deatblade 丢分号 + wep/terra.h 错路径（2026-10-27，KB 04 §6.25 / KB 06 §7.359 Also）

三份 wizards/**/wep/deatblade.lpc 同样丢光分号，且 set_short 颜色串被折行。补回 ; 并接好 short。另：wep/terra.h（及同错的 item/terra.h）把 WEP/ARM 指到不存在的 weapons//armour/，new(WEP "deathblade.lpc") 得 0 后 ->move 炸；改为与 room/terra.h 一致的 wep//arm/（darktown 用本地前缀）。证明：HEAD 语法失败；改后三份均加载。

manex tower8 缺 inherit 分号（2026-10-27，KB 04 §6.25 Also）

wizards/manex/tower8.lpc 写 inherit ROOM 后无 ;，驱动无法编译。补上分号（空房间 stub 保留）。证明：HEAD 失败；改后加载成功。