Nightmare 4 -- porting notes

Source: git clone https://github.com/MudRen/nightmare4 (commit ad4235d477f9cf7bdc67893fd26d6e6722c39e31... actually first commit per its single-commit history, cloned 2026-08-26). The repo's own description: "An influential early lib family. Newly revamped: Unix and Windows driver included together." Top level bundles a full fluffos-2.9-ds2.14 driver source tree, bin/mudos.cfg, and win32/driver.exe alongside the actual mudlib -- all ignored per this project's convention (own driver build at ~/src/fluffos). Mudlib root is the repo's lib/ directory: 1018 raw files, 18MB. Slug nightmare4, number 932, port 40234.

0. Confirmed genuinely distinct from nightmare3

Before investing effort: nightmare3 (already in this collection, port 40208) is the fluffos/nightmare3 fork -- its own Praxis-domain snapshot. This archive ships a completely different example domain (domains/Ylsrim, a desert bazaar town, vs. nightmare3's domains/Praxis/Examples), a different file layout (lib/std/ for base object classes, secure/lib/net/ for a dedicated socket-object layer that nightmare3 doesn't have, class-based Race/Class/ Season data daemons), 568 .lpc files after conversion vs. nightmare3's 822, and its own header versioning that alternates between "Nightmare IV"/"Nightmare IVr2/r3/r6" and "Nightmare V" across files (incremental in-house revisions, not a rename) -- a genuinely different generation/snapshot of the same family, not a duplicate.

1. Conversion

Pure ASCII archive -- no encoding-normalization work needed. convert_lib.sh renamed 568 .c files to .lpc, fixed 125 literal .c" references, converted 26 local angle-bracket includes to quotes, and ran the static->nosave sweep across 498 files with no string-literal or #define nosave static-shim collisions found.

One file entirely escaped this pipeline: secure/lib/connect.real (no .c extension, so convert_lib.sh's glob never touched it). This is real, live LPC source -- the mudlib's own one-time admin-installer (secure/lib/connect.lpc) cp()s this file over itself once the admin account is created, so it becomes the actual live login/registration object for every subsequent connection. It still had 31 raw static keywords on functions (fixed with the same \bstatic\b->nosave sed) and three TYPE array NAME/(TYPE array) dialect sites (see \S3) that only surfaced once compiled standalone -- both bugs would otherwise have surfaced silently, well after this port's own testing, the moment a real admin install ran on a copy that skipped this manual step. Any future archive in this collection should be checked for other non-.c-extensioned live source the mechanical pipeline would miss (grep -rL for text files containing inherit/#include outside *.lpc/*.h) -- secure/sefun/to_object.orig and a couple of Emacs .h~ backups were also found this way but confirmed genuinely dead (never referenced by any cp()/#include/load_object()), left alone.

2. Missing get_root_uid()/get_bb_uid()/creator_file() --

blocked literally every single object load

The single most severe bug in this port, exactly AGENTS.md \S7.2's catalogued class but with an extra wrinkle. This driver was built with PACKAGE_UIDS, which makes set_master() call master::get_root_uid() and (on first load) master::get_bb_uid() -- if either is undefined, the driver exit(-1)s the whole process immediately, with only No function get_root_uid() in master object; possibly the mudlib doesn't want PACKAGE_UIDS to be defined. printed. master.lpc had neither. Added both as flat stubs ("Root"/"Backbone"), matching the identical fix already present in this collection's nightmare3 sibling.

That alone was NOT enough: with both defined, give_uid_to_object() (called on every single object load, not just at master-boot) then requires master::creator_file() too -- entirely missing here. Every load silently failed via a catch()-swallowed error(), so the very first individual-file compile tests (/lib/std/room, /lib/std/item) both "failed to load" with zero output on stdout -- only log/catch/log/runtime (this mudlib's own separate error log, written by master.lpc's error_handler()) showed the real message: *master object: No function creator_file() defined!. Since this codebase's actual ACL model is built entirely on query_privs()/ file_privs() (check_access() in master.lpc), not on driver-level uids -- nothing else in the archive ever calls getuid()/seteuid() for a real decision (confirmed by grep) -- a flat "everyone is Root" stub is a minimal, behavior-preserving fix, not a new ownership scheme:

string get_root_uid() { return "Root"; }
string get_bb_uid() { return "Backbone"; }
string creator_file(string str) { return get_root_uid(); }

master.lpc also used ref (a reserved word on this driver, \S7.2's sibling class) as a plain mapping parameter name in load_access(string cfg, mapping ref)/ref[fl] = ... -- the exact class documented for mortremains earlier this session. Renamed to refmap (3 sites: the .h prototype, the definition, and the one use).

3. The pervasive .c->.lpc filename-slice bug -- this driver

build's whole command/verb/help dispatch architecture assumes a 2-character extension

Exactly the bug class flagged before starting this port (previously found in sunshadow's daemon/command.lpc), except here it is not one isolated instance but the dominant recurring pattern across this entire codebase's dispatch machinery -- Nightmare's command daemon, verb daemon, spell-casting daemon, help-index daemon, room/door ResolveObjectName(), update/clone/create/showtree (wizard file commands), and inventory-key normalization all independently reimplement "strip the trailing extension" using either:

• an unconditional file[0..<3] (assumes exactly a 2-character extension, i.e. .c) -- silently produces "look.l" instead of "look" once files are 4-character .lpc; or
• a guard-then-strip pair, if (file[<2..] == ".c") file = file[0..<3]; -- convert_lib.sh's own literal-.c"-reference fix (correctly) rewrites the quoted ".c" to ".lpc", but leaves the numeric slice widths untouched, so the guard's 2-character file[<2..] comparison against the now-4-character ".lpc" literal can never be true again -- the strip silently stops firing at all, leaving the full .lpc extension baked into every derived command/verb name.

Both shapes are equally fatal, just failing in opposite directions (one corrupts the name, the other leaves it un-stripped) -- either way every command lookup, verb lookup, spell-cast, and help-topic match breaks silently, with no compile error and no crash. Root-caused via a direct compile of /daemon/command (the file /lib/lib/command.c's cmdAll() calls into for every ordinary player command) once the \S2 fix let objects load at all. Swept the whole corpus for this shape (grep for \[0\.\.<3\] and [<2..] ==/!= ".lpc") and fixed every site by widening the arithmetic from a 2-character to a 4-character extension ([0..<3]->[0..<5], [<2..]->[<4..], and the accompanying strlen(x) > 2 guards to > 4):

daemon/command.lpc (command-name dispatch, unconditional -- the most severe: every ordinary command in the entire mudlib routes through this), daemon/verbs.lpc (2 sites: verb registration + reload), daemon/help.lpc (4 sites: the index-builder closure plus 3 more inline slices, and a guarded topic normalizer), lib/detect.lpc (2 sites, spell-cast verb matching), lib/enter.lpc and lib/exits.lpc (ResolveObjectName(), identical in both), lib/props/inventory.lpc (SetInventory()'s key normalizer), secure/cmds/creators/update.lpc (2 sites: the wizard update command's own extension-append AND extension-strip, the append side previously always true post- rename -- update /std/room would have looked for /std/room.lpc.lpc), secure/cmds/creators/create.lpc, cmds/creators/clone.lpc, secure/sefun/to_object.lpc, and cmds/creators/showtree.lpc. lib/history.lpc:118's superficially similar neu[<2..] == "/g" / neu[0..<3] was checked and confirmed unrelated (stripping a 2-character sed-style /g flag suffix from a user-typed regex, not a filename) -- left untouched.

4. Old-MudOS TYPE array NAME declaration dialect -- systemic, not

an isolated slip

This codebase's actual house style writes every array-typed declaration as two words (string array foo, mixed array bar...) rather than string *foo -- 376 sites across 109 files, both for function return types/parameters and local variable declarations, plus a class X array variant (structs-of-arrays) that a plain (string|int|...) array regex misses. Swept mechanically: every (string|int|object|mixed|float|mapping|buffer|function)\s+array\b -> \1 * (one false positive reverted: a string literal in lib/events/dig.lpc's own error message, "...or mixed array, Got: ..."), plus the two class NAME array sites by hand (lib/body.lpc's GetMagicProtection(), secure/daemon/finger.lpc's Marriages). The same dialect resurfaced in secure/lib/connect.real (\S1, escaped the mechanical sweep entirely) as both a declaration and two (string array) casts on RACES_D->GetRaces(1).

5. TYPE * a, b; -- the * binds to the first declarator only

The other half of AGENTS.md's general \S6.3 catalog, found repeatedly: old code intends every comma-separated name in a multi-declarator line to share the pointer/array-ness of the first, but this driver (like standard C) only applies * to the first name -- silent until the second name is later used as an array and hits a hard Bad assignment/ Invalid types to '-' error. Fixed only the instances that actually failed to compile (per this project's policy against blind bulk-fixing every superficially similar declaration): daemon/races.lpc's AddRace() (tmp), daemon/classes.lpc's AddClass() (tmp), daemon/help.lpc's GetIndices() (val, confirmed via its member_array(tmp, val) use requiring an array), lib/chapel.lpc's Religion/Deities/DeityIds (all three, confirmed via Religion[1]/ member_array(deus, DeityIds)/etc.), secure/cmds/players/history.lpc (arr, tmp), and cmds/creators/clean.lpc (items, users). Several superficially identical lines elsewhere (daemon/banish.lpc's loop-counter i, cmds/players/date.lpc, secure/daemon/finger.lpc's Religion, secure/cmds/players/peer.lpc's ret) were checked and left alone -- they compile cleanly as-is, meaning the second name is genuinely used as a scalar there.

6. Bare array declaration -- hard syntax error here, not a silent

no-op

lib/history.lpc's private nosave array History; (no element type) is AGENTS.md's general "bare array x;" class, but manifests as a hard compile error on this driver (unexpected L_IDENTIFIER) rather than the silently-inert declaration documented elsewhere in this project's catalog -- and since lib/history.lpc is inherited straight into the player body class (lib/interface/lib/creator/lib/player/ lib/interactive all failed transitively from this one line), it blocked every player body from compiling at all. Fixed to mixed *History (confirmed via its allocate()/sizeof()/string-indexing uses).

7. ed_start()/ed_cmd()/query_ed_mode() -- this build uses

__OLD_ED__, only the old ed() efun exists

Exactly the catalogued class, hit in two independent files:

• lib/editor.lpc (eventEdit()/process_input()) -- the general-purpose "edit this file, then call me back" mixin used by mail, bug reports, room-building, and the bare ed command. Rewrote against ed(file, exit_fn, restricted): eventEdit() now calls ed(file, "ed_exit", !creatorp()) and stores the caller's callback; the driver invokes the named ed_exit() function on this object when the interactive session ends (replacing the old manual query_ed_mode()==-1 polling in process_input(), which is now unreachable dead code kept only so files chaining editor::process_input() still compile -- the driver's C-level input loop intercepts every keystroke directly for the whole session under __OLD_ED__, bypassing process_input()/input_to() entirely). This file is inherited into the player body class, so before this fix no player body could compile at all -- the single most severe finding in this port after \S2.
• lib/nmsh.lpc (the Nightmare shell mixin, also inherited into the player body) -- write_prompt() called query_ed_mode() to draw a different prompt while mid-edit. Since the driver draws its own prompt for the whole duration of an old-style ed() session (same reasoning as above), this branch can never fire under either ed model; dropped rather than ported.

secure/cmds/creators/lsed.lpc (a wizard utility that drives the editor programmatically, feeding it a whole script of ed commands with no real user typing, to batch sed-style edits across many files) depends on the synchronous, in-process "new ed" API in a way that has no equivalent under __OLD_ED__ at all -- old ed() hands control to the driver's C-level input loop for a real interactive user, there is no way to feed it commands programmatically from LPC. Left as a documented, non-portable gap (see \S10) rather than reimplementing a text editor in LPC.

8. SetLong()/closures -- the same class already seen this session

in sunshadow, independently present here

obj/wed_ring.lpc, obj/portal.lpc, and three domains/Ylsrim/room/ files (armoury, sand_room, weaponry) call the single-argument SetLong() with an old-style closure (SetLong((: RealLong :))) for a dynamic long description, but every implementation in the inherit chain (lib/std/room.lpc, lib/events/look.lpc) only accepted string. Both already had the machinery to evaluate a functionp()/arrayp() value at *read* time (GetInternalDesc()/GetExternalDesc() already branch on functionp(val)/arrayp(val)) -- only the *setters* and their backing fields were narrowly typed. Widened SetLong()/ SetInternalDesc()/SetExternalDesc() to mixed, and the backing InternalDesc field (lib/events/look_in.lpc) from string to mixed (ExternalDesc in look.lpc was already mixed). door.lpc's own two-argument SetLong(side, long) already took mixed long and needed no change. Left lib/interactive.lpc's player-body override of SetLong() narrow (string, doing a strsrch(str, "$N") check) since nothing calls it with a closure and widening it would need an additional stringp() guard for no exercised benefit.

Related, smaller: lib/comp/seal.lpc's eventOpen() called eventPick(who, tool) -- the real signature is varargs mixed eventPick(object who, string id, object tool) (three params: an optional lock/key id string in the *middle*, then the tool). The 2-argument call is missing the id argument entirely, most likely an artifact of eventPick() being extended with a new middle parameter after this call site was written and never updated -- fixed to eventPick(who, 0, tool), preserving the exact same runtime behavior (the function already treats a missing/0 id as "no specific key") while satisfying the compiler's arity/type check. This blocked every lockable door/container in the inherit chain (lib/std/limb/corpse/storage, lib/door, lib/trap, several domains/Ylsrim rooms) from loading.

9. Missing directory breaks a get_dir() +-chain

daemon/help.lpc's LoadIndices() builds the "commands" help index by +-concatenating nine get_dir() calls across nine different command/ verb directories -- but secure/cmds/common/ doesn't exist in this archive snapshot (confirmed: verbs/spells, verbs/undead, cmds/undead are ALSO missing, and the original author already guarded those specific calls with || ({}), just not this one). This driver's get_dir() on a directory that doesn't exist returns something that isn't an array, so mid-chain array + non-array throws Bad type argument to +. Had array and int. at runtime, breaking create() for the whole help daemon -- caught by nothing, since create() isn't wrapped in a catch() anywhere in its call chain, so daemon/help itself never finished loading. Guarded the one unguarded call the same way the original author already guarded its siblings: (get_dir(DIR_SECURE_COMMON_CMDS + "/*.lpc") || ({})).

10. Known gaps -- documented, not invented or mechanically forced

Four files remain uncompilable after the fixes above (a 99.3% pass rate, 564/568, on the full lpcc_check.sh batch sweep); none of them are preloaded in a way that blocks boot (master.lpc's preload() wraps every preload call in its own catch()), and none are reachable from the standard registration/gameplay path exercised in \S11:

• secure/cmds/creators/lsed.lpc -- see \S7; needs the "new ed" synchronous API this driver build doesn't have. A wizard-only batch scripted-edit utility, not core gameplay.
• secure/daemon/remote.lpc (preloaded, but preload failures are caught) -- the "Remote Creator Protocol" daemon, a legacy inbound socket letting a separate client program edit files over a raw TCP connection on PORT_RCP. Every eventWrite(fd, ...) call site passes the connection's raw int file descriptor where the inherited LIB_SERVER::eventWrite(object owner, mixed val, int close) expects the owning socket *object* (owner->GetDescriptor()) -- a pre-existing arity/type mismatch against remote.lpc's own declared base class, not something this port introduced (nothing in the file defines a local, differently-typed eventWrite() override, and no git history exists to show it ever did). Whether old MudOS's looser compile-time type checking simply let this call through uncaught, or this daemon was already non-functional in the original archive, is unrecoverable from a single-commit clone; left uncompiled rather than reworking a niche wizard-only network protocol's calling convention.
• secure/daemon/bugs.lpc -- an in-mudlib bug-tracker daemon explicitly gated #ifdef __PACKAGE_DB__ with the file's own comment "only useable if you can define PACKAGE_DB". This driver build happens to unconditionally predefine __PACKAGE_DB__ (advertising SQL package support), which the original authors evidently never expected any real deployment to enable -- the gated body itself is genuinely unfinished/truncated (a bare ret[row[0]] = (... with no closing braces, an undefined ret variable, a tmp used before declaration in eventAssign(), a hardcoded nightmare.imaginary.com DB host). Not touched -- reconstructing a truncated, never-finished feature would be inventing content, not fixing a driver incompatibility.
• daemon/tmp/A_WEARON.lpc -- one of ~30 tiny runtime-generated cache files under daemon/tmp/ (each does int TYPE() { return SOME_MACRO; }, a classic old-MudOS trick for resolving a #define's value dynamically without a real efun). All the others reference a real macro from include/armour_types.h; this one references A_WEARON, which doesn't exist anywhere in that header (the real macro is A_WEAPON) -- almost certainly a typo baked into this snapshot's cache by whatever process originally generated it, not authored content. Left in place as a harmless, non-preloaded, already- broken cache artifact rather than guessing at and "fixing" a generated file's intended macro name.

11. Verification

Booted the real FluffOS driver (~/src/fluffos/build-debug/src/driver) against config.fluffos -- Initializations complete. reached, clean of any new fatal errors in log/catch/log/runtime beyond the single, already-documented secure/daemon/remote preload failure (\S10). Verified end to end with a raw Python socket client (no tmux_mud.sh/mudclient.py shortcuts):

• Admin install: secure/lib/connect.lpc (the mudlib's real, built-in one-time installer, active only until the first admin is created) walked through completely -- MUD name (fluffos), password + confirm (Mud@2026), display name, gender, real name, email -- correctly added fluffos to both the SECURE and ASSIST groups in secure/cfg/groups.cfg, created secure/save/creators/f/fluffos.o, and self-replaced secure/lib/connect.lpc with the fixed connect.real (confirmed identical via diff afterward) -- exactly the mechanism this codebase intends for admin bootstrap, not a hand-edited save file.
• Admin login (separate connection, post-install): name + password -> three news screens (general/creator/admin, confirming the SECURE group grant actually took) -> real starting room (Ylsrim's central bazaar) -> look (full room description), score ("Creators have no score, get a life." -- correct, intentional creator-exempt response, not a bug), who (shows Arch: Newbie Wanderer Fluffos the unaccomplished, confirming admin rank), say hello world (broadcasts correctly), quit (clean disconnect, [Fluffos quits] logged).
• Full new-character registration (separate connection): name (testchar) -> confirm y/n -> password + confirm -> gender -> display name -> email -> real name (optional) -> race selection (list shows 17 races, pick human) -> newbie news screen -> real starting room, landing correctly alongside the admin -- look/score (correct level-1-human character sheet: town, alignment, age, hunger/thirst/ mood) both producing correct output, quit clean. Removed the throwaway testchar save afterward; kept the seeded fluffos admin account and its realms/fluffos/workroom.lpc, matching this collection's existing convention for nightmare3.
• Driver killed by exact PID after testing (never pkill -f).

12. Genuine outbound network connection at boot

daemon/intermud.lpc (preloaded) is the Intermud-3-protocol daemon; its create() hardcodes Nameservers = ({ ({ "*i4", "204.209.44.3 8080" }) }) and, 2 seconds after boot (call_out((: Setup :), 2)), opens an outbound socket to that address to register with the Intermud nameserver network. Unlike the similar mortremains/ sunshadow-adjacent findings this session (where the hardcoded 1990s target address was confirmed unreachable), a live ss -tnp check during this port's own boot test found the connection actually reached ESTABLISHED state -- that specific address apparently still answers on port 8080 today. This is genuine, intended upstream functionality (not a bug to patch away), but per this project's established convention, nightmare4 should not be swept into high-frequency automated reboot/deep-test loops the way most of this corpus safely can be -- each boot is a real outbound network event to a real, live remote host, more so than the merely-attempted connections documented elsewhere in this collection.

13. WASM status

playable. Booted ~/src/fluffos/build-wasm/src against this lib via scripts/wasm_client.js. secure/sefun/sefun.lpc's dump_socket_status() (a cmds/creators/netstat.lpc-only diagnostic, not on the boot/login path) called socket_status() unconditionally, which is Unknown efun without the sockets package -- fatal at simul_efun compile time, identical shape to this project's ds386/dsII/lima entries. Guarded the function body with #ifdef __PACKAGE_SOCKETS__ ... #else return ""; #endif. After the fix, a full WASM session verified clean: real menu-driven registration (name, confirm, password + confirm, gender, display name, email, real name), race selection (list / pick human), landing in Ylsrim's central bazaar, look/score, and quit. (Initial runs with too-short --idle values between wasm_client.js sends appeared to hang mid-registration at the gender prompt -- reproduced identically on the *native* driver with the same fast timing, confirmed purely a test-harness pacing artifact via a slower --idle 2.0, not a WASM-specific or mudlib bug.)

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

One continuous session against ~/src/fluffos/build-debug/src/driver, a raw Python socket client (no tmux_mud.sh), plus a separate admin (fluffos/Mud@2026) connection for eval-based verification. Registered two full new characters through the real menu-driven flow (name -- rejects any digit, letters/'/- only, 2-15 chars -- confirm y/n, password x2, gender male/female spelled out in full, display name, email requiring a real user@host shape, optional real name, then list/pick RACE), landed in Ylsrim's bazaar both times. Walked bazaar -> n (Kaliid Road) -> w (Kaliid Road near the Fighters' Hall) -> enter hall -> n (through the oak door) to reach Roshd Burlyneck, used ask roshd to describe fighters then ask roshd to join fighters -- class join works exactly as documented (score correctly flips from "Newbie Wanderer ... Drifter" to "Newbie Fighter ... Fighter", skills populates the fighter's primary/secondary/other skill table from a clean zero baseline). Tested the pub economy path (enter pub -> ask lars to list -> correct menu response "I currently supply ale for 3747." -> ask lars to serve ale -> correctly rejected with "You do not have that much in electrum" for a broke new character -- no crash, GetCost() already correctly uses to_int()). Confirmed the bare command forms (list, serve ale typed directly, without ask lars to) are NOT valid verbs in this codebase at all -- SetCommandResponses() is only ever dispatched from eventAsk(), so every NPC "command" here is really an ask NPC to VERB sentence; this is the correct, intended dispatch shape (confirmed by reading lib/sentient.lpc), not a missing-verb bug.

Two real bugs found and fixed (both silent -- zero compile error, zero crash, invisible to lpcc_check.sh and to a clean boot):

1. lib/currency.lpc's GetNetWorth() -- declared to return int but its body sums amt / tmp (a float currency_rate() divisor) into a float net_worth local and returned that float bare, with no to_int() -- the exact §7.121 shape (declared int, real float value, no runtime coercion on this driver). Live caller lib/living.lpc:479's eventSteal() assigns the result straight into a declared-int amt used for random(7*amt)/random(amt)/ AddStatPoints("coordination", random(amt)) in the stealing-skill training path -- not just the cosmetic cmds/creators/stat.lpc "net worth" display. Fixed: return net_worth; -> return to_int(net_worth);. Verified live via admin eval: ob->AddCurrency("gold", 37); ob->GetNetWorth() now returns a real int (intp() true), previously would have been a bare float. (lib/mayor.lpc's near-identical int cost = (currency_rate(...) * Tax); shape at lines 27/49 was checked and left alone -- Tax defaults to 0 and nothing in this archive snapshot ever inherits lib/mayor.lpc at all, confirmed by a corpus-wide inherit grep, so it's genuinely dead/unreachable code, not a live bug.)

2. lib/combat.lpc's Wimpy field (the auto-flee-at-low-health safety threshold) -- a brand NEW bug class, written up as AGENTS.md §7.124. private int Wimpy; is initialized in create() with Wimpy = 0.20; (a bare fraction), but every real consumer in this codebase (cmds/players/wimpy.lpc's player-facing wimpy PERCENTAGE command, cmds/players/score.lpc) treats it strictly as an integer percentage 0-100 (wimpy on sets it to the literal 23). The runtime gate in eventReceiveDamage(), if( Wimpy < percent(hp, GetMaxHealthPoints()) ) return x; (skip fleeing when true), is therefore comparing 0.20 against an integer 0-100 -- true for virtually every nonzero health percentage, so the auto-flee safety net was silently dead for every character from character creation, until the player manually ran wimpy NUMBER themselves. SetWimpy()/GetWimpy() were also declared to return float (matching the buggy literal) instead of int (matching the field's own declaration and every real caller) -- a second, dependent bug: once a real nonzero integer sits in Wimpy, the declared-float return silently widens it, and callers' (int) casts (compile-time-only on this driver, see reference_lpc_int_cast_is_compile_time_only) don't convert it back, corrupting the player-facing display ("Percentage: 20.000000%" instead of "20%"). Fixed both: Wimpy = 0.20; -> Wimpy = 20;, and float SetWimpy(float wimpy)/float GetWimpy() -> int SetWimpy(int wimpy)/int GetWimpy().

Verified live in two independent ways: (a) a brand-new character (freshtwo) created strictly after the fix shows score's mood line as "You are feeling wimpy." and the bare wimpy command correctly prints "Percentage: 20%\nCommand: go out" (a pre-existing character, wanderling, created before the fix still shows "You have wimpy turned off" -- expected, the fix does not retroactively repair an already-corrupted save, same caveat as §7.121/§7.122); this held after a full quit/reconnect cycle, so the fixed default survives a real save/restore round-trip, not just the in-memory session. (b) an admin eval directly exercised the exact comparison used in eventReceiveDamage() against freshtwo's real GetMaxHealthPoints(): simulated 15%-health now correctly evaluates to "should flee", simulated 50%-health correctly evaluates to "should not flee" -- both were "should not flee" before the fix, confirming the safety net was unconditionally dead, not just mis-tuned.

Live combat against a real hostile was not reached in this session: this archive's Ylsrim example domain ships exactly one NPC weak enough for a level-1 test character to safely engage (domains/Ylsrim/npc/traveller.lpc, level 5, non-aggressive, s_bazaar's SetInventory spawns it via unique(file, 1)) -- every other domain NPC (max/shiela, level 15 shopkeepers; lars, level 12 barkeep; roshd/priest, level 45 class leaders) is far too strong for a fair fight, and domains/Ylsrim/npc/ balrog.lpc (an "example simple NPC" file) is never actually placed in any room at all (confirmed by grep -- dead/template content, not reachable). The traveller itself turned out to be blocked by its own intended rarity design, not a bug: daemon/unique.lpc's GetUniqueCopy() enforces an 86400 * rare second real-world cooldown between spawns (rare=1 here, i.e. 24 real hours), and save/unique.o already had a timestamp from this lib's own onboarding-time testing less than a day old, so unique() correctly returned 0 (confirmed live via eval: unique(fn,1) returns 0, catch() shows no error) and the NPC legitimately could not respawn yet -- this is the same "ghost absence is often a feature" shape documented elsewhere in this project, not a bug to patch around. Given the risk of permanently disrupting a shared shopkeeper NPC (or the same 24h-cooldown unique mechanism) for an incremental confirmation, combat plumbing was instead verified end-to-end via the two eval checks above (b) rather than a live fight; a full live fight (and the death/respawn cycle) is flagged here as not verified live, an honest gap rather than a silent skip, should a future session have more wall-clock room (or catch the traveller's cooldown window) to complete it.

All four known cross-cutting bug patterns checked explicitly:

• §7.121 currency float-into-int: found and fixed (GetNetWorth() above); every other money-handling function in lib/currency.lpc/lib/std/barkeep.lpc/lib/std/vendor.lpc/ lib/teller.lpc/secure/sefun/economy.lpc already correctly wraps its float arithmetic in to_int()/to_float() at every call site checked.
• §8.3a private command-dispatch demotion: not applicable to this codebase's architecture at all -- there is no command_hook mechanism; ordinary commands dispatch through daemon/command.lpc (already fixed for the §7.118 filename-slice bug at onboarding) and NPC "commands" dispatch only through eventAsk(). A script-driven check for the general shape (a private function referenced by string name from add_action()/call_out() within the same file, the demonangel-class pattern) found zero matches anywhere in the corpus.
• §7.122 class/marker-item duplication on reconnect: not applicable -- no auto_load/compute_autoload_array/ load_autoload_obj-style reload mechanism exists anywhere in this codebase (confirmed by grep); lib/player.lpc's class quest *Quests/class death *Deaths fields are tracking data, not physical reloadable items.
• §7.123 bare file-scope assignment: one heuristic grep hit corpus-wide (secure/cmds/players/tell.lpc:101's __Morse = ([... ]);), hand-verified as a false positive -- it's a local variable inside the morse() function body, not file scope, and the file compiles and works correctly.

Systematic re-check of the §7.118 filename-slice class (this lib's dominant bug at onboarding, ~15 files): every remaining [0..<N]/[<N..]/strlen(x)-N-shaped slice in the corpus was individually triaged. All were confirmed to be doing something else entirely (stripping a trailing / one character at a time regardless of length in secure/lib/file.lpc, dropping the last word of a explode(str," ") array in daemon/chat.lpc/secure/obj/post.lpc, dropping the last array element in secure/daemon/bboard.lpc, singularizing an English category name for an error message in daemon/help.lpc, trimming an appended / sentinel character in secure/cmds/creators/grant.lpc) -- no further live instances of the dominant onboarding-time bug class remained.

Test characters left as evidence: wanderling (pre-fix, demonstrates the "wimpy turned off" post-restore symptom), freshtwo (post-fix, level 1 human, joined the fighter class, correctly shows wimpy at 20%/"go out"). freshbie/freshbie2/grimtest/ grimtest2/freshtwo-style throwaway registration retries with typos were not all cleaned up; harmless partial/duplicate throwaway accounts from mid-flow registration mistakes (invalid name/email formats rejected before a character object was ever created) are not persisted at all by this mudlib's own registration flow, so no additional cleanup was needed beyond what's listed here.

Sibling note for nightmare3/residuum: check both for the same Wimpy = 0.20-style fraction-vs-percentage literal shape in their own combat/status mixins if they share this lineage's lib/combat.lpc ancestry -- residuum is already known (per §7.118's own writeup) to share the identical filename-slice dispatch bug with nightmare4, so its combat internals are a plausible direct copy too.

Driver killed by exact PID after each reboot (never pkill -f); the Intermud-3 outbound connection noted in §12 was observed reaching ESTABLISHED again on each of the three reboots this session, exactly as before -- no change in that behavior, and the session stayed well within a small number of manual reboots rather than any automated loop.

深度功能测试（§10.7，2026-10-05）— round three: 注册、选种族、进入世界

复测原因：queue 中的 history 序列（tmi2 > nightmare3 > nightmare4 > lpmud141）； 本 lib 之前只有无日期的一轮测试（第 14 节 的 round two 为 2026-08-27）。

流程（原生驱动 build-debug，ulimit -v 6291456 限制内存，PID 精确结束）： banner → 注册（新名 Zephyrq，y，密码 ≥5 字母，male，显示名默认，email， 真实姓名可留空）→ 在同一连接内 pick human（账号要到选种族完成后才保存， 中途断开会留下未保存的半成品，再次登录会被当作新注册）→ 新手新闻按回车 → score、 inventory、quit 正常。新手进入 Ylsrim 中央集市。

存档字母目录的一次性失败（未复现，未修）：首次注册时 ConfirmPassword → SetPassword → save_player 报 Could not open /secure/save/players/z/zephyrq.o.tmp for a save（log/runtime，14:50）。随后该字母目录出现，同一驱动里 Quintara（q 目录原本 不存在）完整注册、进入世界，没有再报错。master.lpc 的 player_object() 会在目录缺失时 mkdir 字母目录，所以这次更像首次启动时的一次性竞争，而不是持续存在的缺陷。log/runtime 里另有一条 n/nativetest.o.tmp 的旧记录，早于本次测试。若将来再出现，先看 player_object() 的 if(!NewPlayer) return 0; 早退分支是否被触发。

运行期日志：config.fluffos 的 debug log file 是 debug.log，但启动后并不生成 该文件；本 lib 只通过 log_file() 写 work/log/runtime、work/log/errors。判断"无错误" 时应读这些文件，而不是 debug.log（参见 KB 08 §10.9）。

编译警告：本次未扫描，待下一轮处理。（更正：meta.json 虽写 submodule-patch，但本 lib 的 work/ 是本仓库直接跟踪的文件，不是 gitlink，lpc_warnings.py nightmare4 可以直接扫，不需要 scratch 基线。）

测试账号清理：Zephyrq、Quintara 的 users 存档、postal 目录，以及由测试 重写的 save/economy.o、save/intermud.o 均已删除或还原；secure/save/players/ 下 的字母目录也已删除（它们是本次测试创建的）。

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。

深度功能测试（§10.7，2026-10-05）— 首次注册存档失败的原因找到了；编译警告清理（机械部分）

1. 首次存档失败（上文“未复现，未修”的那条）原因找到并修复（KB 06 §7.227）。 secure/daemon/master.lpc 的 compile_object() 为新玩家建目录时写成 else if (players 目录不存在) mkdir(players); else if (字母目录不存在) mkdir(字母目录);：父目录不存在时只建父目录、跳过字母目录，于是这位玩家第一次 save_player 打开 players/<字母>/<名字>.o.tmp 失败，密码没有存下；再试一次父目录已在，才建出字母目录。git 不跟踪 secure/save/players/ 下任何文件，所以全新检出、以及网页版每位访客都会遇到。第二个 else if 改为 if（dsI 早已如此）。用只含被跟踪文件的树（git archive + 本次改动）验证：注册 qanine 首次即存档到 players/q/qanine.o，重新登录、走到战士公会 ask roshd to join fighters、score 变为 Fighter、quit 都正常。同样的写法也在 dsII、foundation2、havenmud 修了。 2. 编译警告（机械部分）：scripts/lpc_warnings.py nightmare4 --fix，874 个 nosave 函数改 protected（被别处按名调用的 192 个去掉 nosave 保持 public）、删 34 行 #pragma、删 121 个未用局部变量、补 17 处 varargs、2 处无效转义；lpc_audit_removed_locals.py 复核 170 个被删声明，0 个需要处理。按名动态调用只有 secure/daemon/localpost.lpc（读 secure/cfg/aliases.cfg），其中用到的 query_groups 仍是 public；套接字回调、call_out、input_to 走驱动内部调用，可以调用 protected 函数。整树诊断 1373 → 164，剩下的是结构性警告（同名全局变量重复声明、菱形继承、覆盖函数返回类型不一致、LIB_LIMB 宏重定义），下一步处理。管理员会话（look、who、走到战士大厅、say、smell、eval、update、help）无权限警告。 3. 已有的问题（不是本次引入）：secure/daemon/remote.lpc 编译失败，它按旧版接口调用 eventWrite(fd, ...)，而 secure/lib/net/server.lpc 的 eventWrite 第一个参数是对象；secure/cfg/aliases.cfg 的 creator: 别名调用主控不存在的 query_creators。

深度功能测试（§10.7，2026-10-05，续）— 结构性警告清零，顺带修掉的真问题

启动时警告与错误全部清零；整树扫描 164 → 0 条警告，余下 4 个文件编译失败，都不是本次引入的（见第 6 条）。

1. LIB_LIMB 指向不存在的文件。 secure/include/lib.h 先 include std.h（LIB_LIMB = /lib/std/limb，存在），又把它重定义为 /lib/limb（不存在），后者生效。战斗中肢体被砍断时 lib/body.lpc 的 new(LIB_LIMB) 失败，断肢不会出现，戴在上面的护具也不会落下。删掉 lib.h 里那行；new("/lib/std/limb") 实测可用。 2. **.c→.lpc 转换把 T array a, b 改成 T * a, b**，只有第一个名字还是数组（KB 03）。secure/daemon/finger.lpc 的 Religion 从玩家存档读回却声明成字符串，Religion[1] 取到的是字符编码，finger 的信仰一栏因此不对；lib/magic.lpc 的 send_to、daemon/help.lpc 的 fun、daemon/classes.lpc 的 ots 同样补回 *（这几个运行时不受影响）。 3. 少写的 return 与写错的运算符：lib/living.lpc 的 CanCastMagic() 在没有环境时写成 "You are nowhere!";，没有返回，继续往下用空环境；lib/events/jump.lpc 的 GetJumps() 只写了 Jumps;，永远返回 0；cmds/creators/force.lpc 的 GetErorMessage() 同理；daemon/soul.lpc 的 res == who; 应为赋值，传入种族名字符串时 res 一直是空的。spells/resurrection.lpc 先算出 chance = level + random(治疗技能)，判定却仍用 level，治疗技能完全不起作用，改为用 chance。secure/sefun/sefun.lpc 的 get_stack() 把函数名栈打印了两次而没打印程序名，第一个改为 stack0。 4. 菱形继承（KB 04 §6.10，scripts/lpc_diamonds.py）：lib/comp/container.lpc 不再继承 LIB_RADIANCE，它在 npc、interactive、storage 里那份光亮变量从未被设置（只有物品调用 SetRadiantLight()），容器自身光亮项按原样取 0，玩家、NPC、储物箱的光亮计算不变（实测 NPC SetRadiantLight(3) 加身上点燃的火把 = 10）；lib/living.lpc 不再继承 LIB_SMELL（living 只与 LIB_OBJECT 一起使用，后者已带嗅觉），删去只是转调 smell:: 的 direct_smell_obj()；lib/std/corpse.lpc 多余的 inherit LIB_SMELL 删掉。这与 fluffos 组织维护的 Dead Souls 上游的做法一致。lib/npc.lpc 自己的等级变量改名 NpcLevel（lvs/abilities.lpc 另有一个 private Level），lib/nmsh.lpc 删去从未使用的 CommandFail。同名函数两边都有的，按驱动原本的选择写明（scripts/lpc_name_winner.py，look 类 4 个、CanSell、jar 的 GetShort、pole 的 GetSave）；GetSave() 在 seal 里写成 close 与 lock 两份列表相加，storage 与 jar 返回 item 的完整列表（没有代码调用组合类的 GetSave()，运行时无变化）。 5. 原型与返回类型对齐：body.h 的 SetParalyzed、command.h/nmsh.h/command.lpc 的 GetClient、post.h 的 query_reply_text、update.lpc 的 eventUpdate、sefun.h 的 time_event、move.lpc 的 eventPrint；interactive 的 SetId/SetCapName 返回值与父类一致；room 的 RemoveItem/SetItems、germ 的 eventSuffer、limb 的 GetSaveString 改成父类的返回类型；daemon/include/classes.h、races.h 类定义后多余的 ;；tell 帮助里残留的 ("help", ...) 逗号表达式。 6. /secure/daemon/remote（远程创作者协议）从预载中注释掉（与 fluffos 组织的 Dead Souls 上游、本库同源的 dsI、dsII 相同）。它按 Foundation II 的旧套接字接口写成（eventWrite(fd, ...)），本库 secure/lib/net/server.lpc 的接口是按连接对象的，所以每次启动都编译失败；协议登录后还要加载每位创作者的 ~/adm/remote 对象，本档案里一个都没有。文件保留。havenmud 同样处理。其余编译失败的文件也都是原有的：secure/daemon/bugs.lpc（连接 nightmare.imaginary.com 数据库的未完成原型）、secure/cmds/creators/lsed.lpc（需要驱动没有的 ed_start）、daemon/tmp/A_WEARON.lpc（残留）。 7. 验证：全新启动 0 警告 0 错误；管理员会话 smell、look at me、finger（信仰一栏）、help tell、eval 取 NPC 等级（12）、新建断肢对象、点燃火把后的光亮（火把 7，NPC 3+7=10）、jar 的 GetSave()、GetId()，全部正常。

• 2026-10-14: FluffOS 970d9f09 warns Float value truncated to int (#1426). 7 sites in 5 files: assignments, initializers and returns now spell the truncation to_int(...) (same value). 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" 和 x == "a" && x == "b"：同一个值的比较链永远成立/永远不成立（2026-10-20，KB 06 §7.282）

if (arg != "wan" || arg != "丸") return notify_fail(...) 对任何 arg 都为真，dir == "south" && dir == "north" 永远为假：条件下面的守卫要么永远拦、要么永远不拦。作者要的是“既不是 A 也不是 B”（&&）或“是 A 或 B”（||）。!= 链的 || 改成 &&（仅当条件下面是 return 的拒绝/守卫），== 链的 && 改成 ||（scripts/lpc_fix_cmp_chain.py）。改动：/lib/interface.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 1 处）。改动：/lib/lead.lpc（1 处）。

登录密码超过 255 字节：crypt() 抛错（2026-10-20，KB 06 §7.294）

驱动的 crypt() 对超过 255 字节的密钥直接报错（Error in crypt(), check your salt），老驱动则按任意长度散列。在登录提示符下输入 256 字节以上的密码，密码检查那一行就抛错，连线停在一个裸的 > 提示符上，登录走不完（password 指令的旧密码检查同样）；用 300 个 x 当参数的扫描发现的。/secure/sefun/sefun.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）。

help 菜单的列宽：FluffOS 的 terminal_colour() 会给空的 RESET 换上 ANSI 复位序列（2026-10-21，KB 06 §7.337）

help commands、help races 的编号菜单每格只剩一个字母（[1] a[2] a[3] b[4] b）：/daemon/terminal 的 no_colours() 把整张表（RESET 也是空串）交给 terminal_colour()，FluffOS 会把空的 RESET 换成 ESC[49;49m ESC[0;10m（15 字节）并写回映射，strip_colours() 带着转义序列返回，arrange_string() 少算了这部分宽度，把每格截短。no_colours() 现在把 RESET 映射为退格符再删掉（Dead Souls 3.x strip_colours() 的做法）；help races 按列列出 [1] artrell [2] dwarf ...。修改：/daemon/terminal.lpc。

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

巫师（管理员）把命令敲成不带参数时，驱动把 0 传给字符串参数；下面的命令第一句就把这个 0 交给 efun，运行时报 Bad argument 1 to ...，这条指令从来没有用过。用 scripts/lpcc_bare.py 不登录、在一次 lpcc 求值里以 0 调用 main() 逐个确认后，由 scripts/lpc_fix_bare_arg.py 在第一处使用参数的语句前加守卫：用 GetHelp() 的文件打印 GetHelp() 后 return 1;；没有 help() 的是 return notify_fail("…\n");。改动（3 个文件，括号里是接住 0 的 efun）：/cmds/creators/force.lpc（sscanf）、/cmds/creators/move.lpc（sscanf）、/secure/cmds/creators/banish.lpc（lower_case）。加载检查 HEAD 与工作树对照，没有退步。

注册到不存在函数的 add_action，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 1 条注册（1 个文件，scripts/lpc_fix_dead_add_action.py），动词留给别的处理函数：/lib/bboard.lpc（cmd_save）。最多的函数名：cmd_save 1 处。

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 2 个子目录写日志：log/adm（secure/cmds/creators/call.lpc）、log/harass（lib/interface.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。本库的 log_file() 是 write_file(LOG_DIR + file, text)，目录不存在就抛同样的错误（实测）。把 log/adm、log/harass 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。