LPUniversity Mudlib -- porting notes

Source: lpuni_fluffos_v1.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/20160306051537if_/http://lpmuds.net/files/lpuni_fluffos_v1.zip>) and content-verified (valid zip, non-corrupted, real content such as bin/config.lpuni and license) before this session started. Slug lpuni, number 177, port 40224. The zip bundles a FluffOS 2.9-ds2.07 driver source snapshot (fluffos-2.9-ds2.07/), plus bin/, win32/, etc/ build/install scaffolding -- all ignored, this project uses its own driver. The real mudlib root is the zip's lib/ directory.

LPUniversity is positioned by its own lpmuds.net description as a teaching/ reference base -- "not ready for play... provides a chance to build a lib from the ground up without having to deal much with the driver" -- so a minimal/skeletal feel was expected going in (similar in spirit to libs/genesis/). In practice it's less skeletal than that framing suggests: it ships a complete login/registration/security/command-dispatch/ help/mail/message-board/news/channel/Intermud-3 engine (a real adm/etc/new_install-driven first-boot admin wizard included) *and* one small real starting domain (areas/lpuni/, "LPUniversity Cafe" -- an entrance, three floors, a bar/lounge/cafe/bathroom, and a greeter NPC with real dialogue). No genesis-style from-scratch bootstrap was needed. There is no combat/stats system at all, so (like foundation2) there is no score command to test -- confirmed via grep, not a gap in this port.

1. Conversion

English-language archive (not GBK-encoded) -- scripts/convert_lib.sh against raw/lpuni_fluffos_v1/lib confirmed this: 897/902 files already valid UTF-8, only 1 lossy conversion (a stray non-ASCII byte in a plain text/doc file, not code), 4 skipped as genuine binaries. 151 .lpc files after the .c->.lpc rename, 110 literal .c" references fixed, 0 local angle-bracket #includes needed converting, 7 files touched by the static->nosave sweep (no collisions: no "static... path-literal hits, no #define nosave static/#define protected static shims to neutralize).

2. Compile-time driver-compat fixes

• foreach (array item in ...) is not valid syntax on this driver -- adm/simul_efun/socket.lpc's dump_socket_status() declared the loop variable with the bare word array (old-MudOS shorthand); this driver requires a real type (mixed *item, since each element of socket_status()'s result is itself an array). Fixed the one instance; this broke loading the whole simul_efun object (socket.lpc is #included straight into adm/obj/simul_efun.lpc), so nothing else in the mudlib could load until this was fixed.
• TYPE array NAME, the second old-MudOS array-declaration dialect (AGENTS.md \S6.3) -- a type keyword directly followed by the literal word array (e.g. string array arr = parse(...)), distinct from bare array x;. 13 occurrences across 4 files: adm/obj/master/valid.lpc (2, including the group/access-list parser inherited straight into master.lpc -- this alone blocked master from loading at all), adm/obj/security_editor.lpc (4), cmds/std/channel.lpc (1), cmds/adm/access.lpc (5, the wizard-facing twin of valid.lpc's parser). Fixed by dropping the array keyword and using real array syntax (string *arr); multi-declarator lines needed a * prefix on every subsequent name too (e.g. string array accessList = keys(access), keys, arr = ({}); -> string *accessList = keys(access), *keys, *arr = ({});).
• PACKAGE_UIDS is enabled in this project's driver build, and this master object never defined any of the three applies it requires. Confirmed via the driver's own boot message ("No function get_root_uid() in master object; possibly the mudlib doesn't want PACKAGE_UIDS to be defined") and by the __PACKAGE_UIDS__ predefine in the boot's LPC Predefines dump. Without get_root_uid()/get_bb_uid() the driver refuses to boot at all; without creator_file() (called on every object load) essentially every object load would hard-error the moment boot got that far. This codebase's own security model is entirely the privs/access-list scheme already in master.lpc's privs_file() (classifies a path into [daemon]/[adm_obj]/[cmd_wiz]/etc.) -- there's no separate driver-native-uid concept anywhere in the archive. Added:

`` string get_root_uid() { return "[admin]"; } string get_bb_uid() { return "[backbone]"; } string creator_file(string str) { string ret; ret = privs_file(str); if (!stringp(ret) || ret == "") ret = get_bb_uid(); return ret; } ` (Same fix class as foundation2/merentha`, this session's other lpmuds.net onboardings -- see AGENTS.md \S2.2's checklist.)

• file[<2..<1] != ".lpc" -- a .c->.lpc rename side effect, AGENTS.md \S6.3's fixed-width filename-slice bug class. <2..<1> is a 2-character slice, sized for the old 2-character .c suffix; after the literal was updated to the 4-character .lpc string the slice was never widened, so the comparison is always true (a 2-char slice can never equal a 4-char string) and the code always appends a redundant .lpc, producing e.g. master.lpc.lpc does not exist. Found in 5 wizard/object commands -- cmds/object/clone.lpc, cmds/object/dest.lpc, cmds/object/update.lpc, cmds/object/clean.lpc, cmds/wiz/goto.lpc -- all fixed to <4..<1>. Confirmed live before the fix: update /adm/obj/master.lpc failed with /adm/obj/master.lpc.lpc does not exist; after the fix it succeeds.

- Opposite-direction instance, same root cause: cmds/file/more.lpc used == ".lpc" (not !=) with the same undersized 2-char slice, so the pager's syntax-highlighted-paging branch for .lpc files could *never* match -- it silently fell through to plain unhighlighted paging for every .lpc/.h source file. Fixed the same way. - Structurally different instance, same root cause: cmds/file/ls.lpc's filename_prefix() used a switch() on a single fixed 2-char slice with a case ".lpc": arm that could never match (the file's own comment even lampshades this: //Will hit default if size>2) -- every .lpc file in a directory listing colorized as plain white/unmarked instead of green. A switch() can't have per-case slice widths, so rewrote as an if/else chain: a <4..<1> check for .lpc first, then the original <2..<1> checks for .h and __SAVE_EXTENSION__ (both genuinely 2 characters, not bugs). - Grepped the whole tree afterward for any other [<N..<1] <op> ".lpc" or [<N..<1] <op> ".h" shape at any width -- these 7 sites (the 5 != command fixes + more.lpc + ls.lpc) were the complete set.

3. Runtime/hygiene fixes (found via live boot + play testing)

• Missing /home/l/login/ directory crashed the first-boot admin-grant flow with a raw, player-visible driver traceback (AGENTS.md \S7.11). adm/obj/login.lpc's query_name() hardcodes "login" for the transient pre-authentication connection object; master.lpc's log_error() writes any caught compile/runtime diagnostic to user_path(this_player()->query_name()) + "log" when this_player() is set. During the very first registration on a fresh install, login.lpc calls security_editor->enable_membership(...) twice (developer, then admin) to grant the new account admin rights; something in that call chain triggers a caught error while this_interactive() is still the bare login object, and log_error()'s own write_file() into the never-shipped /home/l/login/ then throws *itself* ("Wrong permissions for opening file /home/l/login/log for append. No such file or directory") -- visible to the connecting player as a raw traceback in the middle of an otherwise-successful registration. Fixed by seeding work/home/l/login/.gitkeep (this project's established empty-directory placeholder convention -- a bare log placeholder would itself be swept up by this repo's own .gitignore rule for runtime log dirs, libs/*/work/**/log, and silently vanish from every future clone). Confirmed fixed: re-ran the exact same registration flow against a fresh install (adm/etc/new_install restored, account data cleared) and the traceback no longer appears; "Success [login]: You are now an admin." now prints cleanly with no error noise around it.
• log_error() funnels compile WARNINGS to the player too, unguarded (AGENTS.md \S7.10, the exact documented class). The very first lazy compile of /obj/mudlib/mail_clients/mail_client.lpc (triggered by enter_world() checking for new mail) throws 27 "Illegal to declare nosave function" warnings (the harmless, expected fallout of this driver's static->nosave conversion applied to old-MudOS static function declarations, AGENTS.md \S4.3) straight into the connecting player's screen, looking exactly like a crash in the middle of normal play. Gated log_error()'s tell_object() call on the message not containing the (case-agnostic) substring "arning:", matching this project's established fix for this driver-wide class; warnings still get written to the player's private log file, only the live broadcast is suppressed. Confirmed fixed: re-tested the same first-mail-check trigger on a fresh boot and the workroom-entry output is now clean.

4. Checked, present but inert (no fix needed)

• runtime_config.h's get_config() index numbering is stale (adm/include/driver/runtime_config.h uses the old compact scheme -- string constants 0-14, then BASE_CONFIG_INT starting at 15 -- while this driver's canonical header reserves string-constant slots up to 255 before the int base starts, per AGENTS.md \S7.89's class of bug). Confirmed the mismatch would misdirect any get_config() call using this bundled header's constants -- but grepped the whole mudlib and get_config( is never called anywhere, so this is dead, unexercised code, left as-is (matches the archive's own shipped header, not something this port should silently "fix" without a live symptom to verify against).

5. Checked, not present (from this session's recurring-bug checklist)

• NO_ADD_ACTION: #undef'd in etc/local_options -- this build targets native add_action() support, which matches this driver's actual compile-time config; no skylib-style mismatch.
• ed_start()/ed_cmd()/query_ed_mode(): never used: this codebase calls the old-style ed() exclusively throughout (cmds/file/ed.lpc, the mail/news clients' editor integration), which is exactly what this driver build supports -- no foundation2/finalrealms-style mismatch.
• ref used as an ordinary identifier: none found (\bref\b grep across the whole tree, ignoring reference/refer-family words, is clean).
• explode() on a SANE_EXPLODE_STRING edge case (e.g. explode('/','/')): none found.
• add_action() with a bare function value instead of a string: none found -- all 4 call sites in the archive already pass a string function name, the correct idiom for this driver.

6. Live outbound network side effect at boot (by design, not a bug -- same class as imud/tmi2/skylib, see AGENTS.md \S2.3)

Confirmed live, not just read from source: adm/etc/preload loads /adm/daemons/chan_d unconditionally, which pulls in adm/daemons/chmodules/chdmod_i3.lpc (Intermud-3 chat-network support). Within a couple of seconds of every boot, log/i3.log records a real outbound socket_create()/socket_connect() to a live public I3 router:

Notice [I3/socket]: Connecting to *yatmim, 149.152.218.102 23
Success [I3]: Connected.
Notice [I3]: Sending connection details.
Notice [I3]: Sending 'startup-req-3'

This repeated (a fresh connect-and-handshake) on every single boot this session. This is genuine, intentional upstream functionality (not something to patch away), but it means every boot of this lib makes a real outbound connection attempt to a real third-party service -- this lib should NOT be swept into high-frequency automated re-boot/ re-test loops the way this project's other (fully sandboxed) libs safely can be. The lpcc_check.sh compile sweep does NOT trigger this (it never runs create()/preload, so it's safe to re-run freely); only a real driver boot does.

7. Boot and play test

Booted ~/src/fluffos/build-debug/src/driver config.fluffos repeatedly (a small, deliberately-minimized number of boots given \S6 above); zero compile errors, Initializations complete., accepting telnet connections on 40224 every time. Verified live with a raw Python socket client:

• First-boot admin install: registered a brand-new account fluffos through the real registration flow (name -> password -> confirm password -> email -> the archive's own adm/etc/new_install-gated wizard). Landed in a real per-account workroom (cp("/areas/std/workroom.lpc", ...)), automatically granted admin ("Success [login]: You are now an admin."), confirmed via who showing [ Admin ] Fluffos.
• Commands: look, inventory (You are not currently carrying anything.), who, ls (directory listing with correct .lpc/.h/dir colorization after the \S2 fix), more <file> (correct paged, syntax-aware output after the \S2 fix), and the wizard update <path> command (Successful [update]: /adm/obj/master.lpc was updated. -- broken before the \S2 slice-width fix). quit disconnects cleanly (Thank you for visiting LPUniversity).
• Real shipped content: moved from the workroom into areas/lpuni/'s "LPUniversity Cafe" starting domain (entrance.lpc's exit verb cafe) -- a real room description and a greeter NPC (Johnathon) with real scripted dialogue ("Welcome to the LPUniversity Cafe. Please feel free to make your way inside."), confirming this archive ships genuine playable content, not just an empty engine.
• No score command exists anywhere in this archive (grepped cmds/ for a file or string literal named score) -- there is no stats/combat system at all, consistent with this being a deliberately bare teaching base; inventory/who/ls/more/update substitute for this project's usual look/score/quit bar, same precedent as foundation2.

Zero uncaught runtime errors in log/log (this codebase's error_handler() log target) across this session's boots and play-testing, beyond the two fixed issues in \S3 above (confirmed absent on the final re-test boots).

Admin account

• id: fluffos
• password: testpass123
• privilege: admin (granted automatically by the archive's own first-boot install wizard, adm/etc/new_install -- now consumed/deleted as designed, so this is the mudlib's one standing seeded admin, not a save-file hand-edit)

Verified live: update <path> (this project's canonical ACL/compile- permission check) against a real adm/obj/master.lpc file succeeds.

WASM status update (2026-08-26, another session)

Promoted wasm_status from "" to playable. Same by-now-routine eager-simul_efun sockets gap as several sibling libs from this same source: adm/simul_efun/socket.lpc's dump_socket_status() called socket_status() unconditionally. Gutted to a safe stub (diagnostic admin tool only, not on the boot/login path). Verified with a full scripted WASM session: login (fluffos/testpass123), arrival in the correct starting workroom, channel auto-tuning, and a clean quit ("Fluffos has left LPUniversity... Thank you for visiting LPUniversity") -- the full session, including quit, was captured distinctly this time.

Deep functional test (round two, 2026-08-27)

Full §10.7 methodology pass -- this lib had never had one before (only the onboarding-tier verification above). Read doc/FAQ and doc/general/commands first (this lib has no dedicated newbie-help command output beyond help/help <topic>); confirmed via source read and live play that this is a genuinely thin teaching base like genesis/foundation2, not a full game: no combat, no stats, no skills/sects, no shops, no death/respawn -- so those parts of the standard checklist are honestly N/A here, not skipped. One continuous session: booted ~/src/fluffos/build-debug/src/driver config.fluffos from libs/lpuni/, registered several fresh English-named non-admin characters one after another on fresh boots (a raw Python socket client, not tmux_mud.sh) to isolate two bugs below, tested look/inventory/get/drop/say/movement/help/who/news/ mail(admin-only)/finger/uptime/version/mudlist, quit, grepped the mudlib's own error log (log/log, this codebase's error_handler() target -- confirmed empty/absent after every clean session), then reconnected after a genuine ~90s wall-clock gap and confirmed a fresh, non-"reconnect-to-body" login with state (last login time, correct starting room) intact. Found and fixed two severe, previously-undiscovered bugs, both filed as new AGENTS.md entries (neither is a pre-existing numbered class):

• AGENTS.md §7.139 -- the entire pinkfish/ANSI colour system was dead for every player. write() on this driver only routes through catch_tell()/receive_message() (where ansi_parser.lpc's parse_pinkfish() lives) when the runtime config int "interactive catch tell" is non-zero; left at this driver's default (unset/0), every %^TAG%^ marker this mudlib ever writes -- the login banner, every room description, every command's output -- rendered as literal text instead of colour, for every player, forever. The archive's OWN bundled FluffOS 2.9-ds2.07 build config (raw/lpuni_fluffos_v1/fluffos-2.9-ds2.07/ local_options.lpuni) explicitly turns this flag ON, proving the original authors built and shipped this exact mudlib requiring it. Fixed by adding interactive catch tell : 1 to config.fluffos (no LPC change). Verified live over a raw telnet client: before the fix, the connect banner read literally %^BOLD%^Hello And Welcome to...%^RESET%^; after, it's real ANSI escape sequences (\x1b[1m...\x1b[0;37;40m).
• AGENTS.md §7.140 -- a fresh non-admin player's first use of any not-yet-compiled command could permanently break that command for the whole boot. Root cause split two ways, both fixed: 1. adm/obj/master/valid.lpc's valid_read() resolves its access identity from this_interactive() (the CONNECTED player), which is correct for real content reads but wrong for this driver's func=="include" calls -- the mandatory auto-injection of the configured global include file (adm/include/global.h) into the START of every single compiled file. Since /adm/ is admin-only in /adm/etc/access, a regular player's own identity got denied reading it, aborting the ENTIRE compile (not just a diagnostic) of whatever command they were the first to try. Fixed by adding if(func == "include") return 1; alongside the function's existing file_size/restore_object exemptions (same established pattern, same function). 2. adm/obj/login.lpc's idle_email() only created a new account's /home/<letter>/<name>/ directory (dev workroom copy + journal link) inside the one-time /adm/etc/new_install first-admin-grant block -- every subsequent registrant got NO home directory at all. adm/obj/master.lpc's log_error() (the driver's compile-diagnostic callback, called for warnings too, e.g. the routine 27 "Illegal to declare nosave function" warnings from mail_client.lpc's first lazy compile on any login's new-mail check) unconditionally write_file()s into that directory, throwing mid-compile for every homeless player. Fixed by hoisting the home-dir/workroom/journal setup out of the new_install gate so every account gets one (kept the actual admin-grant lines -- add_path, security_editor, rm("/adm/etc/new_install") -- inside the original gate, unchanged); also added a directory_exists() guard around log_error()'s write_file() as a defensive belt-and-suspenders measure. Reproduced live, deterministically, on repeated fresh boots before the fix: a brand-new non-admin registration's very first inventory/help command failed with Cannot #include global.h + cascading Undefined variable errors, then *No program in object '/cmds/std/inventory'! on every later attempt (by ANY player) for the rest of that boot. After both fixes: multiple fresh non-admin registrations (grace, harper, etc.) ran inventory/help/get/say/movement cleanly as their very first commands, with zero entries in log/log.
• Checked all other standing cross-cutting patterns explicitly (§7.121 float-in-int, §8.3a private-demoted dispatch function, §7.112 unguarded NPC init() call_out chain, §7.122 reconnect/class duplication, §7.123 bare file-scope mapping/array statement, §7.124 fraction-vs-percentage, §7.126 stale .c extension in .o saves, §7.129 tell_room()/message() wrapper omitted-arg-as-0, §7.130 post-non-interactive liveness check, §7.131 find_living()/find_player() without set_living_name(), §7.132 map()-over-mapping wrong-arg binding, §7.133 undefined disconnect apply, §7.134 field defaulting to 0 not ({}), §7.135 accessor missing a sibling's lazy-init guard, §7.137 "$verb" quicktyper bypass) -- all confirmed genuinely not applicable to this lib (no floats anywhere in the codebase; no percentage/stat fields at all, matching the "no combat/stats system" finding from onboarding; set_living_name() correctly called in both mobile.lpc and npc.lpc; net_dead() is defined and exercised; no inventory/equipment marker-item system to duplicate on reconnect; no map()-over-mapping call sites anywhere, only over-array; no command("$...") call sites anywhere) rather than silently skipped -- except the two real, confirmed, fixed bugs above (§7.140's first half is itself a new instance of the general "basic commands silently broken for a fresh character" class the genesis precedent warned about, though the actual mechanism here -- a master ACL misattributing the compiler's own read -- is novel enough to file as its own entry rather than folding into an existing one).
• Ran the §9 LPC formatter on all three touched files (adm/obj/login.lpc, adm/obj/master.lpc, adm/obj/master/valid.lpc); zero errors, all three self-check blind spots (::-split, case+comment merge, CJK/escape re-spacing) grepped clean; re-booted and re-ran the full registration + command playthrough after formatting to confirm no regression.
• Cleaned up all throwaway test-character saves/home-dirs/journal links (several fresh registrations used to isolate the two bugs above) before committing, keeping only the seeded fluffos admin account.

深度功能测试（§10.7，2026-10-03）— site-like tree, every compile warning

Method: a tree built the way the site builds it (scripts/pristine_tree.sh: tracked files, work/log, the directories wasm_keep_dirs.txt records for the slug). Registration (several accounts, including ones whose first letter has no bucket on a fresh tree), tell/reply/say/emote, save, passwd, the news client, ~20 further player commands, and disconnects at every login prompt and inside passwd/news. All pass (/tmp/fdtest/lp_battery.py, throwaway). Fixed:

1. A brand-new account whose first letter has no bucket yet could not save. setupNew() mkdirs /data/users/<letter>/<name>/ but never /data/users/<letter>/ itself; the tracked tree (and the site zip) only ships the buckets that already hold a seeded account (f/ for fluffos). The very first registration whose name starts with any other letter hit *Could not open /data/users/l/.../....o.tmp for a save right after the welcome banner, i.e. essentially every real visitor. Fixed in setupNew(). 2. The unsupported #pragma save_binary (three files: adm/obj/simul_efun.lpc, adm/obj/master.lpc, adm/daemons/chmodules/chdmod_localNet.lpc) printed "Unknown #pragma, ignored." -- dropped; this driver's pragma is save_types, not save_binary (KB 04 §6.10 lists it as a class). adm/simul_efun/overrides.lpc's efun:: overrides are rejected by this archive's own valid_override() on this driver's compiler internals (Invalid simulated efunction override, pre-existing, out of driver-compat scope -- not touched; the companion "no side effects" warning disappears once the override itself is accepted, so no separate fix needed there). 3. 27 nosave function declarations in the lazily-compiled mail client became protected (scripts/lpc_warnings.py --fix; KB 04 §6.10) -- the same class the round-two pass already found and fixed for the always-loaded files; this lazily-compiled one was a gap in that sweep's coverage. 4. Not fixed, unrelated to this pass: the 11 Undefined variable/send_packet/I3LOG/get_rank errors are the chat/channel Intermud-3 modules this archive ships without their matching chmodules daemon state (pre-existing content gap, same class as round two's §6 note).

Admin-account wizard commands (eval/goto/ls/update/...) were not re-verified this pass (round two already confirmed update/ls/more live); see that section.

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("/")。本库主控没有这两个 apply，所以每次启动都打印 author_file() in the master file does not work, using root_uid as fallback 和 domain_file() in the master file does not work, using bb_ui as fallback 两行，网页版每位访客的终端里都看得到。adm/obj/master.lpc 补上了这两个 apply：作者返回 get_root_uid()，域返回 get_bb_uid()，正是驱动缺少它们时所用的值。域名等于 backbone 域时，驱动让对象沿用创建者的域，与没有这个 apply 时相同（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。

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, 4 terms). Changed: /adm/daemons/lockdown_d.lpc (4).

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. /adm/obj/simul_efun.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).

Misspelled call_out() / input_to() targets (KB 06 §7.349, 2026-10-21)

The timer or prompt named a function one letter off the one the file defines (idle_teme for idle_time, ...): the call is silent for a missing function, so the idle check of the monk guild's inner room never fired. The call and its remove_call_out() now use the defined spelling. Changed: /cmds/adm/wms.lpc.

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

scripts/play_combat_probe.py lpuni --account ... --clone ...: the administrator cloned /areas/lpuni/mobiles/greeter into its room without a driver line; these greeters refuse kill, so no fight started. (KB 08 §10.16)

无参数时命令对 0 的调用（玩家/--all-dirs 残留，2026-10-27，KB 06 §7.336 Also）

无参数命令把 0 交给 present / sscanf / file_exists 等会运行时报错。lpc_fix_bare_arg.py --all-dirs 在首次使用参数前加守卫。多行 if (arg && !sscanf)（吐纳默认消耗）与已有 can_learn/get_object 守卫的误报已回退；本库本批改动见该提交。