Dead Souls I -- porting notes

Source: dsI.zip, a maintained, already-FluffOS-adapted archive from lpmuds.net (a now-defunct LPMud archive site). The live site is down; this exact archive was recovered from the Wayback Machine (<https://web.archive.org/web/20160306035507id_/http://lpmuds.net/files/deadsouls/dsI.zip>) and content-verified (valid zip, intact central directory, 1673 files, non-corrupted) before this session started. Slug dsI, number 178, port 40225. The zip bundles a FluffOS 2.7-ds2.018 driver source snapshot (fluffos-2.7-ds2.018/), plus bin/, win32/, extra/, v22.2b14/ build/install scaffolding -- all ignored, this project uses its own driver. The real mudlib root is the zip's lib/ directory.

1. Duplicate check -- genuinely distinct, not a duplicate

This collection already has two Dead Souls 3.8.6-era copies, ds386 and deadsouls_fluffos. Before building out a full pipeline, checked whether dsI is a third copy of the same codebase:

• Domain count: dsI/lib/domains/ has only 2 domains (default, Ylsrim); both ds386 and deadsouls_fluffos have 8 (amigara, campus, cave, default, examples, learning, Praxis, town, Ylsrim).
• Top-level directory set: dsI/lib/ is missing estates/, open/, powers/, and std/ entirely (present in both other copies) -- 967 files total vs. 4811/4813.
• secure/daemon/master.lpc byte diff: dsI's copy is 711 lines; ds386/deadsouls_fluffos's is ~1180-1200 lines. A real diff shows completely different structure from the first non-comment line on -- different include set (<config.h>/<rooms.h> vs. <runtime_config.h>/ROOMS_H), no COMPAT_MODE/DEFAULT_PARSING/ RESET_ALL feature-flag scaffolding, no LIB_DAEMON inherit, a much smaller global-variable set (5 vars vs. 15+), no MSSP support.
• Self-identification: several files in this archive carry "Dead Souls V" or "Dead Soulsr1/r2 Object Library" header comments (vs. ds386's explicit 3.8.6-era file headers), and lpmuds.net's own site description places this snapshot as closer to Nightmare IV than to the modern Dead Souls lineage.

All of this is consistent with dsI being a genuinely earlier, distinct generation of the Dead Souls codebase, not a re-packaging of the same one -- proceeded with the full pipeline. duplicate_of is null in meta.json.

2. Conversion

English-language archive (not GBK-encoded) -- scripts/convert_lib.sh against raw/dsI/lib confirmed this: 966/967 files already valid UTF-8, 0 lossy conversions, 1 binary file skipped. 583 .c files renamed to .lpc, 127 literal .c" references fixed, 26 local angle-bracket #includes converted to quotes, 511 files touched by the static->nosave sweep.

One file needed the same by-hand treatment as foundation2's login.real: secure/lib/connect.real is not .c/.lpc (skipped by the mechanical sweep) but is real, live LPC source -- the one-time first-boot admin-installation wizard (secure/lib/connect.lpc, \S5 below) rename()s it over connect.lpc itself once the first account finishes registering, and it becomes the mud's real ongoing login handler from then on. It needed both the static->nosave sweep (31 sites) and the TYPE array NAME->TYPE *NAME dialect fix (\S3 below, 1 site: string array races = (string array)RACES_D-> GetRaces(1);) applied by hand, since the mechanical sweeps both key off file extension. Caught live: a second test account's login crashed the connection outright with a bare "*No program in object '/secure/lib/connect'!" once the installer had already consumed connect.real and swapped it into place -- confirmed via log/catch/log/runtime, fixed, verified with a fresh driver boot and a full second registration.

3. Compile-time driver-compat fixes

3.1 ref is a reserved keyword on this driver

Same bug class as foundation2/nightmare3 (§7 catalog item). Only 1 site: secure/daemon/master.h's load_access() prototype used mapping ref as an (unnamed-in-effect) parameter type/name; renamed to refmap to match the actual implementation's parameter name in master.lpc (which was already correctly named resource, not ref -- only the header prototype had the reserved-word collision).

3.2 LIB_UNDEAD gated behind a garbled, always-false #ifdef

The single highest-impact fix in this port. secure/include/ lib.h had:

#ifdef Dead SoulsLPMud
#define LIB_UNDEAD         DIR_LIB        "/undead"
#endif

Dead SoulsLPMud is not a real, definable macro name (and Dead is never #defined anywhere in this codebase) -- the #ifdef always evaluated false, silently leaving LIB_UNDEAD undefined. Every other LIB_* macro in this header is defined unconditionally; there's no evidence this one was ever meant to be conditional. inherit LIB_UNDEAD; in lib/body.lpc then failed with a hard "syntax error, unexpected L_IDENTIFIER, expecting L_STRING or '('" (the driver left the bare, unexpanded identifier where a string was expected) -- and since lib/body.lpc is the base of this codebase's entire living-creature class hierarchy (body -> race -> living -> npc/sentient, plus combat, leader, mayor, teller, trainer, std/barkeep, std/vendor, obj/friend, and every NPC in domains/Ylsrim/), this one bug transitively broke ~25 files across the codebase. Fixed by defining LIB_UNDEAD unconditionally, matching every sibling macro's convention.

3.3 Missing LIB_* macros (3 more, same root cause as 3.2)

Three more LIB_* macros referenced by inherit/new() calls were simply never defined in secure/include/lib.h at all (their target .lpc files do exist -- this is a header-omission bug, not a missing- content one):

• LIB_GERM -> DIR_STD "/germ" (target: lib/std/germ.lpc; used by domains/Ylsrim/broken/{cold,flu}.lpc, both disease-item examples). First guessed DIR_LIB "/germ" (matching most sibling macros' pattern), which compiled but failed at *load* time with "Inherited file '/lib/germ' does not exist!" (caught via log/runtime, not a compile error) -- the actual file lives under lib/std/, not lib/ directly, so the value needed DIR_STD instead.
• LIB_VIRT_LAND -> DIR_LIB "/virtual/virt_land" (target: lib/ virtual/virt_land.lpc; used by lib/virtual/virt_sky.lpc).
• LIB_FTP_CLIENT -> DIR_SECURE_LIB "/net/ftp_client" (target: secure/lib/net/ftp_client.lpc; used by secure/lib/net/ftp.lpc).

3.4 Old-MudOS TYPE array NAME declaration dialect (370+ sites)

FluffOS's grammar does natively support TYPE array NAME as a type-modifier synonym for TYPE *NAME (opt_atomic_type L_ARRAY in grammar.y, confirmed by reading the compiler source before assuming this was a bug) -- but not in every grammatical position. A function *return-type* declaration using it hits a hard parser error:

string array explode_list(string list) { ... }
// error: syntax error, unexpected L_IDENTIFIER, expecting L_ASSIGN or ';' or '(' or ','

Rather than characterize exactly which of the many possible contexts (top-level function return type, local-variable declaration, function parameter, multi-name declaration list, cast expression, class NAME array variant) are safe vs. broken, every instance was converted uniformly to the TYPE *NAME form via a small Python script (handles single names, names with = initializer default values including nested-paren/comma expressions like allocate(3) or ({ 0, 0 }), and multi-name comma-separated declaration lists like private string array Classes, Religion, Deities, DeityIds; -> private string *Classes, *Religion, *Deities, *DeityIds;, fanning * out to every name in the list). Also handled separately: (TYPE array) cast expressions -> (TYPE *) (12 sites, a different regex since there's no trailing identifier), class TYPENAME array NAME (2 sites, e.g. class MagicProtection array GetMagicProtection() -- the base-type detector only matched bare keywords like string/int, not class <Identifier>), and one bare array used alone with no type prefix at all (private nosave array History; in lib/history.lpc -> private nosave mixed *History;).

370 sites across 107 files via the main sweep, 12 more via the cast-form sweep, 2 more via the class TYPENAME array fix, 1 more for the bare-array case, plus 1 more found only later in secure/lib/connect.real once it became live (\S2 above) -- 386 total. Every file recompiled clean afterward; spot-checked a sample against the pre-fix/post-fix diff by hand (lib/chapel.lpc, lib/ magic.lpc, lib/race.lpc, verbs/items/wield.lpc, daemon/help.lpc) to confirm the comma-list fan-out and default-value paren-tracking worked correctly before trusting the full sweep.

3.5 .c-to-.lpc rename side effect: fixed-width filename-slice

arithmetic (16 sites, including the command-dispatch table itself)

Same bug class as discworld/foundation2/skylib/lpuni -- code written to strip/check a 2-character .c extension via file[0..<3] (drop last 2 chars) or compare file[<2..] == ".c" (last 2 chars) silently breaks once every file is 2 characters longer (.lpc). Checked every slice-comparison-against-a-literal-suffix pattern in the tree, not just the common one, and found instances in both directions (stripping *and* appending):

• daemon/command.lpc's eventRehash() -- the command-dispatch table builder itself: cmd = file[0..<3] (built from get_dir(path + "/*.lpc") results) turned "look.lpc" into "look.l" instead of "look", meaning every ordinary player/creator/admin command would have silently stopped matching typed input had this shipped unfixed -- caught by lpcc's own strict compile of files depending on this return value being used consistently elsewhere, and by reasoning through the bug class from the foundation2/discworld precedent before it could bite at runtime. Fixed to file[0..<5].
• Extension-stripping via map(): daemon/help.lpc (4 sites: LoadIndices()'s anonymous stripper function plus 2 inline map()s, GetHelpByIndex()'s topic-normalization check), daemon/verbs.lpc (2 sites: eventReloadVerbs()'s string-arg branch and its verb-name extraction from a bare filename), lib/detect.lpc (2 sites, both in the "detect magic" skill's spell-name guessing, see \S4 below for a related runtime crash fixed in the same code), lib/enter.lpc/lib/exits.lpc (ResolveObjectName(), identical helper duplicated in both files), lib/props/inventory.lpc (SetInventory()), secure/cmds/ creators/update.lpc (cmd()'s own .lpc-suffix strip), cmds/ creators/showtree.lpc.
• **Extension-*appending* (the inverse bug)**: x[<2..] != ".lpc" (last-2-chars compared against a 4-char literal, always true) meant .lpc got appended even when a user already typed it, producing foo.lpc.lpc -- secure/cmds/creators/update.lpc (a second, different site from the one above), secure/cmds/creators/ create.lpc, secure/sefun/to_object.lpc, cmds/creators/ clone.lpc.

All fixed to the 4-character-aware form ([0..<5] / [<4..]). Verified live: update <path> (both with and without an explicit .lpc suffix) correctly reports Ok/File not found rather than silently matching nothing.

3.6 runtime_config.h: bundled copy is stale but harmless here

The bundled secure/include/runtime_config.h has drifted from this driver's actual get_config() index numbering in several slots (confirmed by diffing against ~/src/fluffos/src/include/ runtime_config.h: index 1 is __ADDR_SERVER_IP__ here vs. a retired slot on the real driver, index 9 is __MAX_LOCAL_VARIABLES__ here vs. retired, etc.) -- the skylib-class bug from the catalog. Checked and confirmed harmless for this archive specifically: the only get_config() calls anywhere in the tree (secure/lib/net/{ftp, server}.lpc, secure/lib/file.lpc) use exactly __MAX_BYTE_TRANSFER__ (index 16) and __MAX_READ_FILE_SIZE__ (index 17), and those two specific macro names/values are identical between the bundled header and the real driver's -- the drifted slots are simply never read. Not touched, to avoid an unnecessary risk of introducing a *new* mismatch.

3.7 OLD_ED: driver build vs. archive's new-style ed API

This project's driver build defines OLD_ED (#define OLD_ED in src/local_options), which compiles out ed_start()/ed_cmd()/ query_ed_mode() (core.spec's #else branch) and leaves only the older single-efun ed(fname, exitfn, restricted) form -- but this archive's lib/editor.lpc and secure/cmds/creators/lsed.lpc were written against the newer API (same gap as foundation2/ finalrealms, opposite direction from what the AGENTS.md catalog item literally describes, which assumes the archive itself predates the new API -- here the archive is *newer* than this driver's build config).

• query_ed_mode(): added a secure/sefun/ed_compat.lpc simul_efun shim (byte-identical fix to foundation2's), approximating it via in_edit(this_object()) (available under OLD_ED): returns 0 while editing, -1 otherwise -- lossless for every real caller in this codebase, which only ever branches on -1 vs. not.
• lib/editor.lpc: rewrote to the same pattern as foundation2's fixed version -- eventEdit() now calls the old ed(file, "__ed_exit_dispatch", !creatorp()) directly (the driver's own comm layer routes raw input straight to its C-level editor while a session is active, so this object no longer drives the edit loop itself), with a new __ed_exit_dispatch() callback replacing the old process_input()-based loop. One convenience lost: the original used ed_cmd("$a") right after starting to drop mortals/empty-file edits straight into insert mode; there's no LPC-level way to feed a just-started real ed() session a synthetic command, so editing now always starts at ed's normal command prompt.
• secure/cmds/creators/lsed.lpc (a scripted, non-interactive batch editor -- "run this list of ed commands against a file"): has no OLD_ED equivalent at all. Unlike a single interactive ed() session (fire-and-forget with an exit callback) or query_ed_mode() (approximable via in_edit()), there is no LPC-level way to feed a running OLD_ED session commands one at a time and read back its output, which is this tool's entire design. Left as a graceful "not supported on this driver build" message rather than left uncompilable -- a creator-only scripting convenience, not part of this project's core player/creator verification path.

4. Runtime (non-compile-time) bugs found and fixed

4.1 daemon/help.lpc's LoadIndices() crashed on every single boot

Not caught by lpcc_check.sh (the file compiles and loads fine standalone) -- only surfaced because lpcc --batch also *executes* each object's create(), and because this file is transitively reachable during a real boot (it's not itself preloaded, but gets loaded on first help use, and this project's batch check loads every file regardless). LoadIndices()'s "commands" index builder chains several get_dir(DIR_X + "/*.lpc") calls with +:

tmp = get_dir(...) + get_dir(...) + ... + get_dir(DIR_SPELL_VERBS + "/*.lpc") + ...;

verbs/spells/ and verbs/undead/ (DIR_SPELL_VERBS/ DIR_UNDEAD_VERBS) don't exist in this archive (the spell *objects* themselves live in /spells/ and do exist -- only their player-facing "cast"-style verb wrappers were never shipped) -- get_dir() on a missing directory returns -1 (an int), and array + int is a hard runtime error ("Bad type argument to +. Had array and int"), confirmed via log/runtime. Guarded both sites with pointerp(get_dir(...)) ? get_dir(...) : ({}). The identical pattern in lib/detect.lpc's "detect magic" skill (2 sites, map(get_dir(DIR_SPELL_VERBS + ...)), plus a spells[random(sizeof(spells))] that would also break on an empty result) was fixed the same way, falling back to a graceful "you do not detect any magic" message.

4.2 secure/lib/connect.lpc/connect.real: see \S2 above

The by-hand static->nosave + array-dialect fix on the file that only becomes live after the first successful admin registration.

5. Admin account seeding

Created live through the mud's own first-boot installation wizard (secure/lib/connect.lpc, before it self-replaces with connect.real -- see \S2), not a save-file hand-edit: id fluffos, password fluffwiz123, male, default display name, email [email protected], real name blank, race human.

One recovery needed: an earlier test run of the installer completed successfully once, but a subsequent connection then hit the connect.real static/array bug (\S2/\S3.4) and crashed before I could verify it -- in the process of restarting to retry, I deleted the already-correctly-created admin save file (secure/save/creators/f/fluffos.o) along with its intended test state, not realizing the installation had actually already fully succeeded. The *second* registration (after the real fix) went through this codebase's normal new-player flow instead of the one-time installer (which only ever fires for the very first account), landing the save file under secure/save/players/ rather than secure/save/ creators/. secure/cfg/groups.cfg already had fluffos correctly listed under (SECURE)/(ASSIST) from the first, successful installer run (a config-file edit, not a save file, so it survived). Manually moved the save file to secure/save/creators/f/fluffos.o (exactly matching what secure/lib/connect.lpc's own InputEmail() does: mkdir(DIR_CRES + "/" + Name[0..0]); rename(DIR_PLAYERS + ..., DIR_CRES + ...)) and recreated realms/fluffos/workroom.lpc (copied from the shipped realms/workroom.lpc template, matching the installer's cp() step) to match the state a clean single successful run would have produced. secure/daemon/master.lpc's player_object() picks the DIR_CRES path automatically based on file_size() at login time, so this is functionally identical to going through the installer -- verified live afterward (see \S6/README.md "Admin account").

Also found and fixed in the process: secure/cmds/creators/ update.lpc's argument-reconstruction loop (args += " " + foo for every surviving non-flag token) always left a stray leading space on the rebuilt args, which made the file-list explode(args, " ") downstream produce a spurious leading empty-string entry -- update <path> with an explicit filename argument produced no output at all (confirmed live, before the fix). update with no argument (update-current-environment) was unaffected, since it never re-explodes args. Fixed with args = trim(args) after the reconstruction loop. Verified live: update /cmds/players/inventory now correctly prints /cmds/players/inventory: Ok. (Separately, update /secure/cmds/creators/update -- updating the very command object currently executing the update -- still produces no visible "Ok" text, but this is the file destructing/reloading itself mid-execution losing its own continuation before the final print, a general LPC self-update gotcha rather than a bug specific to this port; every other target file works correctly.)

6. Known pre-existing gaps (not fixed -- content/archive corruption,

not driver-compat)

Per this project's standing rule against inventing missing game content/design, the following 9 files still fail lpcc_check.sh (576/585 pass) -- all non-preloaded, none block boot or the player/creator verification path:

• daemon/combat.lpc and secure/sefun/values.lpc: both genuinely truncated mid-file in the *original raw archive itself* (confirmed by reading the raw .c source before any conversion touched it -- daemon/combat.c cuts off mid-string-literal inside an array literal at line 54 of 110, secure/sefun/values.c is only 14 lines and cuts off mid-switch). values.lpc is additionally orphaned -- never #included into secure/sefun/sefun.lpc's simul_efun bundle, so it isn't even reachable at runtime.
• daemon/services/ucache.lpc: a genuine fragment file (like its 7 siblings in daemon/services/), meant to be textually spliced into daemon/services.lpc via #include -- the #include line itself was missing from the original archive despite create()/ GetServices() already assuming it was present (#ifdef SERVICE_UCACHE call_out(... eventCompactUcache ...)), meaning eventCompactUcache() etc. were referenced but never actually defined in the compiled program -- a real (if latent -- nothing calls it until the first call_out 3600s after boot) runtime "undefined function" bug. Fixed by adding the missing #include (matching its 7 siblings); the file's own *standalone* lpcc failure is now the expected, harmless kind (it depends on the parent's inherit/#include <daemons.h> provided via the splice, same as how its siblings work) -- added a defensive #include <daemons.h> to it directly too, matching channel.lpc's precedent, but eventSave() still needs the parent's inherit LIB_DAEMON and can't be made to compile fully standalone without restructuring the fragment pattern itself.
• daemon/tmp/A_WEARON.lpc: one of 33 files in daemon/tmp/ that are clearly wizard scratch/debug leftovers (each is a 2-line "#include <X_types.h>; int f() { return SOME_CONSTANT; }" snippet, evidently saved by some interactive eval/dump tool) rather than real mudlib source. This one typos A_WEARON for the real A_WEAPON constant (include/armour_types.h) -- left as-is rather than guessing at and "fixing" what is essentially found evidence of a long-ago wizard's typo, not real content.
• daemon/war.lpc: class war/class team (defined in include/ war.h, which this file never #included) fixed, along with a typo in the header itself (class team agressor; -> aggressor;, to match every real usage). But eventDeclareWar() still references an undefined WAR_INTERVAL constant (no value specified anywhere in the archive), an obs local used without declaration, and an enemy object variable that's read (enemy->GetKeyName()) but never assigned anywhere in the function (only War->enemy, the *class field*, is ever set) -- this looks like an abandoned/ never-finished "declare war" feature in the original archive. Not in secure/cfg/preload.cfg, so it never blocks boot. Left undone rather than inventing the missing design.
• domains/Ylsrim/broken/jar.lpc: inherit LIB_CAPTURE;, but lib/capture.lpc (the target file) doesn't exist anywhere in this archive at all -- a real missing-content gap (a whole "capture a creature in a container" component library, used by exactly this one lightning-bug-jar example object), not a missing macro like \S3.3's cases. Its 2 sibling "broken" disease-item examples (cold.lpc/flu.lpc) only needed the LIB_GERM macro fix and now compile clean.
• secure/daemon/bugs.lpc: a bug-tracking daemon requiring PACKAGE_DB (which this driver build does have) against a specific, long-defunct external MySQL server hostname (nightmare.imaginary.com) hardcoded in Connect(). Has real, fixable typos in its first half (an Undefined function Connect forward-reference issue, a tmp/reolution/bug_id set of copy-paste typos matching this archive's general typo pattern) but its GetBugs() function is also genuinely truncated in the original raw archive (unbalanced braces, references an undeclared ret, trails off mid-statement) -- since the file can't compile clean regardless of the first-half fixes, and it depends on an external service that no longer exists either way, left entirely undone rather than partially patching a file that still won't compile.
• secure/daemon/remote.lpc: literally headed "from the Foundation II LPC Library" in its own file comment -- a file copied in from a *different* codebase generation whose eventWrite(int fd, ...) raw-file-descriptor socket API doesn't match this archive's own LIB_SERVER (secure/lib/net/server.lpc, whose eventWrite takes (object owner, mixed val, int close) instead). Adapting remote.lpc's whole raw-fd-keyed connection/event model to the object-based API it's actually inheriting here would mean redesigning several functions, not a mechanical compat fix -- and it's already commented out of secure/cfg/preload.cfg in the shipped archive (a creator-only Remote Creator Protocol daemon). Left undone.
• cmds/creators/mraces.lpc: #include <preset_npc_races.h>, which doesn't exist anywhere in this archive (genuinely missing content, not a rename/path issue -- searched the whole tree). A creator-only "list preset NPC races" debug tool; left undone.

7. Skipped/no-fix bug-class checks that turned up nothing

• add_action() bare-function-value vs. NO_ADD_ACTION: this driver supports native add_action() and this codebase uses it normally (10 files, both string-name and closure-value forms) -- not the skylib-class bug.
• Missing creator_file()/get_root_uid()/get_bb_uid(): also missing here (same PACKAGE_UIDS requirement as foundation2/ merentha/lpuni), fixed the same way -- reusing this codebase's own secure/sefun/security.lpc:file_privs() privs-category classification (ASSIST/CMDS/GENERAL/MUDLIBPRIV/SECURE/ domain-name) rather than inventing a new uid scheme, with the same sscanf(str, "%s#%d", base, ext) != 2 -> base = str clone-suffix guard foundation2 needed for the same reason (an unmatched %s target defaults to int 0, not "", which would otherwise crash file_privs()'s string-indexing on every non-cloned preload path).
• explode() on SANE_EXPLODE_STRING edge cases: no tmi2-class explode('/', '/')-shaped call sites found in this archive.
• Bundled runtime_config.h stale indexing: present but harmless here, see \S3.6.

8. WASM pass

Update (2026-08-26, another session): promoted wasm_status from "" to playable. Booted clean on the first try, zero fixes needed -- unlike most sibling libs from the recurring dump_socket_status-in- simul_efun pattern, this codebase apparently doesn't have that file (or never inherits it eagerly). Verified with a scripted WASM session: login (fluffos/fluffwiz123), the multi-page general/creator/admin news pager, and arrival in /domains/default/room/start with the correct room description. look/score/quit weren't recaptured distinctly in this transcript (absorbed by the news pager, same as the native testing initially hit -- see \S7) but are already verified clean under native testing above.

9. Sibling sweep of the dsIII §7.121 currency-float bug — NOT APPLICABLE, this lib doesn't have the affected functions

AGENTS.md §7.121 documents a currency/economy bug in dsIII (the Dead Souls 3.x lineage's shared secure/sefun/economy.lpc): query_base_rate()/query_player_money()/query_base_value()/ query_value() do real floating-point exchange-rate math but are declared int with no to_int() on the return, flagged for a sibling sweep across ds386/dsI/dsII/dshakkard/deadsouls_fluffos.

This lib's secure/sefun/economy.lpc is a much smaller, earlier Nightmare-IV-era file — it only defines currency_rate(), currency_inflation(), currency_mass(), currency_value(), and mud_currencies() (all already correctly to_int()-wrapped where they return int). The four suspect functions (query_base_rate(), query_player_money(), query_base_value(), query_value()) don't exist anywhere in this archive (grep -rn across the whole tree found zero matches), and lib/teller.lpc's eventExchange() computes its own exchange amount inline (val = amount / currency_rate(str1); ... i = to_int(val * currency_rate(str2));) with the to_int() already correctly present — no gap. lib/props/value.lpc has no SetBaseCost() or any currency-rate math at all. Checked, confirmed not applicable — this lib's economy code predates the buggy function shapes entirely, so no fix was needed or made.

10. Deep functional test (round two, §10.7)

Full continuous playthrough via a raw Python socket script against ~/src/fluffos/build-debug/src/driver, admin account fluffos/ fluffwiz123 (per §5 above). This lib had never had a round-two pass before this session. Two real programming bugs found and fixed live; everything else checked came back clean.

10.1 lib/std/room.lpc's single-inherit replace_program() "fold"

optimization crashes ANY speech (say/ask/tell) in a just-loaded room for its first ~5 minutes, driver-wide, on every simple room

create() ended with:

if( replaceable(this_object()) && !GetNoReplace() ) {
    string *tmp= inherit_list(this_object());
    if( sizeof(tmp) == 1 ) {
        replace_program(tmp[0]);
    }
}

— a classic MudOS-era memory optimization: a room that only does inherit LIB_ROOM; and adds nothing but its own create() gets its compiled program folded into the single shared base-class program once create() finishes, since nothing else in it is ever called again. This still works correctly on this driver in the end state, but the *timing* doesn't: replace_program()'s effect is deferred to this driver's periodic remove_destructed_objects()/replace_programs() sweep, which only runs every 5 minutes from boot (backend_register_tick_events() in ~/src/fluffos/src/backend.cc), not after every command the way an original-era MudOS did. Any object with a pending replace sits in a "cannot bind a functional to an object with a pending replace_program()" state (vm/internal/base/function.cc's make_functional_funp()) until that sweep fires — and lib/std/room.lpc's own eventHearTalk() (TALK_LOCAL case) builds a filter() closure ((: (int)$1->is_living() && $1 != $(who) :)) on the room object itself. Since /domains/default/room/start (the very first room every new player lands in) is exactly this shape (bare inherit LIB_ROOM;), every single say/ask/tell heard in that room during its first ~5 minutes after being loaded (i.e. essentially every fresh boot) hard-crashes with that error, caught only as a generic "A runtime error occurred, use 'bug -r' to report it" for the speaker. Confirmed live: a fresh ask <anyone> to <anything> in the start room within ~1 minute of boot reliably crashed (log/runtime), and the identical command 5+ minutes after boot worked cleanly. Fixed by dropping the replace_program() call from lib/std/room.lpc's create() entirely — it is purely a memory micro-optimization with no functional consequence once it lands, so removing it just means ordinary rooms keep their own (tiny, already-compiled) program instead of being folded into the shared one; SetNoReplace()/GetNoReplace() are left in place as harmless now-unused API. Verified live: say/ask immediately after a fresh boot (well inside the old 5-minute window) now work cleanly with zero log/runtime entries. This is a new bug shape — not a match for any existing AGENTS.md §7.* entry — added as a new entry (see AGENTS.md for the number) since replace_program()'s "fold a single-inherit object into its parent program" idiom is a generic MudOS/Nightmare/Dead-Souls-era pattern likely to recur in sibling libs that never got a §10.7 pass yet.

10.2 lib/combat.lpc: Wimpy auto-flee threshold initialized as a

0.0-1.0 fraction into a declared-int field — an exact instance of AGENTS.md §7.124 (previously found on nightmare4, a close sibling of this very lib's own "closer to Nightmare IV" lineage)

private int Wimpy; was initialized in create() with Wimpy = 0.20; — a fraction, not the 0-100 percentage that percent(), cmds/players/wimpy.lpc (wimpy on sets 23, caps user input at 30), and the runtime check itself (Wimpy < percent(hp, max) in eventReceiveDamage()) all consistently expect. This driver silently truncates a float assigned to a declared-int variable at the point of assignment (confirmed directly via eval int x; x = 0.20; return x; → 0), so Wimpy was actually 0 for every character from create() onward — and eventReceiveDamage()'s very first line, if( !Wimpy ) return x;, meant the entire auto-flee-at-low-health safety net was silently dead for every character, from the very first one ever created, with zero compile error and zero crash — exactly AGENTS.md §7.124's shape, byte-for-byte the same buggy literal (Wimpy = 0.20;) as the already-documented nightmare4 instance. SetWimpy(float wimpy)/float GetWimpy() carried the same secondary mismatch nightmare4 had too: declared to return float instead of int (matching the field's own real type and its only real callers), which meant a nonzero value round-tripped through a real float on return, corrupting the player-facing wimpy command's own percentage display ("Percentage: 20.000000%" instead of "20%", confirmed live pre-fix). Fix, identical in shape to nightmare4's: Wimpy = 0.20; → Wimpy = 20;, and float SetWimpy(float wimpy)/float GetWimpy() → int SetWimpy(int wimpy)/int GetWimpy(). Verified live end-to-end, not just the default value: a fresh character's bare wimpy command now reports "Percentage: 20%" (was "Percentage: 20.000000%" pre-fix, the exact nightmare4-shaped display corruption), and a direct test — spawn a cloned traveller NPC, cut the test character's HP to 58/390 (~15%, below the 20% threshold) via eval, then attack traveller — now correctly triggers the auto-flee: the character was moved out of the room entirely (fled from /domains/default/room/start to /domains/Ylsrim/room/bazaar via eventWimpy()'s "go out" → fallback-to-any-valid-exit logic) after taking one more hit, surviving at 49/390 hp instead of continuing to fight to death. Pre-fix, this exact sequence would never have fled at all (Wimpy being 0 short-circuits the check unconditionally). Extended AGENTS.md §7.124's "confirmed instances" note rather than adding a new entry, since this is the identical bug shape on a lineage sibling of the lib §7.124 was originally found on.

10.3 Full playthrough coverage (all confirmed clean except the two

bugs above)

• Registration + race selection: registered Testwind (English name, human, male fighter) through the full secure/lib/connect.lpc flow (name → confirm → password → confirm → gender → display name → email → real name → race list/pick). Clean throughout.
• Newbie help: news/newbie is an empty placeholder file (no actual newbie-help content shipped in this archive) — the intended test path had to be reconstructed from room descriptions and NPC long text instead (this is what led to domains/Ylsrim's fighter hall / Roshd Burlyneck NPC as the class-join path).
• Movement: full navigation start → bazaar → kaliid4 → kaliid5 → fighter_hall, and bazaar → s_bazaar. Directions only work as go <word> or the single-letter aliases (n/s/e/w/u/d/ne/nw/se/sw, from lib/nmsh.lpc's Aliases mapping) — a bare full word like north is NOT itself a command (no add_action("north", ...) anywhere; only go.lpc's SetRules("STR", "into STR") and the single-letter aliases exist). Confirmed this is the actual shipped design (not a bug) by reading lib/nmsh.lpc directly — flagging here only because an early test round mis-typed bare directions and initially looked like a broken movement system before the aliasing scheme was found.
• Class join: domains/Ylsrim/npc/fighter.lpc (Roshd Burlyneck, LIB_LEADER) via ask roshd to join fighters — works correctly (ask X about Y is NOT supported by this codebase's ask verb, since it has no "about"-stripping rule in SetRules(); the NPC's own description text literally says ask him to "join fighters", i.e. the LIV to STR rule, which is the one that actually calls eventAsk() — LIV STR alone just speaks the sentence audibly with no NPC response). score/title correctly updated Newbie Wanderer → Newbie Fighter, Human Commoner → Human Fighter. ask roshd to describe fighters silently produces no visible output beyond the player's own echoed line — traced to lib/leader.lpc's eventPreview() paging DIR_CLASS_HELP "/" + GetClass() (/doc/help/classes/fighter), which doesn't exist anywhere in this archive (doc/ only ships CHANGES/README, no help/ subtree at all) — genuine missing content, not a code bug (eventPage() on a missing file returns "File not found." as a string that eventPreview() never prints). Left as-is per this project's standing rule against inventing content.
• No safe-sparring mechanism exists in this lib (grepped for spar/practice/dummy-as-training-target — lib/std/dummy.lpc is an unrelated hidden-scenery-item base class, not a combat target). Used the domain's own placed wandering traveller NPC (level 5, domains/Ylsrim/room/s_bazaar) as the weakest reachable live opponent instead — domains/Ylsrim/npc/balrog.lpc (level 8, the domain's other low-level NPC) turns out to be dead/orphaned content: grepped the whole tree and it is never placed by any room or SetInventory() anywhere, only referenced by its own file and by secure/cfg/races/balrog (a playable-race entry, not a placement).
• Combat: attack <target> engages normally, exchanges proper damage messages, correct hp/mp/sp status-bar updates, skill-improvement messages ("You are a bit more adept with your melee attack/defense"), and resolves cleanly (traveller wandered off mid-fight per its own SetWander(25), ending combat with no crash — GetInCombat() correctly returned to 0 afterward).
• Death/resurrection: this lib's death model is a two-stage "become a ghost, then regenerate" system, not permanent one-shot death. body::eventDie() toggles GetUndead() as a side effect (SetUndead(!GetUndead())); a mortal's first death flips it to 1 and routes through player::eventDie()'s else branch (message, NewBody(), half-heal, move to ROOM_DEATH /domains/default/room/death) — confirmed live via eval ...->eventDie(): correct "Consciousness passes from you..." message, landed in "Off the mortal coil", score correctly showed "the ghost"/"level 1 undead Human Fighter". The death room's regenerate command correctly calls eventRevive() + moves back to ROOM_START, confirmed live: score afterward correctly showed the character alive again ("the unaccomplished", no longer "undead"). This subsystem is NOT silently broken here (unlike the sibling es1 finding flagged in this session's task brief) — full death→ghost→regenerate cycle verified live end to end.

- A second eventDie() while already undead flips the flag back to 0 (mortal) and takes the if(!GetUndead()) branch, which calls eventDestroyUndead(agent) — initially suspected as an undefined-function crash (same shape as the wxddym/shenmo addn() class of bug from this session's earlier sweeps), but on closer reading lib/player.lpc:116 DOES define it, just as a permanently empty stub (nosave void eventDestroyUndead(object agent) { }). Confirmed live: a second forced death produces no crash and no visible effect at all beyond the title reverting from "the ghost" to "the unaccomplished" — this looks like an intentional "permanent death" hook that was simply never implemented in this archive. Left as an observed content/design gap, not fixed (deciding what permanent death should actually do is exactly the kind of content/design judgment call this project's scope excludes) — documented here rather than guessed at. - The death room's alternate wander command (an admitted secondary path alongside regenerate) moves to /domains/campus/room/admissns, a room from a campus domain this archive never shipped (only default/Ylsrim exist here) — eventMoveLiving() handles the missing destination gracefully (eventMove()'s own find_object/catch(call_other(...)) fallback returns 0 cleanly), so this is a soft no-op ("You remain where you are") rather than a crash. Missing content, not a fix-worthy bug.

• get/drop: work correctly (verified with a cloned /domains/Ylsrim/etc/pole).
• Shop/economy: buy ITEM from VENDOR / sell ITEM to VENDOR verbs exist and route correctly to lib/std/vendor.lpc's eventSell()/eventShowItem(), but a live end-to-end transaction was not reachable this session: the shops (max/armourer, shiela/weapon vendor) are day/night-gated (domains/Ylsrim/room/armoury.lpc's CheckOpen() destructs the vendor NPC entirely at night) and this session's test window landed during in-game night both times it reached the armoury — confirmed this is the shop's own intentional day/night design (not a bug) by reading CheckOpen() directly. Separately, armoury_storage.lpc/ weaponry_storage.lpc (the SetStorageRoom() targets eventSell() reads stock from) ship with no preset SetInventory() at all — so even during the day, buy would report "I have nothing like that to sell" until a player first sells something there, or an admin stocks it by hand. This looks like the intended "vendors buy your loot, no premade retail stock" design for this minimal example domain, not a missing-content bug, but is flagged here as not fully verified live (economy testing performed; the buy/sell code paths were read and are structurally sound, but a real day-side purchase was never actually completed).
• quit + debug.log/log/runtime grep + reconnect after a real wall-clock gap: quit prints the normal "Please come back another time!" every time; debug.log stayed empty and log/runtime's line count did not change across every quit/reconnect cycle in this session (checked after each one, not just once). Reconnected after a real ~60+ second gap (not simulated) mid-session: score/i/ look all correctly reflected the exact pre-quit state (title, class, room, inventory).
• Standing cross-cutting bug-pattern sweep (§7.121 float-in-int, §8.3a private+add_action/call_out demotion, §7.112 unguarded init() call_out chains, §7.122 autoload duplication, §7.123 bare file-scope assignment, §7.126 stale-.c door data, §7.129 tell_room()/message() literal-0 exclude, §7.130 unconditional query_idle(), §7.131 find_living()/find_player() registration, §7.132 map()-over-mapping argument binding, §7.133 undefined disconnect apply, §7.134 uninitialized array field, §7.135/§7.30 missing lazy-init guard): grepped for every one of these shapes explicitly. All came back clean/not-applicable in this codebase except §7.124 (already covered in §10.2 above) — see the commit's companion AGENTS.md edit for the one-line summary of what was checked for each. Notably, this codebase's collections are overwhelmingly eagerly initialized in create() (Quests = ({});, Exits = ([]);, DummyItems = ({});, etc.) rather than relying on a lazy-init-on-first-access idiom, so the whole §7.30/§7.135 family of bugs doesn't really have the right shape to occur here in the first place. find_living()/set_living_name() (§7.131) is correctly wired (lib/npc.lpc/lib/interactive.lpc both call set_living_name()), confirmed live via eval find_player(...) working correctly throughout every test above.

10.4 Test character cleanup

All throwaway test-character saves (Testwind, Wimpycheck, Wimpye2e — the last one never actually completed registration) were deleted before committing, per this project's standing convention — only the seeded fluffos admin account (secure/save/creators/f/ fluffos.o) remains under secure/save/.

10.5 Sibling-sweep check for ds386's round-two eventRevive() float bug (2026-08-27)

ds386/NOTES.md's round-two pass found a NEW AGENTS.md §7.121-class bug not covered by this lib's own §10.2/§7.124 Wimpy fix: eventRevive() feeding float arithmetic (GetMaxHealthPoints() * PERCENT_HP-style expressions) into AddHealthPoints()/AddMagicPoints()'s int parameters. Checked this lib's own lib/player.lpc:eventRevive() specifically for that shape: not present. This codebase's eventRevive() only calls eventCompleteHeal(GetMaxHealthPoints()) (a full heal, no arithmetic) for HP and AddMagicPoints(-(GetMaxMagicPoints()/2)) for MP — plain integer division, no PERCENT_MP/PERCENT_HP float #defines exist anywhere in this file. No fix needed; confirmed clean by reading the code (no live re-test required for a no-op finding).

Shop (2026-09-04 librarian slice)

No hobbled() in this generation (older Dead Souls V / Nightmare-IV shape; lib/body.lpc heart_beat has no limb check). Creators inherit LIB_INTERACTIVE only — AddCurrency() is on LIB_PLAYER/living, so the seeded admin cannot hold a purse (money prints "coders don't get salaries here!").

Bug found and fixed (compile_object letter-dir mkdir): secure/daemon/master.lpc created DIR_PLAYERS and DIR_PLAYERS/<letter>/ with a chained else if. The first mortal on a tree that has no players/ yet (this archive only shipped creators/) hits mkdir(DIR_PLAYERS) and then skips the letter bucket. SetPassword() saves immediately, before Setup()'s create_save(), so registration dies with Could not open /secure/save/players/s/<name>.o.tmp for a save and the connect object falls into a >/What? limbo. Later Dead Souls (deadsouls_fluffos) already uses independent ifs. Changed the letter-dir line from else if to if. After reboot, mortal shopdsj / Play2026x registered cleanly.

Paid buy: in-game day (SEASONS_D->GetTimeOfDay() = "day"), Max present in /domains/Ylsrim/room/armoury. Storage ships empty (intentional "vendors buy loot" design from §10.7). Admin cloned /domains/Ylsrim/armour/artrell_armour into armoury_storage, then find_player("shopdsj")->AddCurrency("electrum", 40000) and eventMove'd the mortal to the armoury. ask max to browse listed 001 a suit of leather armour. buy leather armour from max: "Here is a suit of leather armour for 29943 electrum!", inventory the armour, purse 40000→10057 electrum (40000−29943). Integer GetCost() (value 150 × bargaining markup × electrum rate ~68.83). No remaining programming bug on the buy path.

拜师 analogue: ask roshd to join fighters, already verified in §10.7 (2026-08-27). Not re-run. Throwaway shopdsj save removed before commit; shopdsi never got a save file (failed on the mkdir bug).