Dead Souls 3.0 -- porting notes

Source: a ZIP archive (ds3.0/, 5553 files, WINDOWS_INSTRUCTIONS.txt at top level) from mudbytes.net, a now-dead site. Recovered via the Wayback Machine (<https://web.archive.org/web/20190802014424id_/http://mudbytes.net/files/download/2733/>), re-verified live (HTTP 200, real ZIP magic bytes, 5553-file/ds3.0/ top-level-dir match) at the start of this session before extracting. The zip bundles fluffos-2.18-ds07/ (a driver source snapshot), bin/, win32/, extra/ build/install scaffolding -- all ignored, this project uses its own driver. The real mudlib root is the zip's lib/ directory (bin/mudos.cfg's own master file : /secure/daemon/master confirms it). Slug dsIII, number 182, port 40229.

1. Duplicate check -- distinct release, but the closest sibling by far

This collection already has FOUR Dead Souls entries: dsI (~1996-98, Nightmare-IV-era, 2 domains), dsII (2006, "Dead Souls II", 4 domains), and ds386/deadsouls_fluffos (3.8.6, 2014-era, 9 domains, confirmed identical codebase generation to each other). Before building out a full pipeline, checked whether "Dead Souls 3.0" repackages any of the four:

• Self-identification: doc/RELEASE_NOTES has a dated version history culminating in a ---- 3.0 ---- entry ("Released 2009-12-30... Driver: added Kalinash's fix for Solaris on 64 bit SPARC... Added commands: anchor, unanchor..."), preceded by 2.11a20-2.11a22 entries -- a real, self-contained, later point release, not a copy of dsII (Nov 2006) or a re-save of ds386 (2014). README.txt's own title text is a stale "Dead Souls 2... Released December 2005" banner (last touched Jan 2008) that predates this zip's actual RELEASE_NOTES contents -- a rolling-tarball snapshot where the README just wasn't kept in sync, not evidence this is actually the 2006 "II" release.
• Domain count: 7 domains (campus, cave, default, examples, Praxis, town, Ylsrim) -- more than dsII's 4, and a STRICT SUBSET of ds386/deadsouls_fluffos's 9 (missing only amigara and learning, both added in later 3.x point releases per their own RELEASE_NOTES entries). dsI has only 2. 4598 files under lib/ here vs. dsII's 1937, ds386's 4727.
• secure/daemon/master.c comparison -- the decisive evidence: 1157 lines here vs. ds386's 1157 (identical length) and dsII's 781/ dsI's 749 (a different, much smaller shape entirely). A real diff against ds386's raw archive shows only 8 changed lines total -- all tiny game-balance/logging tweaks (a debug_message() string, a get_stack(1) vs get_stack() arg, a clean-up percentage 26 vs 34, an array-size cap 500 vs 1024, an error-counter variable) -- vs. dsII's real, substantive divergence (added globals, a rewritten load_access() body). Against deadsouls_fluffos's raw copy, the same file differs by 223 lines -- meaning this archive's master.lpc is closer to ds386's than ds386 and deadsouls_fluffos are to EACH OTHER (189 lines apart, two nominally-same-version 3.8.6 copies from different sources).
• Whole-tree file-identity check: of the 4555 relative paths shared between this archive and ds386's raw copy, 4136 (90.8%) are byte-identical; against deadsouls_fluffos the rate is only 63.3% (2874/4542). 43 files exist only here (mostly domains/campus/adm/* admin scratch tools and cached domains/town/virtual/sky/*.c coordinate-instance files); 172 exist only in ds386 (new content added in later point releases, including the two newer domains).

This is a genuinely distinct release -- not a byte-identical duplicate, since 419 of the 4555 shared files differ and 43 are unique here -- but it sits MUCH closer to ds386/deadsouls_fluffos in the lineage than dsII ever did: this is best understood as an early-2010s snapshot of the exact same "3.x" branch that later became 3.8.6, captured about 4-5 years and several point releases earlier. duplicate_of is null in meta.json, but the whole fix catalog below was ported from ds386's own NOTES.md wholesale (verified per-file via diff against ds386's raw archive before applying each fix) rather than rediscovered from scratch, since nearly every driver-compat bug turned out to be byte-identical.

2. Conversion

English-language archive (public domain, from mudbytes.net) -- scripts/convert_lib.sh against raw/ds3.0/lib confirmed this: 4562/4598 files already valid UTF-8 (pure ASCII), only 5 lossy conversions, 28 binary files skipped. 2290 .c files renamed to .lpc, 603 literal .c" references fixed, 31 local angle-bracket #includes converted to quotes, 1429 files touched by the static->nosave sweep.

2.1 One source file was genuinely MIXED-encoding, not GB18030 -- and the lossy conversion broke its string-literal syntax

secure/sefun/translate.lpc's FLUBS array (the same multilingual "drunk talk" curse-word list ds386/dsII both have) contains:

• Two Finnish words with literal Latin-1 bytes ("kyrpä", "räkä", 0xE4 = ä).
• One word that was already valid UTF-8 in the raw archive: "ХУИТА" (Cyrillic, 5 characters, \xd0\xa5\xd0\xa3\xd0\x98\xd0\xa2\xd0\x90).

A single GB18030 pass mangled BOTH differently: the Cyrillic bytes (already-correct UTF-8) got mis-decoded as GB18030 multi-byte sequences into garbage CJK characters, and the Latin-1 ä bytes got silently dropped as invalid, which ate the closing quote immediately after kyrp too ("kyrp\xe4" -> "kyrp" was NOT what happened -- the actual result was "kyrp,"r鋕,"shawtay", a real syntax break: an unterminated string swallowing the next comma, then an invalid bare identifier r鋕). This is a genuine, driver-compat-relevant compile failure, not just cosmetic mojibake -- confirmed by a first driver boot failing simul_efun's compile entirely (*No program in object '/secure/sefun/sefun'!). Fixed by hand: re-decoded the Finnish words as Latin-1 and re-typed the Cyrillic word as literal UTF-8, leaving everything else in the file (99% pure ASCII) untouched. Also individually re-decoded with iconv -f ISO-8859-1 -t UTF-8: doc/CREDITS, doc/help/players/credits (a Lars Pensjö name-with-diacritic), and domains/town/txt/mystery.txt (a ¯\(°_o)/¯ ASCII-art shrug). One more lossy file, secure/save/functions.o, turned out to be a stale function-signature cache (secure/daemon/function.lpc's RestoreObject()/file_exists() guard handles a missing file gracefully and rebuilds on demand) -- deleted rather than hand-repaired, same precedent as ds386's functions.o/books.o cache files.

3. Driver-compat fixes -- ported wholesale from ds386's catalog, verified per-file

Every fix below was checked against ds386's raw archive first (diff on the specific file) to confirm the bug is byte-identical before porting ds386's exact fix; all were.

• Missing get_root_uid()/get_bb_uid()/creator_file() master applies (AGENTS.md section 7.2, and the creator_file() extension first cataloged on ds386): this driver's PACKAGE_UIDS build requires all three; missing entirely. Added byte-identical stub implementations (all constant "root"/"backbone") to secure/daemon/master.lpc, same placement as ds386.
• log_error()'s load_object(WEB_SESSIONS_D) re-entering the compiler mid-compile (AGENTS.md section 7.60's pattern): same fix as ds386 -- swapped to find_object().
• A stale runtime_config.h INT-slot-numbering mismatch (AGENTS.md section 7.89, ds386 is the origin case): this lib's own secure/include/runtime_config.h computed BASE_CONFIG_INT = BASE_CONFIG_STR + 15 instead of the canonical driver's RC_BASE_CONFIG_INT = RC_LAST_CONFIG_STR + 1 (=256) -- every get_config(__MAX_EVAL_COST__)/__MAX_BYTE_TRANSFER__/ __MAX_READ_FILE_SIZE__/__MAX_STRING_LENGTH__ call (identical symbol set to ds386's) would read the wrong config slot. Fixed by dropping in the driver's canonical header, same as ds386.
• OLD_ED vs. this archive's split ed_start()/ed_cmd()/ query_ed_mode() API: lib/editor.c and secure/cmds/creators/lsed.c are BYTE-IDENTICAL to ds386's raw copies, and lib/interactive.c/lib/nmsh.c differ from ds386's by only one unrelated line (a valid_event() security check added later in interactive.c) -- so ds386's exact rewrites were copied/ported directly rather than re-derived: lib/editor.lpc (full file copied from ds386's fixed version), secure/cmds/creators/lsed.lpc (full file copied, stubs the batch-editor feature as unavailable), lib/interactive.lpc's dead ed_cmd(".");ed_cmd("x");ed_cmd("Q"); quit-path block removed (redundant under OLD_ED's automatic save_ed_buffer() on destruct), lib/nmsh.lpc's query_ed_mode()- gated custom ed-prompt branch collapsed to dead code (if(0){...}, unreachable since the driver prints its own ed prompt directly from C under OLD_ED).
• #define __DIR__ "..." colliding with the native compiler predefine: 13 files here (domains/{town,default,Ylsrim}/virtual/ *.lpc) still had the old manual #define __DIR__ idiom that ds386's OWN copies of these same files had already dropped in a later point release (ds386's town/default/Ylsrim virtuals have no __DIR__ define at all -- confirms this archive predates whatever point release removed them, consistent with section 1's "same branch, earlier snapshot" finding). Removed all 13; the native __DIR__ produces an identical value.
• secure/cmds/admins/sconv.lpc's #ifndef __FLUFFOS__ guarding the wrong symbol (this driver predefines bare FLUFFOS, not __FLUFFOS__): byte-identical raw file to ds386's; same one-line fix (__FLUFFOS__ -> FLUFFOS). Left the other ~9 files using the same guard untouched, matching ds386's "no error, no fix" precedent -- all compiled and ran clean.
• secure/include/lib.h's LIB_REMOTE macro pointing at a nonexistent path (/lib/remote when the real file lives at secure/lib/net/remote.lpc): same bug, same fix as ds386. LIB_CAPTURE (/lib/capture, no implementation anywhere in the archive) has the same pre-existing gap as ds386 -- left unfixed, breaks only domains/Ylsrim/broken/jar.lpc (a file literally shelved in a broken/ subdirectory in the original archive).
• **SetLong/SetExternalDesc/SetInternalDesc declared string-only when their Get*() counterparts already support functionp()/ arrayp() dynamic-description values** -- 44 real call sites use the function-value form here (more than ds386's reported failure count, since this archive's cave/examples/Praxis domains have their own additional users of the idiom). Widened SetLong/SetExternalDesc in lib/events/look.lpc, SetInternalDesc+InternalDesc's declared type in lib/events/look_in.lpc, and lib/std/room.lpc's SetLong wrapper, all to mixed/mixed * -- exact same fix and files as ds386 (its own lib/include/{room,interactive,door}.h forward declarations were NOT widened either, matching ds386's working, tested precedent that the header prototypes aren't enforced against the concrete override here).
• **daemon/books.lpc's LoadChapters()/lib/std/book.lpc's eventLoadChapters() declared mixed * (array) when the real payload is a mapping**: same bug, same 3-declaration fix as ds386 (daemon/books.lpc, lib/std/book.lpc x2).
• domains/Praxis/supply2.lpc's set_pre_exit_functions( ({...}), ({...}) ) (a #define-aliased SetProperty(string, mixed) call given two arrays): byte-identical file to ds386's (differs only in whitespace); every sibling room in the same directory already has the identical call commented out. Applied ds386's exact fix (comment out with the same explanatory note).
• secure/sefun/sockets.lpc's foreach (array item in finalsocks) -- the same dump_socket_status-in-simul_efun array-dialect pattern flagged as recurring across sibling libs. Fixed to foreach (mixed *item in finalsocks).

4. TYPE array NAME dialect -- 166 files via an automated fixer, plus 7 hand-caught shapes the fixer's regexes didn't cover

Wrote a quote/comment-masking Python regex fixer (handles single-name declarators with/without default values, multi-name comma lists, bare array NAME/array *NAME, and (TYPE array) cast forms) and ran it across the whole work/ tree: 166 files changed on the first pass (secure/cmds/admins/admintool.lpc's array *classes;/array *races;, secure/sefun/sockets.lpc, secure/sefun/sefun.h, and 163 more). Then a full driver boot + a standalone re-grep surfaced 7 more shapes the regex set didn't anticipate, each fixed individually and verified there was exactly one instance:

• lib/body.lpc:1810: class MagicProtection array GetMagicProtection(){ ... } -- a class TYPENAME array FUNCTION RETURN TYPE, not a variable declaration. The generic fixer's "bare array NAME" rule actually MIS-FIRED here (matched the space before array regardless of the preceding marriage/MagicProtection token, since its negative lookbehind only checks the single character immediately before array) and produced the WRONG result (class MagicProtection mixed *GetMagicProtection() -- inserting a spurious mixed). Caught and corrected to class MagicProtection *GetMagicProtection(){ ... } by hand; grepped the whole tree for the same class \w+ mixed \* mis-fire pattern and confirmed no other instances.
• secure/daemon/finger.lpc:23: private class marriage array Marriages; -- same class TYPENAME array shape, in a variable declaration this time; hit the SAME fixer mis-fire (private class marriage mixed *Marriages;). Corrected to private class marriage *Marriages;.
• lib/guard.lpc:6: private nosave array PendingGuard = ({}); -- a bare array NAME = default (no element type, WITH an initializer); the fixer's bare-array rule only matched a terminator of ;/,/) immediately after the name, not =. Fixed to private nosave mixed *PendingGuard = ({});.
• lib/std/story.lpc:20,65: array GetTaleKeys(); / array GetTaleKeys(){ return keys( Tales ); } -- bare array as a FUNCTION RETURN TYPE (prototype and implementation). Fixed both to mixed *GetTaleKeys().
• secure/sefun/sockets.lpc:14: foreach (array item in finalsocks) -- covered in section 3 above, same root pattern as ds386's recurring dump_socket_status-class foreach.
• secure/cmds/players/history.lpc:32: string *arr, tmp = ({}); -- the ORIGINAL (pre-conversion) source was string array arr, tmp = ({});, a multi-name declarator list WITH a default value on the second name. The fixer's multi-name rule required no = in the matched span, so it never touched this line at all, and it slipped through to a real compile failure (Type mismatch ( string vs mixed * ) when initializing tmp) since a bare string *arr, tmp only stars the FIRST name -- tmp stayed scalar string while assigned an array. Fixed to string *arr, *tmp = ({}); (both starred, matching the old dialect's "applies to every name in the list" semantics, same as ds386's NOTES.md documents for its own admintool.lpc finds).

All 7 fixes verified individually: a clean re-grep for each pattern class found zero remaining instances, and a full driver boot + full lpcc_check.sh batch sweep both came back clean of these specific error signatures afterward.

5. Live verification

Booted clean (Accepting telnet connections on 0.0.0.0:40229., Initializations complete., zero fatal errors, zero uncaught compile errors in log/debug.log).

Confirmed THREE independent real outbound network side effects at boot -- more than any prior Dead Souls-lineage lib in this collection, and the third one is NEW to this archive specifically:

1. IMC2 (secure/daemon/imc2.lpc, preloaded, DISABLE_IMC2/ DISABLE_INTERMUD both 0 in secure/include/config.h): makes a real socket_connect() a couple seconds after boot. Confirmed live (ss -tnp showed ESTABLISHED ... 204.209.44.3:8080). 2. Intermud-3 (daemon/intermud.lpc, preloaded): a second real connection attempt, confirmed live (ss -tnp showed SYN-SENT ... 74.207.247.83:5000). 3. NEW: an automatic release-notes HTTP fetch on every single boot, with zero player interaction required. secure/daemon/autoexec.lpc is ITSELF preloaded, and its create() schedules eventRun() via call_out(..., 1) -- one second after preload completes, with a hasrun guard so it only fires once per boot. eventRun() unconditionally does catch( load_object("/secure/cmds/creators/dsversion")->cmd());, and dsversion.lpc's cmd() immediately opens a real socket to dead-souls.net (97.107.133.86:80) and fetches /RELEASE_NOTES, overwriting doc/RELEASE_NOTES_HTTP with the live response. Confirmed live and reproducible on a bare idle boot with NO login at all: doc/RELEASE_NOTES_HTTP's mtime changed within ~14 seconds of driver start with zero connections made, and its content is a REAL, CURRENT response from a still-live server -- the fetched response includes release notes up through "3.9, Released 2020-04-08", meaning dead-souls.net itself is not dead (unlike mudbytes.net, this project's actual archive source). A SEPARATE creator command, secure/cmds/creators/rss.lpc, has the same eager-create()-time network behavior (registers RSS feeds from lpmuds.net/feedburner.com via call_out((: Setup :), 2)) but is NOT itself preloaded -- it only fires if something actually loads the file, which an lpcc --batch compile sweep does (merely compiling secure/cmds/creators/rss.lpc as one of thousands of files in the sweep triggers its create() and, 2 seconds later, its own real HTTP/RSS fetches) -- confirmed by doc/RELEASE_NOTES_HTTP and tmp/imc2.log both getting freshly rewritten after this session's lpcc_check.sh runs, with no driver boot in between. This means BOTH a live boot of this lib AND a bare lpcc --batch compile sweep of this lib can make real, uncontrolled outbound network connections to third-party servers -- extend the existing "don't sweep into high-frequency automated re-boot loops" caveat (already standard for dsI/dsII/imud/foundation2-class libs) to cover automated/batch COMPILE sweeps of this lib too, not just driver boots. doc/RELEASE_NOTES_HTTP and tmp/imc2.log were reset to their originally-shipped/pristine state before every commit in this session; a future re-test will regenerate fresh copies again on its own next boot/sweep, which is expected and not a bug.

Tested with a raw Python socket client (scripts/mudclient.py), all required flows in one continuous session:

1. First-boot admin installation wizard (secure/lib/connect.lpc): username fluffos -> password/confirm Mud@2026 -> default display name -> gender male -> blank real name -> [email protected] -- completed successfully; the mud self-shut-down as designed (connect.real's shutdown() + destruct(master()), matching dsII/ds386's installer behavior -- a driver restart is required before the new admin can log in). secure/cfg/groups.cfg and secure/include/config.h correctly ended up with fluffos in place of the ADMIN/DEBUG_NAME placeholder tokens -- unlike dsII, this archive's shipped groups.cfg/config.h already had the correct placeholder tokens (not a hardcoded original-author name), so no hand-fix was needed here. 2. Reboot + admin login: fluffos/Mud@2026 -> arrived in /realms/fluffos/workroom with wizard robe+hat, general/creator/ admin news tiers all paged correctly, who-equivalent (people, auto-sourced via .profile) showed the admin present, score produced a correct full character sheet ("You are First Admin Fluffos (neutral)... a level 1 Human Explorer..."), update (no argument, current-room form) reported Ok (warnings-only compile, no errors), quit disconnected cleanly with correct inventory-removal messages. 3. Ordinary new-player registration (tested with Qinbing, an ASCII name -- this archive enforces a strict A-Z a-z ' - name-charset check with an explicit rejection message, so a real Chinese name is out of scope here by design, same adapted bar as ds386's own testing): name -> confirm (y) -> age-gate (y, "Are you 13 years of age or older?") -> password -> confirm -> gender -> [display-name step skipped, CUSTOM_FORMAT is off by default] -> email (validated, a blank/malformed first attempt is correctly rejected and re-prompted) -> [real-name step skipped, REAL_NAME is off] -> race selection (list/help RACE/pick RACE -- 8 races: dwarf, elf, half-elf, half-orc, halfling, human, orc, poleepkwa) -> news pages -> landed in "The start room" (the default domain's entry point) with a correct ASCII-art room map and an HP/MP/SP status bar. Re-logged in as the same character on a second connection: Reconnected. (confirms the reconnect path is a distinct, working code path from fresh registration, not just registration succeeding once), score produced correct output, quit clean. 4. update <path> on the 4 lpcc-flagged "eval-cost" failures (section 6 below) confirmed 3 of the 4 are batch-sweep artifacts, not real bugs -- see that section for the one (domains/cave/room/ common1) that is a genuine, but pre-existing and non-blocking, content-side performance bug.

2282 of 2291 .lpc files pass a full lpcc_check.sh batch compile sweep (99.6%); the 9 that don't are broken down in section 6.

6. Known pre-existing gaps and sweep artifacts (not fixed -- content/archive design or tooling artifacts, not driver-compat)

• /obj/stargate: #include this is here on purpose to prevent this from loading -- a deliberate author anti-load trap, same class as ds386's identical file.
• /obj/area_room and /open/prog: builder-tool scaffolding templates never meant to compile standalone in place (area_room.lpc is copied elsewhere by the room-building tool before its relative #include resolves; open/prog.lpc is written by daemon/defines.lpc at runtime with a $S placeholder later substituted) -- same non-bug class as ds386's identical files.
• /domains/campus/chamber/elevator and /realms/template/area/ room/sample_room: fail with ONLY a "Redeclaration of global variable" warning and no further error text -- confirmed harmless by an identically-warned sibling passing clean in the same sweep (domains/campus/chamber/tree.lpc for the first, any of dozens of LIB_ROOM-inheriting rooms for the second) -- the same isolated-single-file-compile artifact class ds386's NOTES.md documents (AGENTS.md section 10.4).
• /domains/town/room/shop2 and /realms/template/area/room/ sample_room (again): fail with "Eval interrupted... cost limit reached" in the BATCH sweep, but PASS when compiled standalone via a single-file lpcc --batch run outside the shared-budget batch session -- confirmed a feedback_lpcc_sweep_eval_cost_artifact-class false positive (the whole lpcc_check.sh sweep shares one continuous eval-cost counter across thousands of files; a room whose create() clones a large SetInventory() list can tip over the limit purely because of what ran immediately before it in the batch, not because the file itself is slow). Also verified shop2 live via admin update -- no error.
• /domains/cave/room/common1: fails BOTH in the batch sweep AND standalone -- a genuine, pre-existing performance bug, but confirmed present byte-identically in daemon/races.c in ds386's own raw archive too (not something this port introduced). Root cause: common1.lpc's create() -> SetInventory() clones an orc_female1 NPC -> SetRace("orc") -> NewBody() -> daemon/races.lpc:534's SetCharacterLimbs(), which hits "Too long evaluation" (5,000,000 usec) even though the actual per-race data is tiny (13 HAND/LIMB lines total for "orc"). AddRace()'s LIMB-parsing branch (daemon/races.lpc:365-366) has an actual duplicate-append typo -- res["Limbs"] = ({ res["Limbs"]..., limb }); immediately followed by res["Limbs"] += ({limb});, appending each limb twice -- present identically in ds386's raw copy, i.e. a real original-archive bug, not conversion fallout. Whether that duplicate- append is the actual root cause of the 5-second hang (vs. some other expensive step in the NPC/race/body creation chain) was not fully isolated; per this project's "programming bugs only, and even then verify before applying" discipline, and because domains/cave/ is NOT preloaded and NOT on the core registration/login/play path this pass verified, this was left as a documented, non-blocking gap rather than speculatively patched. A future deep-play pass into the cave domain specifically should budget time to actually isolate and fix this (or confirm it's cosmetic/rare enough to leave).
• /domains/Ylsrim/broken/jar: LIB_CAPTURE has no implementation anywhere in the archive -- same pre-existing gap as ds386 (the file is literally shelved in a broken/ subdirectory in the original archive).
• /secure/cmds/admins/opcprof: calls the opcprof() efun, which doesn't exist in this driver build at all -- same as ds386, a creator-only diagnostic tool, not on any core path.

7. WASM pass

Measured 2026-09-03 against the shared ~/src/fluffos/build-wasm. secure/sefun/sockets.lpc was already stubbed (the ds386 sibling treatment this section originally predicted). No further mudlib-side compile fix was needed. Verified with scripts/wasm_client.js: fluffos / Mud@2026 into Fluffos' workroom, score showed "First Admin Fluffos", clean quit. IMC2/I3/HTTP version-check still try outbound connects at boot — do not loop-reboot. Shop/combat/death were not exercised this pass.

深度功能测试 / Deep functional test (round two, AGENTS.md §10.7)

First real *playthrough* pass on this lib (prior sessions only verified registration + look/score/quit + the admin install wizard, per section 5 above). Played a full, continuous session as an ordinary new player on the native driver (build-debug), using a second admin connection only for a controlled death test at the end. This is also the first lib in the Dead Souls lineage (dsI/dsII/ds386/ dshakkard/deadsouls_fluffos) to get a §10.7 pass -- none of the siblings have a "深度功能测试" heading in their own NOTES.md yet, so the bug found below (and its fix pattern) should be checked against all of them, not just this archive.

Newbie path: read chapter 1 in handbook (repeated through "chapter N" for the whole book) is this lib's own onboarding doc. Chapter 4 spells out the intended early game explicitly: level up on weak monsters in "the newbie mansion" (reached from the village, past a newbie-only gate guard and a locked mansion door -- the real way in for a level-1 character is external: grab the ladder from the gardener's shack west of the mansion, drop it under the mansion's open second-floor window, and climb ladder/enter window), sell loot to a shopkeeper (this archive's actual town vendors are James at the Healers' Guild and Lars at the pub -- the handbook's own "Otik" example is stock Dead Souls boilerplate text, not hand-tailored to this archive's actual NPC names, a content detail, not a bug), bank with Zoe, and eventually join a class (fighter/mage/cleric/thief) via a guild NPC once a gating quest (Orcslayer) is done.

Test character: id Qintestds, ASCII name (per this archive's own strict A-Z a-z ' - name-charset validation, same as ds386's and this archive's own admin-install testing in section 5 -- a Chinese name is out of scope here by design), password Abc12345, human Explorer -- kept (not cleaned up) as a representative playthrough character, currently a ghost-then-regenerated corpse sitting in "The start room" at half HP/MP/SP after the controlled death test below. Save file: work/secure/save/players/q/qintestds.o.

Bug found and fixed

Silver/gold/every other in-game currency silently turns into a float the first time it passes through ANY currency-exchange calculation, corrupting the player's Currency mapping permanently -- secure/sefun/economy.lpc's query_base_value()/query_value()/ query_base_rate()/query_player_money(), plus one sibling miss in lib/teller.lpc's eventExchange() and one in lib/props/value.lpc's SetBaseCost().

• Symptom: reproduced live buying a 2-silver bottle of water from Lars (the pub keeper) with buy water from lars. The very next attempt to learn a spell from Herkimer (ask herkimer to teach light) printed: "Light costs 100 silver and you only have 88.000000." -- a raw float leaking into player-facing text instead of a clean integer silver count. Confirmed at the data level too: after a save, the raw player file literally contains Currency (["silver":88.000000,...]) -- not a display bug, the stored balance itself is now a float and stays that way forever (every future addition/subtraction against an already-float mapping value stays a float in this dialect, since a declared int type is compile-time only and never coerces an assignment -- see this project's own reference_lpc_int_cast_is_compile_time_only precedent and the quest_times_percent_operator corpus sweep for the identical class of bug in other libs).
• Root cause, traced end to end: lib/std/barkeep.lpc's eventSell() (invoked by every buy X from <barkeep NPC>) does x = query_value(ob->GetBaseCost(), query_base_currency(), GetLocalCurrency()); ... who->AddCurrency(GetLocalCurrency(), -x);. query_value() and its helper query_base_value() (secure/sefun/economy.lpc:59-70 before the fix) are both declared to return int, but their bodies do real floating-point exchange-rate math (amount * rate, baseval / rate, where rate always comes from ECONOMY_D->__Query(type,"rate"), a value the economy daemon stores and returns as a genuine float) and return that float without ever calling to_int() -- unlike their own two sibling functions in the exact same file, currency_mass() and currency_value(), which already correctly wrap their return in to_int(). Once a float slips into x, AddCurrency(string type, int amount) stores it straight into the Currency mapping (lib/currency.lpc's Currency[type] += amount), permanently changing that currency's runtime type for the rest of the character's life. The bank's currency-exchange command (lib/teller.lpc:227-229, ask zoe to exchange N x for y) has the identical gap on its own separate line -- i = val / currency_rate(str2); -- missing a to_int() that the very next line in the same function (charge = to_int(i / (100 / GetExchangeFee()));) already has, so it independently reintroduces the same corruption even with economy.lpc fixed. lib/props/value.lpc's SetBaseCost(string currency, int amount) two-argument form (used pervasively for pricing items, e.g. SetBaseCost("silver", 10) in dozens of files) has the same gap at its own point of origin: Cost = i * rate; with no to_int(), contaminating an item's *listed* price before it ever reaches a vendor transaction.
• Fix: added to_int() around the return value of query_base_rate(), query_player_money() (also switched its internal accumulator x from a lying int to an honest float that only gets truncated at the very end, for clarity), query_base_value(), and query_value() in secure/sefun/economy.lpc; wrapped lib/teller.lpc's i = val / currency_rate(str2); in to_int(); wrapped lib/props/value.lpc's Cost = i * rate; in to_int(). Every fix matches the to_int()-wrapping convention already established by sibling functions in the very same files -- this is closing an inconsistency the original code already knew the pattern for, not inventing a new one.
• Verified: rebuilt nothing (LPC-only change), ran the lpc-syntax formatter on all three touched files (all three §9 blind-spot checks came back clean), rebooted the native driver, reconnected as Qintestds. The character's pre-existing silver balance is still shown as 63.000000 in the raw save file after the fix -- expected and left as-is: the fix stops NEW corruption, it cannot retroactively un-float a value already written to a save file (that would need a one-time data migration, out of scope for a programming-bug fix). Proved the fix actually works by exercising a brand-new currency the character had never held: opened a bank account with Zoe and did ask zoe to exchange 20 silver for gold (the exact previously-broken eventExchange() path) -- after a save, the raw player file shows "gold":2 with no decimal point, next to the still-float legacy "silver":63.000000 in the very same mapping literal, proving the fix closes the corruption at its source. log/debug.log stayed clean (no compile/runtime errors) through the whole sequence.
• Sibling libs to check: ds386, dsI, dsII, dshakkard, and deadsouls_fluffos share this same Dead Souls Object Library codebase and, per this archive's own section 1 lineage analysis, are very likely to have byte-identical or near-identical copies of secure/sefun/economy.lpc, lib/teller.lpc, and lib/props/value.lpc -- worth a quick diff check and the same three-line fix on all of them the next time any gets a §10.7 pass or economy-focused re-test.

What was tested and confirmed working

• Registration: full flow (name -> confirm -> age gate -> password + confirm -> gender -> email, validated -- a blank/malformed first attempt correctly re-prompts -- -> race pick, list/pick human) -- landed in "The start room" with the starting Player's Handbook, t-shirt, and jeans, score and i both correct immediately after.
• Movement/exploration: the default domain's start room only has a down exit into the campus domain (LPC University, a separate builder-tutorial zone) -- this is SetNoModify(1)-locked, deliberate content structure, not a bug. Walked from there through campus, across "University Square", north along a connecting path into the actual town domain (village intersection, Saquivor Road, the mansion), and confirmed door/open door mechanics, a climb ladder/enter window alternate-entry puzzle (the mansion's front door is locked and needs a key; the intended level-1 route is external, via a ladder retrieved from the gardener's shack), and a newbie-only gate-guard bypass (PreExit() in domains/town/room/gate.lpc, correctly let a level-1 character "sneak past" with a flavor message).
• Combat: no player-reachable safe-sparring dummy exists in this archive (domains/default/npc/dummy.lpc sits in a creator-only arena room gated by CanReceive(); domains/campus/npc/dummy.lpc is never placed in any room at all -- both are inert example content, not a bug, just not wired up for ordinary players). Fell back to the handbook's own recommended weak target: rats in the mansion kitchen (domains/town/room/mansion_room7.lpc, SetLevel(1), SetMaxHealthPoints(10)). A full kill rat fight ran cleanly for several dozen exchanges (both sides mostly missing at level 1, unarmed -- slow but correct, matches the handbook's own explanation of skill-based to-hit), correctly updated melee attack/melee defense skill percentages in stat from use, and ignore all + moving away correctly disengaged without further loss once clear of the room.
• Skill/class acquisition, two paths: asked Herkimer (the Mages' Guild guildmaster) directly to join -- correctly refused pending the Orcslayer quest ("First you must prove yourself worthy...", herkimer.lpc's JoinGuild()). Asked him to teach light (the organic, no-guild-required spell-purchase path) -- correctly gated on affordability, which is exactly the interaction that surfaced the currency bug above.
• Shop/economy: bought a claritin pill (10 silver) from James at the Healers' Guild and a bottle of water (2 silver) from Lars at the pub -- both transactions correctly moved the item and deducted silver, i/inventory updated immediately. Opened a bank account with Zoe (ask zoe for account, 5-silver minimum deposit) and successfully exchanged currency (ask zoe to exchange 20 silver for gold) -- see the bug section above for what this surfaced and how it was verified fixed.
• Death/respawn: no reachable-in-budget wild monster could kill a 420-HP level-1 character in reasonable real time (the rat fight above shows just how low the to-hit rates are at level 1), so this was triggered directly and deliberately via the seeded admin account (fluffos/Mud@2026) on a second connection: eval object ob = find_player("qintestds"); return ob->eventDie("test smash"); against the non-admin test character (never the admin's own body, which would hit lib/creator.lpc's wizard-exempt eventDie() override instead and prove nothing). Confirmed the full cycle live: death message + ASCII "YOU ARE DEAD!" art, score correctly showed Qintestds the ghost, undead race, and reduced max MP while dead; regenerate correctly rematerialized the character at roughly half HP/MP/SP, applied the expected XP penalty (50 -> 38, score's "more experience points to advance" figure moved from 450 to 462 accordingly), and dropped the character back at the same default "start room" respawn point. log/debug.log stayed clean through the whole death/regenerate sequence.
• quit / debug.log / reconnect after a real gap: quit produced the correct "Please come back another time!" message and clean worn-item-removal text; log/debug.log showed zero errors immediately after (this project's own §7.16/§10.7 lesson: a clean-looking quit message does not by itself prove nothing broke server-side -- checked anyway, and it really was clean here). Waited a genuine 90-second wall-clock gap, reconnected with the same login/password, and landed back exactly where the character had quit (First Village Bank, same worn items re-equipped), with score correctly reflecting realistic additional food/drink decay over the elapsed real time.

Minor observation, not fixed (cosmetic, not a programming bug)

domains/town/room/mansion_uhall1.lpc (and likely other rooms using the same window-as-exit idiom) defines "window" as BOTH a raw SetExits() key and a SetEnters() entry. The room's own GenerateObviousExits()-generated text lists "window" as if it were a plain bare-word direction right alongside down/west/east/south ("Obvious exits: down, west, east, window, south, enter window"), but typing bare window actually fails ("There is no go window here.") -- only enter window (or, per verbs/rooms/go.lpc's own help text, presumably go window) actually works. Root cause: lib/nmsh.lpc's bare-word-to-"go "-command pre-aliasing only covers the fixed classic-cardinal-direction set (n/s/e/w/ne/nw/se/sw/u/d/out), never arbitrary custom exit names, while lib/std/room.lpc's GenerateObviousExits() blindly implode()s every raw exit key into the display text regardless of whether it's bare-word-usable. This looks like it would affect every room in the whole Dead Souls lineage that names a custom exit outside the cardinal set (not something introduced by this archive), and it's misleading-but-harmless UI text, not a crash or wrong efun call -- documented here per the project's own scope discipline (ERR_THERE_IS_NO is a normal, non-fatal parser rejection, not a bug signature) rather than "fixed" by guessing at what the room author actually intended.

Sibling sweep of ds386's round-two eventRevive()/room.lpc/combat.lpc/beggar.lpc bugs (2026-08-27)

ds386's own round-two deep-functional-test pass (this NOTES.md's own sibling file) found and fixed four bugs and flagged all of them for a sweep across the whole Dead Souls 3.x lineage: (1) SEVERE eventRevive() feeding float arithmetic into AddHealthPoints()/ AddMagicPoints()'s int parameters (AGENTS.md §7.121 class), (2) the MudOS-era replace_program() fold in lib/std/room.lpc's create() (AGENTS.md §7.141), (3) the Wimpy fraction-vs-percentage bug in lib/combat.lpc (AGENTS.md §7.124 class), and (4) a minor GiveMap() reentrancy gap in domains/town/npc/beggar.lpc. Checked all four here:

• Bug #1 (eventRevive() float corruption): NOT PRESENT, confirmed clean. lib/player.lpc's eventRevive() heals back HP/MP/SP via plain integer division (AddMagicPoints(-(GetMaxMagicPoints()/2)), AddStaminaPoints(-(GetMaxStaminaPoints()/2)), AddHealthPoints(-(GetMaxHealthPoints()/2))) -- this codebase never adopted the PERCENT_MP/PERCENT_HP float-#define percentage scheme ds386/dshakkard/deadsouls_fluffos use here, so there is no float expression feeding the int parameters at all. No fix needed.
• Bug #2 (replace_program() fold in lib/std/room.lpc): PRESENT, fixed identically to ds386/dsI. Removed the replaceable(this_object()) && !GetNoReplace() → inherit_list() → replace_program(tmp[0]) fold from create() entirely (pure memory micro-optimization, no functional effect once applied since this driver defers replace_program()'s effect to a periodic backend sweep, which crashes any closure built on a freshly-loaded room, e.g. eventHearTalk()'s TALK_AREA filter, for roughly its first 5 minutes). SetNoReplace()/GetNoReplace() left in place as harmless now-unused API.
• Bug #3 (Wimpy fraction-vs-percentage): PRESENT, fixed identically to ds386. lib/combat.lpc:65 had Wimpy = 0.20; (declared private int Wimpy, so the float literal silently truncated) instead of Wimpy = 20;, plus SetWimpy/GetWimpy mistyped float instead of int. Fixed both. Verified live: booted (~/src/fluffos/build-debug/src/driver config.fluffos from libs/dsIII/), logged in as the existing admin account (fluffos/Mud@2026), ran wimpy -> Percentage: 20%, a clean integer.
• Bug #4 (GiveMap() reentrancy in domains/town/npc/beggar.lpc): PRESENT, fixed identically to ds386. Added the same !present("town map",ob) delivery-time re-check inside GiveMap() that already existed at schedule-time in init().

Verified live end-to-end: fresh boot compiled clean (zero new errors in log/debug.log, only the same pre-existing benign warning classes already catalogued), say hello world worked immediately with no crash, and wimpy showed the fixed 20%. Killed the test driver by exact PID when done.

Source pattern for all four checks: ds386/NOTES.md's "Deep functional test (round two, 2026-08-27)" section.

Sibling sweep: become <class> SetClass -> ChangeClass fix (2026-09-01)

Ported the fix from riftsds (AGENTS.md §7.195/§7.196; commit 11216f003b1) -- see ds386/NOTES.md's identical write-up (same session, same batch) for the full bug mechanism. Summary: the stock Dead Souls "Praxis" demo guild-join rooms called SetClass() directly instead of ChangeClass(), so the multi-class privilege gate (high_mortalp(), which excludes creators too) always silently rejected a fresh explorer's first class despite a success message printing.

Confirmed present here, byte-identical to ds386/riftsds's pre-fix files: all six domains/Praxis/*_join.lpc files called SetClass(); secure/include/compat.h was missing the same three query_name/query_cap_name/query_gender mappings.

Fix applied: six SetClass() -> ChangeClass() edits, three compat.h accessor lines added -- identical to ds386.

Live-verified with the seeded fluffos admin (Class reset to "explorer" via eval between attempts, to force the exact creator-excluded-from-high_mortalp() scenario the bug hits): become fighter/become cleric/become mage all now print the real success text and score confirms the genuine class change ("You are a level 1 Human Fighter"/"Cleric"/"Mage"). become monk/kataan/ rogue do not work, but this is the identical pre-existing, shared-lineage missing-class-data gap documented in ds386/NOTES.md (secure/cfg/classes/ here also only ships explorer/fighter/ mage/cleric/thief -- confirmed, same as riftsds and ds386), not a regression or a miss in this fix.

Committed only the six join-file fixes and the three compat.h mappings; incidental save-file churn from this session's boot (seeded fluffos admin's Class field, player list, mudinfo, snoop, preload saves, RELEASE_NOTES_HTTP refetch) was left uncommitted. Killed the test driver by exact PID when done.

Shop (2026-09-04 librarian slice)

Otik's general store (/domains/town/room/shop) is live on this 3.0 snapshot — the handbook's "Otik" example is not boilerplate here. Seeded admin fluffos / Mud@2026 (88 silver on the creator save) goto'd the store. buy wooden torch from otik completed a paid sale: Otik said "Here is an old wooden torch for 60 silver!", inventory showed the torch, money 88→28 silver. Purse math is integer. Do not use bare buy torch — /domains/town/obj/rayovac.lpc still carries "torch" in its id list. 拜师 analogue ask herkimer to join at /domains/town/room/magic_guild is still Orcslayer-gated ("First you must prove yourself worthy..."), same as the earlier §10.7 pass; re-confirmed live this slice.

Bug found and fixed (heartbeat hobbled(this_player())): same shape as deadsouls_fluffos/brassring/dsII. lib/body.lpc heart_beat called hobbled(this_player()); during heartbeat this_player() is 0, so disable.lpc did 0->GetMissingLimbs() every 2s. Changed to hobbled(this_object()) and added if(!objectp(ob)) return "No missing limbs."; after the decls in secure/sefun/disable.lpc. After this boot, look/goto/buy worked with no idle collapse. Verb-file hobbled(this_player()) checks in go/jump/enter/climb were left alone (those run in command context). Same one-line fix applied this session to dshakkard/riftsds/ds386 (dsI has no hobbled()).

深度功能测试（§10.7，2026-10-03）— compile warnings

Method: scripts/lpc_warnings.py dsIII --fix (KB 04 §6.10) over all 2298 .lpc files, then a pristine-tree (scripts/pristine_tree.sh) live boot: admin login (fluffos/Mud@2026), movement between the workroom and the sample room, NPC load, and a direct eval MASTER_D->log_error(...) probe. Clean boot, no crash, no "insufficient permission" apply-rejection anywhere in the console.

Fixed, mechanically, across ~1390 files: every nosave-function (the historic static→nosave conversion hitting functions, not just variables)→protected/dropped-if-called-cross-object, every unused local variable, every unknown escape sequence. See the commit log for the itemized breakdown; this is this lib's share of the corpus-wide warning-is-a-bug policy (AGENTS.md Conventions, 2026-10-03).

Fixed by hand, five distinct arg-count/return-type families (KB 04 §6.10's varargs-on-override pattern; none change call-site behavior since LPC already tolerates a call with a different arg count than declared):

• domains/default/npc/horse.lpc: dropped two redundant direct inherits (LIB_MOUNT/LIB_DOMESTICATE) already reached through LIB_SENTIENT→LIB_NPC/LIB_LIVING — the safe, same-file diamond case.
• lib/std/germ.lpc's eventSuffer() family: base changed void→ int (its only caller discards the return value); six disease overrides (ecoli/lice/cold/h1n1/flu/fleas) got varargs; rage.lpc's bare return; became return 0;.
• eventTurnOn()/eventTurnOff() family: varargs on five files' overrides (jennybot, turret, rocketpack, wristcomp, drone2) plus generator.lpc's eventTurnOn()-only override.
• create() family (six files): checked each override against the *actual* base it inherits before fixing, since unlike the two families above these don't share one common base. Five get varargs; secure/lib/net/remote.lpc is the one case where the *base* has more required args than its two overriders (realms/{fluffos,template}/adm/remote.lpc), so the fix lands on the base instead.

Two real bugs found and fixed:

• secure/daemon/imc2.lpc's GetChanInfo() built a copy of chaninfo and then returned the literal 1 instead of the copy. Grepped every file for a caller: none exist today, so this is currently unreached, but it's a real bug fixed at the root.
• scripts/wasm_keep_dirs.txt had zero entries for dsIII. Since git doesn't track log/ and the WASM site's zip-boot only pre-creates directories listed there, every live site boot has been silently failing every log_error() write ("Wrong permissions ... No such file or directory") and dumping the raw warning/backtrace text onto the player's screen instead — this is what actually surfaced the "Redeclaration of global variable" spam visibly during this session's own boot test rather than it going to a bedroom-quiet-but-dead log. Ran python3 scripts/gen_keep_dirs.py --merge dsIII (targeted single-lib mode, no 15-minute corpus rescan) and live-verified the fix: created work/log/errors by hand in the running pristine-tree boot, then confirmed a repeat eval MASTER_D->log_error(...) wrote cleanly to log/errors/fluffos with no error, both on screen and by reading the file on disk.

One self-caught mistake, documented so the class is on record: secure/sefun/fuzzymatch.lpc had been "fixed" in an earlier slice by moving alen/blen/dist inside each function's #ifdef __FLUFFOS__ guard (this driver never defines __FLUFFOS__, so the guarded block, and everything only used inside it, is dead on this build — KB 04 §6.10's documented ifdef-branch caveat). That earlier pass missed that mixed foo, baz = alen+1; on the next line is *also* only used inside that same guarded block, and had deleted it outright instead of moving it in too, leaving a dangling while(baz) reference in the (currently dead) branch. Caught on a pre-commit re-read of the diff, not by the compiler (the branch doesn't compile here, so lpcc never saw it either) — restored the declaration inside the guard in both affected functions. Grepped the rest of this session's unused-locals diff for the same shape (a deleted declaration whose *use* is also inside an #ifdef a few lines below the deletion point) and found no other instance.

Explicitly deferred, not rushed — a fresh full scan after all of the above shows 2193 distinct diagnostics remain (2298 files compile, 10 pre-existing/deliberate failures unrelated to this session: obj/area_room.lpc, obj/stargate.lpc, open/prog.lpc, secure/cmds/admins/opcprof.lpc — a missing customdefs.h, a file whose own header says "here on purpose to prevent this from loading", and an opcprof efun this driver build lacks):

• 1671 "Redeclaration of global variable" warnings across 929 files — two unrelated mixins declaring the same global name; the driver forces the second to nosave to avoid a save-key collision. Cosmetic when both sides were already nosave; a real miss only when a side that wasn't nosave silently loses persistence. Top clusters are the read/touch/smell/listen/look event mixins pulled in via storage.lpc/flask.lpc, and base_armor.lpc. This needs its own dedicated pass, not a rushed one.
• ~190 "inherited from both" diamond-inheritance warnings in the same storage.lpc/flask.lpc/base_armor.lpc/client.lpc clusters — each is a judgment call about which definition should win, unlike the clean same-file horse.lpc case already fixed.
• ~20 remaining "Number of arguments disagrees"/return-type-mismatch instances beyond the three families already fixed (GetHelp, eventReleaseObject, SetProtection, CanRelease/pre_north, CanReceive, clean_up, eventCatch, eventReceiveObject, CanGet/GetLong, eventJump, GetExternalDesc, SetCoordinates, eventRead/eventSocketClosed, eventReceiveCommand, CanSell).
• A handful of one-off warnings (Function pointer returning string constant is NOT a function call, Non-void functions must return a value, Types in ?: do not match, a few Previous function prototype ... does not match) scattered across domains/Praxis/*, lib/props/save.lpc, daemon/classes.lpc, lib/pager.lpc, and others — each needs its own individual read, not a mechanical fix.

深度功能测试（§10.7，2026-10-03, addendum）— corrections and follow-up

Correction to the 2026-10-03 section above: its sentence "Grepped the rest of this session's unused-locals diff for the same shape … found no other instance" was wrong. The grep was by name, not by scope. A scope-aware check (a deleted declaration whose name still occurs, inside the same function body of the original file, on a line under an #if/#ifdef) found 23 more: the declaration had been deleted because the driver does not compile the branch that uses it, which silently breaks the branch for the day it is compiled. All 23 were restored *inside* their inactive branch (daemon/map.lpc, domains/Praxis/roots.lpc, lib/body.lpc, lib/cedit.lpc ×2, lib/chamber.lpc, lib/lvs/position.lpc, secure/cmds/creators/gauge.lpc, secure/daemon/autoexec.lpc, secure/daemon/imc2.lpc, secure/daemon/master.lpc, secure/lib/net/http.lpc), the same fix fuzzymatch.lpc needed, and scripts/lpc_warnings.py now refuses to delete such a local (KB 04 §6.10).

Follow-up fixes (same day): 15 (: "name" :) function pointers in domains/Praxis/* (a string constant, not a call, in this driver) → (: name :) with a forward prototype; domains/town/obj/ebutton1/ebutton2 car->SetDoor(1) → SetDoorClosed(1) (KB 06 §7.211); lib/virtual/virt_land.lpc SetCoordinates(int x, int y, int z) renamed SetVirtualCoordinates (it clobbered LIB_ROOM::SetCoordinates(mixed) in every virtual room; the only callers of the 3-int form are in the file itself); return-type/prototype families (limb.lpc GetSaveString, room.lpc RemoveSearch, eventDeteriorate in obj/light/obj/whip, oob.lpc eventRead, corpse.lpc SetClass/GetClass); 29 more varargs and 14 bare return; by the tool; secure/cmds/admins/opcprof.lpc now reports that the opcprof() efun is gone from FluffOS. Oracle against the pre-session commit (8ef71a2827e): 1385 nosave→protected flips, no other change except the intended SetCoordinates rename.

Remaining diagnostics: 1671 Redeclaration of global variable (929 files) and 418 inherited from both, from 46 source files (17 redundant leaf inherits, 8 function overlaps, 4 partial overlaps, 17 name collisions; queued), and the four files that never compiled (obj/area_room, obj/stargate, open/prog, secure/cmds/admins/opcprof before this change).

深度功能测试（§10.7，2026-10-03, structural pass）— one file reached through two inherit paths

The last open cluster of the compile-warning pass (KB 06 §7.213 has the mechanism). 1671 Redeclaration of global variable + 418 inherited from both X and Y warnings came from 46 source files; the other ~900 files repeated them. dsIII now prints 5 diagnostics, all in the three files that never compiled (obj/area_room, obj/stargate, open/prog; type.h's OBJECT redefinition is open/prog.lpc including it next to compat.h).

Fixes, by kind:

• 16 leaf files inherited something their other inherit already contained (LIB_READ after LIB_ITEM/LIB_ARMOR, LIB_SMELL after LIB_BED, LIB_SHOOT in pcannon, LIB_AMBIANCE after LIB_SHOP in healer): scripts/lpc_diamonds.py --fix-redundant --apply removed the line. healer lit lamps now count (the plain ambiance getter had replaced room.lpc's).
• Name collisions between private variables: npc.lpc lost a dead Unique (the real one is LIB_UNIQUENESS's), interactive.lpc an unused globaltmp, autosave.lpc renames LastSave→LastAutoSave (props/save.lpc has its own), pipe.lpc BurnRate→PipeBurnRate, room.lpc counter→RoomCounter and Search→DefaultSearch (inventory.lpc and events/search.lpc have theirs; the room's own AddSearch/GetSearch keep using the room's), exits.lpc Dir→RoomDir (enter.lpc has one), corpse.lpc Class→LivingClass (damage.lpc's is the weapon class; corpses do not save), donate.lpc Owner→DonateOwner (domesticate.lpc has one), bot_limb.lpc lost its copy of limb.lpc's stank, base_storage.lpc its unused my_save (worn_storage.lpc renames its own to WornSave), players.lpc/pwatch.lpc/drone.lpc/the door module declared a name twice in one file, and the nine creator modules prefix their scratch globals (globalstr → file_globalstr, ...) the way dsII's do: staff.lpc and glasses.lpc inherit all nine.
• Functions defined by two inherits, resolved in the class that inherits both and spelled the way the inherit order had already chosen: player.lpc CanGet() (living's, which living::direct_get_obj() needs; interactive's "is a living being!" never was in effect), chamber.lpc/vehicle.lpc GetDoor/SetDoor/GetEnter/ResolveObjectName (exits') and SetLanguage (read's), eventTurn() of the four activatable/turnable bots (the later inherit, LIB_ACTIVATE/LIB_TURN), GetSave() of the two poles (item::GetSave(), as in dsII), worn_storage.lpc SetSaveRecurse() (persist's, which its create() relies on).
• trainer.lpc init(): base_trainer and sentient both define it, the later inherit won, so base_trainer's init() — it forgets a student who left in the middle of a lesson — never ran and "I am already training you!" stuck for good. It now calls both. Live, before and after: leave Radagast's room mid-lesson, wait out one tick, come back → before: "I am already training you!" forever; after: "You will have to start your studies anew, Fluffos." and the next lesson starts.
• Dummy hybrids (campus/obj/trashcan, town/obj/riverwater, town/obj/seawater) inherited LIB_BASE_DUMMY and LIB_STORAGE/LIB_FLASK, i.e. every item mixin twice (100 warnings each). What LIB_BASE_DUMMY adds to an item is isDummy() and SetInvis(1); the leaves define those and drop the inherit (and with it the door/knock/scratch verbs of LIB_ENTER: "knock on river" now gets the parser's refusal, "enter river" says "You can't enter that." instead of "...!"). The eight places that asked inherits(LIB_BASE_DUMMY, ob) (verbs/items/look.lpc ×3, room.lpc, flask.lpc ×3, sefun/get_object.lpc) ask ob->isDummy(), which every dummy and hybrid answers.
• LIB_OOB (secure/lib/net/oob.lpc) inherited LIB_SOCKET and LIB_CLIENT, both LIB_DAEMONs (28 warnings, and OOB_D another 24 because it inherited LIB_DAEMON next to LIB_CLIENT). The class now inherits LIB_CLIENT only and carries the six LIB_SOCKET members it used (Descriptor, Owner, GetAddress, GetDescriptor, eventCloseSocket, eventSocketClosed, SocketWrite for socket::eventWrite); OOB_D calls client::create().

Radiance (a bug the oracle exposed, not a warning): container::GetRadiantLight() already adds GetBaseRadiance(), the object's own glow, but npc.lpc and storage.lpc still added object::/item::GetRadiantLight() on top, so SetRadiantLight(7) gave 14 for an NPC, chest, bed or chair and 7 for an item or a player. Both return the container's value now (7); worn_storage did not count its contents at all (armor's getter won) and returns it too. No shipped object sets a radiance on those classes. The container formula itself (own glow attenuated by ambient: 0 at ambient 20 for a 7) is untouched.

Verification. Golden-master oracle (scripts/ds_oracle.py, KB 08 §10.12), 266 + 18 programs (every lib/ program and 61 leaf files) probed on HEAD and on the working tree: normalised GetSaveString() unchanged everywhere, no getter changed besides random per-creation stats; function definers change only where the resolution is now explicit (list above), duplicate variables disappear, the radiance rows above, healer's lamp-in-room 30→35, and the three hybrids lose the enter/knock/scratch functions. Live on a pristine tree, HEAD and edited side by side: trashcan (look at/in, get/put, "firmly attached", knock, open), river and sea (look, fill, drink, "stays in place", jump into sea swims), coffin and hoverpod (enter/out), pole GetSave(), sign and note read, plasma cannon wear/remove, the trainer scenario above, the OOB service port (oob-test, oob-begin, oob-file replies are byte-identical to HEAD) and the OOB client (clone to a local listener: same oob-begin packet, the socket closes after oob-end), a fresh mortal (13+ prompt, race, start room, score, inventory) fighting the dodo with damage both ways, and the admin command battery with no runtime error on the console. The formatter (KB 08 §9) was not run: this tree is not formatter-clean, it rewrites nearly every line of the 58 touched files.

深度功能测试（§10.7，2026-10-03, addendum 2）— a regression of mine in Praxis stone.lpc, and a corrected claim

Found while verifying the same Praxis files in ds386 (eval new("/domains/Praxis/obj/misc/stone")->GetLong()).

• domains/Praxis/obj/misc/stone.lpc: once the (: "long_func" :) pointer became a real pointer (the 15-pointer fix earlier today), look at stone ran long_func(), which printed its text with message() and returned nothing; GetExternalDesc() then added to the 0 it got back and the player saw a run-time error trace (Left hand side of += is a number (or zero)). long_func() now returns the text (the same fix as ds386). The other 14 pointers' targets were each called once live in ds386 (exit checks, at_desk, la_jail, listen_council, the spider's death_func, ...): none raises an error.
• Correction to the structural-pass notes and to the comments in trashcan/riverwater/seawater: LIB_BASE_DUMMY added more than isDummy() and the invisibility. It also brings LIB_ENTER, LIB_KNOCK and LIB_SCRATCH, and the oracle's fn - rows for these three objects (about 30 direct_*/event*/Set* functions) are that loss, not noise. Nothing in the lib calls them on these fixtures, and the live transcripts show knock on trashcan/knock on river/open river/enter river answering "You can't ..." before and after (KB 06 §7.213 kind 5 now says so). The comments in the three files now say it too.
• One of the "5 remaining diagnostics" was not a dead file: secure/include/type.h redefines the OBJECT macro that compat.h already defined ("/lib/std/item" against the bit mask 0x10; nothing in the lib uses either), and socket_err.h/socket.h do the same with ERROR_STRINGS (32 against 33). Each #define now follows an #undef, which keeps the include-order behaviour the lib always had. dsIII is at 4 diagnostics: obj/area_room.lpc, obj/stargate.lpc and open/prog.lpc (two errors).

深度功能测试（§10.7，2026-10-04）— Praxis comma-form function pointers

The 30 comma-form pointers in the 21 Praxis files and the target fixes made in ds386 (see its NOTES section of the same date: SetSearch handlers return their text, the helm asks GetMorality(), test_invis takes the viewer, look_at_vault/look_at_manhole restored as text) carry over here: scripts/lpc_twin_port.py adopt ds386 dsIII --pre HEAD --apply (18 files are token twins: 12 byte-identical, 6 spliced into dsIII's own spacing; farm, west_road2 and yard carry dsIII's own code and were edited by hand the same way). All 21 files compile without a diagnostic, and a live spot check (search field, search mounds, look at mound/manhole/vault/treats, the planning room, a voting hall) answers as in ds386.

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。secure/daemon/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。

• 2026-10-14: FluffOS 970d9f09 warns Float value truncated to int (#1426). 13 sites in 10 files: assignments, initializers and returns now spell the truncation to_int(...) (same value); x op= <float> converted the float before the operation (zeroing x for a factor below 1) and is now x = to_int(x op ...): /lib/teller.lpc（total +=）, /secure/daemon/players.lpc（seed *=）. Changed objects loaded before and after in fresh drivers, no regressions (scripts/lpc_fix_float_trunc.py, KB 04 §6.10).

x != "a" || x != "b" and x == "a" && x == "b": comparison chains of one value that can never differ (2026-10-20, KB 06 7.282)

if (arg != "wan" || arg != "pill") return notify_fail(...) is true whatever arg is, dir == "south" && dir == "north" is never true: the guard under the test never opens, or never closes. The author meant "neither of these" (&&) or "either of these" (||). != chains joined by || became && (only where the statement under the test is a return, a refusal or a guard clause), == chains joined by && became || (scripts/lpc_fix_cmp_chain.py). What it was: prompt / editor messages were never marked MSG_NOWRAP (msg_class == "prompt" && msg_class == "editor"). Changed: /lib/interface.lpc (1).

if (x = CONST): a comparison written as an assignment (2026-10-20, KB 06 7.290)

if (arg = "button" || arg = "pill"), if (!arg || arg = "") return 0;, else if (check = 0 && tempB != ""): = was typed for ==, so the condition is the constant: the branch is always taken for a non-zero number or any string (even "") and never for 0, and the variable is overwritten; the driver does not warn. Every such = became == (scripts/lpc_fix_cond_assign.py, 10 terms). Changed: /cmds/creators/cryptogram.lpc (1), /cmds/creators/rot.lpc (1), /domains/campus/npc/jennybot.lpc (1), /domains/default/weap/grenade.lpc (1), /domains/default/weap/plasma.lpc (1), /domains/default/weap/rpg.lpc (1), /domains/default/weap/torpedo.lpc (1), /lib/lead.lpc (1), /secure/sefun/mappings.lpc (1), /verbs/builders/reload.lpc (1).

A login password over 255 bytes: crypt() raises (2026-10-20, KB 06 7.294)

The driver's crypt() raises (Error in crypt(), check your salt) for a key over 255 bytes, where the old driver hashed any length. A password of 256 bytes or more typed at the login prompt threw an error in the password check and left the connection at a bare > prompt, so the login never completed (the old-password test of password the same); found by the sweep shape of 300 x. /secure/sefun/sefun.lpc now defines a simulated efun crypt(key, salt) of the same name: only an over-long key is cut (to 255 characters, then character by character until sizeof(to_buffer(key)) <= 255: the limit is in bytes, a CJK character takes three) before it is handed to efun::crypt(); every ordinary password hashes as before and the saved hashes are untouched (scripts/lpc_add_sefun.py crypt).

The suffix strip after a directory read (2026-10-20, KB 06 7.296)

After the .c -> .lpc rename the names read with get_dir("*.lpc") were still cut by two characters ($1[0..<3]): baguadao.lpc became baguadao.l, and every path built from it failed to load (stories, battlefields, events, help indexes, player-made item data, summon lists). Now [0..<5] (scripts/lpc_fix_getdir_strip.py). Changed: /daemon/help.lpc (2).

help <topic> on a plain clone, telnet and setreboot without an argument (2026-10-21, KB 06 7.212, 7.336)

help <topic> loads /daemon/help, whose LoadIndices() adds get_dir() results with +: secure/cmds/common, verbs/spells and verbs/undead are directories the archive ships empty and a git clone does not have (git keeps no empty directory; the site recreates them from scripts/wasm_keep_dirs.txt). Booted without them, the first help <topic> printed A runtime error occurred (Bad type argument to +. Had array and int) and the half-built daemon answered every later topic with "no results". Every directory is now read through DirList() (get_dir(pattern) || ({})); help look, help telnet, help races and help commands answer with the four directories removed (PLAY_RMDIRS). The mortal telnet with no argument raised Bad argument 1 to sscanf() Expected: string Got: 0 (the bare-verb sweep: A runtime error occurred), and the usage line it meant to print came from this_object()->help(), a function no object defines: it tests !args || now and prints GetHelp(); the administrators' setreboot called the same missing help() and went on without a return, it prints its syntax and returns now. Changed: /daemon/help.lpc, /secure/cmds/players/telnet.lpc, /secure/cmds/admins/setreboot.lpc.

Creator and administrator commands with no argument, the A* padding (2026-10-21, KB 06 7.336, 7.338)

The creator / administrator verbs had never been run, only compiled. Swept as the seeded administrator (play_sweep_lib.py SLUG --wiz --account fluffos:Mud@2026 --shapes one), seven of them raised on the bare command: areaclone and cryptogram (sscanf(0, ...)), pause (lower_case(0)), html (read_file(0)), encre (player_save_file() runs before the argument is looked at and raises Bad argument 1 to save_file()), resolve and wget (the daemons RESOLV_D->eventResolve(0) and WGET_D->eventGet(0) run sscanf(0, ...) / strsrch(0, ...)). Each tests the argument first and prints its GetHelp(). The A* sefun's find_path() (the creator command astar) died with (s)printf(): Null pad string specified because it pads the map with sprintf("%N'\27's", ""), a pad string that has no display width in FluffOS; it pads with spaces and replaces them, and astar prints its demo route again. The sweep must skip removeraces (it empties the race table: every NPC created afterwards fails in its heart beat). Changed: /secure/cmds/builders/areaclone.lpc, /cmds/creators/cryptogram.lpc, /secure/cmds/admins/pause.lpc, /cmds/creators/html.lpc, /secure/cmds/admins/encre.lpc, /cmds/creators/resolve.lpc, /secure/cmds/creators/wget.lpc, /secure/sefun/astar.lpc.

add_action() registrations of functions nobody defines, second round: objects behind a macro path or that the static scan cannot settle (2026-10-21, KB 06 §7.349)

The first round's static scan could not read objects that inherit through a macro path (a #define ROOM STD_DIR "room/room" whose body starts with another macro) or an inherit it could not resolve, and skipped them as UNSURE; the second round fixed the resolver, rescanned, and asked function_exists() of those objects in a running driver. An add_action("fn", "verb") whose function is defined neither by the object, nor by its headers, nor by its parents, nor by any object derived from it (scripts/lpc_scan_callout_targets.py, descendants checked transitively, then function_exists() in a running driver, scripts/lpc_verify_handlers.py). The driver walks the actor's sentences from the newest one: when that sentence's function does not exist it raises *Function for verb 'drop' not found. and tries no other handler of the verb (an older dead sentence is passed over silently). Removed 3 registrations in 2 files (scripts/lpc_fix_dead_add_action.py), the verbs are left to their other handlers: /secure/obj/machine.lpc (check_it), /secure/obj/memo.lpc (clean_memo, erase_memo). Most frequent: check_it x1, clean_memo x1, erase_memo x1.

Combat probe with the seeded administrator (2026-10-22): clean

scripts/play_combat_probe.py dsIII --account ... --clone ...: the administrator cloned /domains/default/npc/tree, /domains/Praxis/cemetery/mon/ghosta, /domains/Praxis/obj/mon/balrog, /domains/campus/npc/charly, /domains/town/npc/thief, /domains/Ylsrim/npc/balrog into its room and attacked each for 20 s (6 fights read): no driver line and no error text on the screen. (KB 08 §10.16)

new(MISSING)->move() / carry_object(MACRO(..))->wear() chains guarded (2026-10-22, KB 06 §7.73 addendum)

A statement new(PATH)->move(...); / carry_object(ARMOR_D("x"))->wear(); whose path names no file of this archive: new() answers 0 and ->move() on it raises Bad argument 1 to EFUN call_other(). In create() the object never loads, in reset() every spawn below the first missing one is skipped on every reset. 2 statement(s) in 2 file(s) (new) are now { object new_ob = new(PATH); if (new_ob) new_ob->move(...); } (scripts/lpc_guard_carry_object.py --new: the path is resolved with the macros the file itself sees, an ambiguous macro or a virtual-object path is left). Load check of the 2 changed files, HEAD vs tree: 2 load at HEAD, 2 load now (0 newly loading, 0 regress).

Pointers (: name :) to functions the object never defines (2026-10-22, KB 06 §7.355)

A (: name :) in an item_desc / inquiry / chat_msg_combat names a function that neither the object, its headers nor its parents have a body for: it compiles, and the use raises *Undefined lfun pointer called: name (look THING / ask NPC about TOPIC fail on the spot, a combat or idle chat pointer fails inside the NPC's chat() and the rest of that heart beat is skipped). scripts/lpc_scan_callout_targets.py --kinds=lfun. 2 more function(s) stay missing (other 2): the prototype is there and the body is lost, or it was never written, and no sibling lib's copy of the file can be copied as it is (it lacks the function, needs helpers this lib does not have, or has another signature); nothing invented: /secure/lib/net/ftp.lpc:eventCmdPass, /secure/lib/net/h_ftpd.lpc:eventCmdPass.

log/ subdirectories nobody created: the logging write_file() / log_file() raises (2026-10-22, KB 06 §7.358)

The code of this lib writes logs into 3 subdirectories of log/: log/harass (lib/interface.lpc), log/imc2_server (secure/daemon/imc2server/server_log.h), log/secure/log (secure/lib/connect.lpc). Neither the archive, a git clone nor the site pack has them (no local boot wrote into them, so gen_keep_dirs.py never recorded them), and write_file() raises *Wrong permissions for opening file /log/dlog/money for append. "No such file or directory" for a missing directory: the function that logs stops on that line and what came after it (the reward, the flag, the next step) never runs. This lib's log_file() is write_file(LOG_DIR + file, text) and raises the same error for a missing directory (measured). log/harass, log/imc2_server, log/secure/log are now in scripts/wasm_keep_dirs.txt (scripts/lpc_keep_log_dirs.py; the site and the pristine tree create the empty directories from it). Proof (scripts/lpcc_eval.py, an eval object under /adm/daemons or similar writes /log/zzdirprobe/..., fresh processes): the measured table in KB 06 §7.358.

Bare wizard command with a 0 argument (2026-10-27, KB 06 §7.336 Also)

After the permission gate, the command still handed 0 to sscanf / read_file / find_living. lpc_fix_bare_arg.py adds if (!arg) (or help(me) / GetHelp()) before the first use. See this commit for the files touched in this lib.