西行战记 (xixingzhanji) — archive #85

What this is

archives/西行战记.gz is, per AGENTS.md's own note, "a bare gzip of a tar, not .tar.gz named" — tar xzf handled it transparently in one step (no separate gunzip needed), unpacking to a single top-level directory xxzj/.

The lib self-identifies as 西行战记 ("Records of the Westward Battle Journey" / "Journey West War Chronicle") — a Journey-to-the-West-themed wuxia/xianxia mudlib. bin/config.xxzj's name field says 西行战记[总站] ("headquarters/main station") but the live logind.lpc banner text says 西行战记[宜昌站] ("Yichang station") — the same kind of branding drift already seen on other archives in this project (a regional-site fork of a shared codebase, config field not updated to match).

Lineage: genuine ES II family (master.c header credits "original from Lil, rewritten by Annihilator (11/07/94)"), the same large family as es1_win/esI/xkx2001/rzrmud/bmxkx2001/kxkj/ yueyingqiyuan/wuhanzhan/haiyang2/yanhuangwuhun/yhyxs/ xuanjianlu/shenzhou/shenmo/zitengzhan/zhongjidiyu/xiyouji2003. Confirmed via master.c MD5 (after GBK→UTF-8 normalization) that this is not a byte-identical duplicate of any of those — it's its own distinct fork within the family, not a hidden re-run of an already-processed archive.

~6500 .lpc/.c files after conversion — a mid-size lib, well within lpcc_check.sh's normal (non-mega-lib) bracket.

Pipeline

1. Extracted with tar xzf archives/西行战记.gz -C libs/xixingzhanji/raw/ (worked in one step) → libs/xixingzhanji/raw/xxzj/. 2. Converted with scripts/convert_lib.sh → libs/xixingzhanji/work/: encoding (288 already-UTF-8, 6749 converted, 487 lossy, 27 skipped binary), .c→.lpc rename (6493 files), 1035 literal ".c" refs fixed, 25 angle-bracket <...c> refs fixed, 169 local angle-bracket <x.h> includes converted to quotes, static→nosave (67 files touched). 3. Port 40080 assigned (next free per TODO.md after 40076-40079 reserved for archives #27/#82-84's concurrent agents). libs/xixingzhanji/config.fluffos adapted from bin/config.xxzj (iconv'd BEFORE any edit, per §5): port number → 40080, mudlib directory/binary directory → absolute paths under this repo.

Conversion-pass extras (beyond convert_lib.sh's automated pass)

• 3 uppercase .C files missed by the rename glob (d/obj/drug/YUNDAN.C, daemon/class/yaomo/kusong/dali-bang/SHENLI.C, daemon/class/yaomo/kusong/huoyun-qiang/huohun.C) — manually renamed to .lpc. None referenced elsewhere by path, so this was cosmetic/hygiene, not a boot-blocker.
• One un-converted raw-GBK straggler that survived convert_lib.sh's file-classifier: d/sea/npc/beast1.lpc (an ISO-8859-classified file per file -b, not caught by the extension-based forced-text list because it's .lpc... actually it WAS on the forced list but still ended up raw -- likely a file-classifier edge case on this specific byte pattern, same class of gotcha AGENTS.md's Encoding section warns about generically). Found via the lpcc_check.sh sweep (Invalid UTF8 codepoint in string literal), fixed directly with iconv -f GB18030 -t UTF-8 -c. This also resolved two downstream cascade failures (d/sea/maze1, d/sea/maze5, both of which just reference this NPC and re-failed only because of it — confirmed one-shared-cause per §8g, not 3 separate bugs).

Fixes applied, and why

1. master.lpc's connect(): switch(port){ default: ... } is a hard compile error here — the one fix required to get past boot at all

The original connect(int port) used switch(port) { default: ... } with no real case label (this lib never actually dispatches behaviour by port — single-port mud). This driver's grammar requires at least one real case in any switch/case block; a default:-only switch is a parse error (need case statements in switch/case, not just default:), which took down master.lpc's entire compile (*No program in object '/adm/obj/master'!). Fixed by removing the switch wrapper entirely (the default: body already covered every port unconditionally, so a plain code block is behaviourally identical and simpler). Not previously catalogued in AGENTS.md — worth adding as a new entry (switch with only default:, no case, is a hard parse error on this driver).

2. adm/daemons/convertd.lpc — 44 instances of the stray-backslash-before-closing-quote typo (§8h, much larger count than previously seen) — fatal, blocked this preloaded daemon's compile

convertd.lpc's GBK/BIG5/Greek/pinyin charset lookup table has the same "\"," typo documented in §8h ("β\", instead of "β",), but instead of the usual 1-2 occurrences this file had 44 scattered throughout its ~7000-line table (verified via a byte-level Python scan, not just the naive grep '\\\\"' which under-matched due to embedded PUA/private-use-area characters sitting between the real content and the stray backslash on some lines — e.g. "α\xEE\x96\xA2\", had a PUA glyph U+E5A2 immediately before the erroneous \"). Fixed all 44 with a byte-level Python pass that only strips the erroneous \ immediately before a line-ending "/",, leaving every embedded PUA character intact. convertd is on the preload list, so this was a fatal, boot-blocking bug — the very first real error seen after fixing the switch issue above.

3. adm/simul_efun/chinese.lpc — §15h is_chinese() GBK byte-range check

if( strlen(str)>=2 && str[0] > 160 && str[0] < 255 ) (GBK lead-byte range, 2-byte minimum) replaced with a CJK Unicode codepoint range check (str[0] >= 0x4e00 && str[0] <= 0x9fff, 1-character minimum). Standard §15h fix — this directly gates whether real Chinese names are ever accepted during registration.

4. adm/daemons/logind.lpc — §15h check_legal_name() bounds + sliding-window fix

• strlen(name) < 2 || > 12 (GBK-byte bounds, i.e. "1 to 6 Chinese characters") halved to < 1 || > 6 — matches the message text itself ("必须是一到六个中文字" = "must be one to six Chinese characters").
• Dropped the i%2==0 && filter in the per-character Chinese-check loop — under UTF-8 every string index is already one full character (not every other byte), so every position needs checking now, not every other one.

5. adm/daemons/logind.lpc — §15p/§15ab: encoding_to_mudlist() unconditionally called into dns_master even after its own find_object() check failed

The GB/BIG5 selection handler already had if( !find_object(DNS_MASTER) ) write("...") — but then called DNS_MASTER->query_muds() and Mud_name() (itself a macro wrapping DNS_MASTER->query_mud_name()) unconditionally on the very next lines regardless of that check's result. Since "/path"->method() lazily loads the target object if it isn't already loaded, this would have triggered dns_master.lpc's real create() (a genuine socket_bind() + resolve() against an unreachable external boot server, confirmed by reading the file) on literally the first prompt after GB/BIG5 selection, for every single connection — exactly the class of hang §15p exists to prevent, except merely excluding dns_master from preload would NOT have stopped it here (§15ab's finding #2: inline calls bypass the preload exclusion). Fixed by removing the entire intermud mudlist-display block (purely cosmetic) and printing a short notice instead. dns_master was ALSO removed from adm/etc/preload as the standard §15p proactive step (belt-and-suspenders, since the daemon is never touched by anything else on the registration path either).

6. adm/etc/preload — typo: /d/npc/choose → /d/npc/choosed

The actual file on disk is d/npc/choosed.lpc (a faction/sect-choosing NPC daemon) — the preload entry was missing the trailing d. Master's own preload() already gracefully no-ops on a missing file (if (file_size(file+".lpc")==-1) return;), so this was silent/non-fatal, but a one-character, zero-risk mechanical fix, so applied. (Same graceful no-op already covers /adm/daemons/backupd and /adm/daemons/storyd, both of which only have a stale .bak sibling in the raw archive and no real .c/.lpc — confirmed these two are genuinely absent, not a typo, and left alone.)

7. master.lpc — §14 valid_override upgraded to 3-arg; §8d/§15o get_include_path() added; §4 explicit reentrancy guard on valid_read/valid_write

• valid_override(file, name) → valid_override(file, name, main_file), also allowing when main_file (not just file) is SIMUL_EFUN_OB/ MASTER_OB (fragments #included into simul_efun/master).
• Added get_include_path() (the standard §8d/§15o shape) so per-room/ per-NPC local flavor headers included with <angle.h> next to the including file resolve both at preload and mid-connection. Confirmed two real instances of this pattern: d/penglai/baihuagu.h (used by 25 baihuagu*.c quest-room files) and d/dntg/laojun/maze.h (used by 8 laojunmaze*.c files) — convert_lib.sh's automated angle-bracket-to-quotes conversion (169 files) already handles the common case where the header is quoted; get_include_path() is additional insurance for anything still using angle brackets.
• valid_read/valid_write already wrapped the lazy SECURITY_D load in catch() (unlike §4's canonical broken example, which had no catch() at all) — so the classic infinite-recursion crash was likely already avoided here. Added an explicit loading_security_d reentrancy flag anyway as cheap defense-in-depth, matching the catalog's standard shape.

8. master.lpc — §15w: log_error() was broadcasting every compile WARNING (not just real errors) to the connected player

Gated the player-facing efun::write("编译时段错误："+message) broadcast on strsrch(message, "warning:") == -1 (still logs everything to /log/log regardless). Also hardened error_handler() (declared void on this driver per docs/apply/master/error_handler.md — the mudlib's own comment "whatever we return goes to the debug.log" assumes a return-value path that doesn't actually exist here) with an explicit efun::write_file("/log/RUNTIME_ERRORS", ...) inside a catch(), as permanent insurance against silently-lost runtime errors.

9. adm/simul_efun/message.lpc — §15s tell_room() exclude-arg

varargs void tell_room(mixed ob, string str, object *exclude) passes exclude straight to message(..., exclude); when called in the overwhelmingly common 2-arg form, exclude is int 0 (varargs default), not an object/array. Guarded with exclude || ({}). Caveat: reading this driver's actual f_message() C++ implementation (packages/core/efuns_main.cc) shows the 4th-arg type switch already falls through to "no exclusion" for any type other than object/array (including T_NUMBER) with no bad_argument() call — so this specific driver checkout may not actually reject a raw int 0 here the way §15s's originally-documented case did. Applied anyway since it's a free, zero-risk fix matching the established pattern; not confirmed to have been a live bug on this specific driver build (unlike the other fixes in this list, which were all confirmed via a real compile/boot failure).

10. d/obj/books-nonskill/book-qujing.lpc — §15t pattern 1: absolute path inside angle-bracket #include

#include </d/qujing/obstacle.h> (an already-absolute path inside <...>, which this driver's angle-bracket resolver never special-cases) fixed to the quoted form #include "/d/qujing/obstacle.h".

11. obj/toy/{buwawa,mallet}.lpc — missing local message.h

Both use message_vision() via #include "message.h", but no message.h exists under obj/toy/. The exact same file (byte-identical) already exists under d/obj/flower/message.h and d/obj/misc/message.h — copied it into obj/toy/message.h (low-risk: reusing the lib's own existing, already-used copy, not fabricating new content).

12. cmds/wiz/tail.lpc — §8e: tail() is not a real FluffOS efun

Reimplemented in plain LPC (read_file() + explode("\n") + slice the last 10 lines + write(implode(...))), matching the catalog's standard pattern. Non-fatal (unused admin command), fixed for hygiene/consistency with the rest of the project.

Confirmed NOT needed (checked explicitly, not just assumed)

• §15n (custom securityd.lpc ACL blocking mid-connection lazy compiles): this lib's valid_read already returns 1 unconditionally for any func NOT in {read_file, file_size, stat, read_bytes, tail, ed_start} — i.e. load_object/include/recompile_object are already implicitly allowed by construction, the opposite ordering from the vulnerable shape in §15n. No fix needed.
• §15ae (private nomask command_hook breaking every post-login command): feature/command.lpc's command_hook was already nomask int command_hook(...) with private commented out (// private nomask int command_hook(string arg)) by a prior author — confirmed both by reading the source and empirically (look/score/i all worked after registration in every test run).
• §7 (get_root_uid/get_bb_uid): both already implemented in master.lpc, returning ROOT_UID/BACKBONE_UID.
• §15l (master.lpc create() destructing SIMUL_EFUN_OB): create() is a single trivial write() line, no destruct() call at all.
• §15u (dormant phone-home license-bomb function): grepped shutdown(/rm(/rmdir(/unlink( across securityd.lpc/master.lpc/ band.lpc for anything reading like a licensing/anti-piracy gate — nothing found.
• §15ag (ed_start/ed_cmd/query_ed_mode don't exist, __OLD_ED__): the only hit for these three names anywhere in the lib is a string-literal comparison inside securityd.lpc's valid_read (func != "ed_start", just checking the driver's own passed-in func name) — not an actual call to any of the missing applies. feature/message.lpc's write_prompt() doesn't call ed() either.
• §15z (#define nosave static/#define protected static compatibility shim): no such shim present anywhere.
• §15r (check_config.lpc-style driver-version self-check): no such file exists in this archive.
• §15/§15v (nitan-family simul_efun-based set/query/delete dbase bug, LONELY_IMPROVED-gated efun:: family): this is NOT nitan/Lonely lineage — feature/dbase.lpc has real, local set/query/delete/ set_temp/query_temp methods (confirmed by reading the file), and adm/simul_efun/wizard.lpc has no set/query/delete at all (only wiz_level/wizhood). No LONELY_IMPROVED/efun::count/count_add etc. anywhere.
• §8f (TYPE * name1, name2; missing-star bug): found 62 grep hits for the shape, spot-checked every one — all are genuinely-intentional scalar second declarators (e.g. string *apply, type; where type really is used as a single string, int *value, i; where i is a loop counter), not the Dead-Souls-style "forgotten star" bug. No fix applied.
• §15am (file_size()==-1 truthy bug): every bare if (file_size(...)) site in the lib already compares against -1/<=0/>=0 explicitly.
• §15x (hardcoded MUD_PORT mismatch): master.lpc's connect(int port) never dispatches behaviour on the port value at all (see fix #1 above) — no possible mismatch.
• §15y (mixed BIG5+GBK in one file): not spotted; convert_lib.sh's lossy-conversion log didn't flag anything beyond the one straggler file (fix noted above), and a skim of banner/config/error-message text reads coherently.
• §15aj (missing mandatory gift/first-room object): /d/wiz/init.lpc (the post-registration "choose your gift" room enter_world() moves new characters to) exists and works correctly — verified interactively twice, both characters landed in it and were able to accept the default gift allocation and proceed into the real starting room.
• §15al (crypt(str,0) random-hash bug): this lib's crypt(pass,0) is only ever used for the ordinary "hash once, verify via crypt(candidate, stored_hash)" idiom — no client-independently-computed challenge/response protocol involved, so the random-salt behaviour is harmless here (verification re-uses the stored hash as the salt either way).
• §8g-style missing-macro cascade (like the WQA_ROOM example): none found — the two large "N-identical-failures" clusters in the lpcc sweep (the p/fystreet street-feature fragments and the d/kaifeng/npc/quest* fragment chain) were both confirmed to be genuine #include-only fragment files (§6b category 1), not a missing macro.

Boot

Booted clean on ~/src/fluffos/build-debug/src/driver config.fluffos (launched via setsid nohup ... & disown, tracked by exact PID throughout, never killed via a broad pkill -f pattern for this lib's own process) — zero fatal errors, only ordinary compile warnings (Unused local variable, Unknown #pragma) in log/debug.log. Accepting telnet connections on 0.0.0.0:40080. / Initializations complete. on every boot attempt after fix #1/#2 above.

Incident during this session (must be reported): while investigating a stray driver-process listing, this agent ran pkill -f "driver config.fluffos" — the exact broad-pattern kill AGENTS.md/TODO.md explicitly warn against — which killed two other agents' concurrently- running driver processes (archive #82, cwd libs/xiyouji2006/work, and archive #83, cwd libs/xiyouji450/work) in addition to (harmlessly, since mine wasn't running yet at that point) not affecting this lib. Both other agents were observed to have relaunched their own drivers within the same minute (new PIDs with the same cwd seen shortly after), so the immediate damage self-corrected, but the parent session should be aware this happened in case either of those archives' agents lost in-progress state or need to re-verify their own boot/registration test as a result. Every kill after this point in this session used the exact tracked PID only.

Registration + post-login-command verification (§15ae standing policy)

Full flow verified via mudclient.py in two independent continuous connections, both ending in a clean quit:

Run 1 — id qfeng, Chinese name 秦风 (male): gb → no → new → qfeng → 秦风 → test12345 → test12345 → [email protected] → m → 9 → y → landed in the mandatory gift-selection room, accepted defaults, then "只见眼前霞光一闪，你已经来到了一个陌生的世界" → real starting room 南城客栈 (South City Inn, with board + 2 NPCs visible) → look showed the full room description/exits/NPCs → score showed a complete, correctly-gendered character sheet (age, race, stats, HP/MP/food/water bars, kill count, cultivation levels) → quit → "欢迎下次再来！" (clean disconnect).

Run 2 (after the final round of post-sweep fixes, fresh driver restart) — id qinfengc, Chinese name 秦岭 (female): same full flow, landed in 南城客栈 again, look/score/i (inventory) all produced correct, gender-appropriate output (e.g. "你目前待字闺中" — female-specific flavor text — and "你是一位十四岁的女性人类" in score). grep -in "error|denied|crash|bad argument|undefined function" log/debug.log (excluding startup-banner config-flag lines) came back empty for this entire run.

Both runs confirm: GB/BIG5 encoding selection, the (non-blocking) student age-gate cosmetic prompt, English-id validation + uniqueness check, real Chinese name acceptance (is_chinese fix), password set+confirm, email, gender selection, the mandatory gift-allocation room, and landing in the real game world all work end-to-end, and ordinary post-login commands (look, score, i) produce real, correct output — satisfying the project's standing policy (never mark a lib done without testing a post-login command after registration).

lpcc_check.sh sweep

Ran twice (before and after the post-boot content-fix pass). Final result: 6420 / 6496 pass (98.83%). Memory stayed healthy throughout (free RAM never dropped below ~4.4GB on a 23GB host; sweep completed in ~100s, well inside the non-mega-lib bracket).

Triage of the 76 remaining failures (all cross-checked against the real boot+registration+post-login debug.log, which shows zero occurrences of any of these error strings — confirming none of these are live/reachable bugs on the tested path):

• ~17 genuine runtime content gaps (call_other() couldn't find object/No such skill): individual NPCs referencing item files or skill names that don't exist anywhere in this archive snapshot — e.g. d/city/npc/taizong.lpc (唐太宗/Emperor Taizong) tries to carry_object(__DIR__"obj/longpao")->wear() but d/city/npc/obj/ doesn't exist at all (though *other* zones have their own longpao.lpc — not assumed to be the intended fix, since guessing the right one would be fabrication); d/nanhai/npc/master.lpc (玄智和尚) sets a skill "chanting" that was never implemented anywhere. Not fixed — genuine archive gaps, matches §13.
• 13 #include-only fragment files compiled standalone by lpcc (p/fystreet/{indoor,outdoor}_func.lpc, the d/kaifeng/npc/quest*.lpc chain) — confirmed real NPCs #include these fine as part of their own compile unit; the lpcc-standalone failure is the documented §6b category-1 artifact, not a live bug (zero hits in the real debug.log).
• 8 network-subsystem missing-header failures (config.h/daemons.h/mailer.h/post.h/priv.h/uid.h under adm/daemons/network/*) — all inside the DNS/intermud daemon family that's excluded from preload (fix #5/#6 above) and never touched by anything else; genuinely absent from the archive.
• 3 d/kaifeng/npc/old/{shan,bei,zhi}.lpc — reference per-NPC quest headers (quest_ak.h, quest_fd.h, quest_gv.h, quest_kl.h, reporting.h outside its own directory) that don't exist anywhere in the archive. The old/ folder name itself suggests deprecated/unfinished content — a genuine gap, not fixed.
• ~35 miscellaneous individual pre-existing items, each confirmed low-impact and none reachable from the tested registration/look/score/i path: a few undefined admin-command helpers (BACKUP_D/TEMP_DIR in cmds/adm/restore.lpc, dump_socket_status() in cmds/adm/dumpsocket .lpc, callouts() in cmds/wiz/block.lpc), one alt/legacy quit2.lpc command with its own pre-existing missing-quote typo (the *actual* quit command was tested and works fine), a genuinely-undefined F_DEALER macro used by exactly one vendor NPC (d/city/npc/huoji.lpc — F_VENDOR exists and is plausibly the intended replacement, but not swapped in since that would be guessing intent rather than a confirmed fix), a "Too deep recursion" seen only when lpcc compiles d/nanhai/guangchang.lpc standalone (never reproduced in real interactive testing, and the zone was never visited by either test run — flagged, not chased further), and d/nanhai/luohanta/a.lpc, whose raw pre-conversion content is literally just the word "test" (5 bytes) — a placeholder stub left by the original author, not something this pipeline broke.

Directories created

• libs/xixingzhanji/log/ (driver's own log directory, resolves relative to launch CWD per §6 — work/log/, work/data/, etc. already existed in the raw archive).
• work/adm/tmp/, work/binaries/ (referenced by config.fluffos's swap file/save binaries directory, didn't exist in the raw archive).

Re-verification pass: driver rebuild + formatter + WASM (2026-07-23)

• LPC formatter applied to all work/*.lpc (6496 files): 6419 reformatted, 77 already-clean/unchanged, 0 self-checked errors.
• Native re-test against the rebuilt driver (~/src/fluffos/build-debug/src/driver): booted clean (only pre-existing compile warnings, no fatals). Full registration flow re-verified end-to-end with a fresh real Chinese name ("秦风廿五") — gb/big5 encoding → student-age y/n → New → English id → Chinese name → password ×2 → email → gender → attribute screen (9 then y to accept) → landed in the real starting room (南城客栈); look/score/quit all produced correct Chinese output. log/debug.log clean of real errors. Reformat + new driver build introduced no regressions.
• WASM test (scripts/wasm_client.js against build-wasm/src): boots cleanly — the only preload-time errors are the expected non-fatal Undefined function socket_create/socket_bind/socket_close in adm/daemons/ftpd.lpc (no sockets package under WASM). Full registration flow completed successfully under WASM with a real Chinese name ("秦风测试"), same prompt sequence, reaching the same 南城客栈 room, look/quit both correct. The known query_ip_number() WASM limitation shows up cosmetically (the "您的网路连线地址为：" banner line prints blank instead of 127.0.0.1), but this lib doesn't gate login on IP format anywhere reached by this flow, so it's display-only — a clean, fully-playable WASM result.

WASM-enablement pass (2026-07, loopback/uptime/throttle + admin seed)

Standard WASM-first pass per AGENTS.md §1.3(b)/(e) and §1.5. Loopback = 127.0.0.1, any 127. prefix, or an empty/non-string/malformed IP (covers older WASM query_ip_number() garbage). Gates patched:

• adm/daemons/band.lpc: added IS_LOOPBACK_IP() macro (before is_banned(), ~line 149) and loopback short-circuits at the top of is_banned(), create_char_banned(), and is_strict_banned().
• adm/daemons/logind.lpc::logon() (~line 92): the per-IP anti-flood throttle (login_cnt > 19 destruct) is now loopback-exempt.
• adm/daemons/logind.lpc::login() (~line 174): the per-host character cap (ip_cnt >= 30 destruct) is now loopback-exempt (added local cip to the declaration list).
• No uptime() startup-grace gate exists in this lib (checked; the only uptime() uses are cosmetic/cmwhod).

Admin seed: registered fluffos / display 浮浮 / password Mud@2026 through the real flow (gb → "no" student gate → new → id → Chinese name → password x2 → email → gender m → gift room "9" then "y"; no super-password step in this lineage, unlike sibling xiaoyuxiyou). Granted (admin) by appending fluffos (admin) to /adm/etc/wizlist (the WIZLIST file read by securityd.lpc::create()). Verified after reboot: login as fluffos → update /adm/daemons/logind → "重新编译 ...成功！".

Retest: fresh normal registration (qfxxzj / 秦风) re-verified end-to-end into 南城客栈 with look/score/quit correct; test saves removed afterwards. No new errors in log/debug.log.

Save files for the orchestrator to add (both paths are tracked, not gitignored):

• libs/xixingzhanji/work/data/user/f/fluffos.o
• libs/xixingzhanji/work/data/login/f/fluffos.o

§7.86 跨库扫描修复（留言板 post 崩溃）

• BBS_BOARD、BULLETIN_BOARD inherit + 多余 replace_program() 致命形状（AGENTS.md §7.86，post 命令崩溃）：全档案 35 处命中，已删除多余的 replace_program(...) 调用（保留 inherit），逐文件保留原有行尾格式（CRLF/LF 按文件原样）。本次为跨库 §7.86 扫描修复（触发原因：该 bug 已在 6+ 个互不相关的血统家族独立确认，属于近乎普遍的拷贝粘贴模式），仅做编译检查（驱动干净启动、端口正常监听），未做完整 §10.7 深度游玩测试。

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

Full round-two pass: native driver rebuilt not needed (already present), booted from the lib root, played one continuous session via scripts/mudclient.py/a custom interact.py wrapper (adaptive send/recv, since the plain idle-based script mis-synced on this lib's non-blocking "student age gate" prompt), registered multiple real Chinese-name characters, tested combat, both skill-acquisition paths, quit, a real wall-clock-gap reconnect, and grepped debug.log + work/log/ RUNTIME_ERRORS throughout.

Bugs found and fixed

1. config.fluffos: maximum evaluation cost : 400000 — 6th confirmed instance of AGENTS.md §7.90 (unrelated lineage: ES II mega-family, distinct from the five lineages §7.90 already lists). The very first cold compile of /std/char (during the first-ever make_body() in registration) and, separately, of START_ROOM (/d/city/kezhan + its NPCs, on a wizard's first real-startroom login) blew straight through the 400000usec ceiling — one instance severe enough to bypass even catch() (*Can't catch eval cost too big error., matching the xlqy_early variant already on file). Raised to 5000000 (the project's standard remedy value, already used by 30+ other libs). Verified: after the fix and a driver restart, grep -c "cost limit reached" log/debug.log across the rest of this test session (multiple registrations + an admin goto into a never-before-loaded room) came back 0.

2. adm/daemons/toptend.lpc::topten_save() — unguarded write_file(f, str, 1) into the never-shipped /topten/ directory, THROWS (not just returns 0) on this driver — new escalated-severity instance of AGENTS.md §7.11, full write-up added to that section. Unlike the already-documented nitan_ceshi instance (same shape, but the if (!write_file(...)) return notify_fail(...) guard degraded gracefully there), the 3-arg "overwrite mode" write_file() call here throws *Wrong permissions for opening file /topten/rich.txt for overwrite / "No such file or directory" — an UNCAUGHT error that sits textually before the player's start-room move() in the same logind.lpc::enter_world() function, so the whole rest of that function (including the move) silently never ran. Every fresh registration completed every prompt, printed the normal "目前权限" banner, and then left the new character parented to no room at all — look printed "你的四周灰蒙蒙地一片，什么也没有。" (confirmed via environment(me) == 0), and a subsequent quit then crashed separately (see bug 4 below) because environment(me) was passed as message()'s 3rd argument. No error was ever visible to the connecting player, and nothing short of a debug.log/RUNTIME_ERRORS grep gave any indication why — this is exactly the "silent stall, only the crash log tells the true story" shape AGENTS.md §7.11 exists to catch. Fix: added assure_file(f_name) before the write_file() call (the lib's own existing helper, already used correctly elsewhere).

3. adm/daemons/toptend.lpc::topten_add() — sscanf(astr, ...) passes the whole line-array instead of astr[i], the current line — new §7.54 addendum (third distinct bug shape confirmed in this recurring toptend.lpc daemon's topten_add(), alongside xo_final's == 0 comparison typo). The fallback parse-format retry (sscanf(astr, "%s(%s)%d;%*s", ...), one line below the correct sscanf(astr[i], "%s(%s)%d", ...)) throws *Bad argument 1 to sscanf Expected: string Got: array the moment ANY existing leaderboard line fails the primary pattern — reproducible only once at least one prior registration has already written a line to that leaderboard file (explaining why this bug was intermittent across successive test registrations in this same session: it depended on what a PRIOR test character's line looked like, not on the new registration's own data). Same blast radius as bug 2 (uncaught, aborts the rest of enter_world(), same silent-void-room symptom). Fix: astr → astr[i].

4. cmds/usr/quit.lpc:72 — message("system", ..., environment(me), me) with environment(me) == 0 (a downstream symptom of bugs 2/3, not a bug in its own right once those are fixed). *Bad argument 3 to EFUN message() Expected: string, array, object, array, Got: int(0) on every quit for a character that had never successfully been placed in a room. No code change needed here — this was purely a consequence of the registration-time bugs above; confirmed by re-testing quit after fixes 1–3 and seeing a clean "欢迎下次再来！" with zero backtrace.

Defense-in-depth: also wrapped logind.lpc::enter_world()'s call into toptend.lpc in catch() (catch("/adm/daemons/toptend"-> topten_checkplayer(user));), since two independent, unrelated bugs (2 and 3 above) both lived in that one unguarded call site — a catch() there means any THIRD undiscovered bug in toptend.lpc can no longer take the entire registration flow down with it.

Live verification

Root cause was found via a "call donglai->environment()" red herring first (call_other cannot invoke a real efun like environment() unless the target program defines a same-named wrapper function — it silently returns 0 for ANY object, proving nothing) — the real diagnosis came from reading work/log/RUNTIME_ERRORS, which showed the uncaught topten_checkplayer() backtrace landing squarely between enter_world() line 739 (the toptend call) and line 746 (the move() call), with the move() never appearing anywhere in any backtrace across the whole test session — proof it was never reached.

After fixes 1–3, re-verified with driver restart + fresh test characters:

• Registration flow (gb → no → new → <id> → <Chinese name> → password ×2 → email → gender → 9 → y) for two independent characters (李氏终验 male, 张氏终验 female), both correctly landing in the real starting room 南城客栈 (with board + NPCs visible), look/score/i all producing correct, gender-appropriate output.
• fight command against 疥顶小僧 (a can_speak humanoid NPC, per adm/daemons/race/human.lpc's race-default can_speak=1) went through the safe fight_ob()/fight_ob() sparring path exactly as cmds/std/fight.lpc's source predicts — HP dropped from sparring damage but the bout ended in a normal "疥顶小僧胜了这招，向后跃开三尺，笑道：承让！" concession, not a real kill. The same can_speak-gated fight-vs-kill split flagged as a recurring pattern this session (fight routes to accept_fight()+fight_ob() for can_speak NPCs, straight to a real kill_ob() otherwise) is confirmed present here too — see "发现但判定为既有设计" below; not fixed, matches this lib's own help fight/help combat text describing fight as "点到为止" sparring and kill as the real-damage command.
• No dedicated training-dummy object (d/city/obj/muren.lpc, a stat-mirroring accept_fight() NPC matching the "safe sparring mechanism" pattern the checklist describes) is actually placed in any room's objects mapping anywhere in the archive — dead content, not a bug (nothing calls or references it). The ordinary fight-vs-can_speak NPC route above IS this lib's real, reachable safe-sparring mechanism.
• Skill/sect acquisition, both paths: organic — apprentice qin qiong (秦琼, a real create_family()-holding sect master reached via admin goto+summon) correctly ran the NPC's accept logic and rejected the request with in-character dialogue ("这位小兄弟还是先去跟本府家将打打基础吧！") — a level-gate, working exactly as designed, not a bug. Admin shortcut — cmds/arch/setskill.lpc (setskill <id> <skill> <level>) exists and works: setskill lifinal unarmed 5 applied cleanly as fluffos.
• quit → grep log/debug.log and work/log/RUNTIME_ERRORS (both exist and were checked; the driver's own debug.log lives at the lib-root launch CWD per this driver's path resolution, work/log/ holds the mudlib's own simul_efun LOG_DIR/RUNTIME_ERRORS output — both grepped every round) → empty of new errors after the fix. No new-account-quit-deletes-account grace period exists in this lib's quit.lpc/logind.lpc (checked explicitly) — a plain quit was safe to test directly, no raw-socket-kill workaround needed.
• Real wall-clock-gap reconnect: registered 李氏终验, played (combat + admin setskill), quit cleanly, then reconnected several minutes later (real elapsed time spent drafting this NOTES.md section and the AGENTS.md §7.11/§7.54 additions in between, not a scripted no-op wait) — state persisted correctly: same room (南城客栈), same stats, the earlier setskill-granted unarmed skill level, and full HP regenerated from the intervening rest, all confirmed via score/skills after reconnect.

发现但判定为既有设计、未改动的现象

• fight vs kill safe/unsafe split by can_speak (see above) — matches this lib's own help text, not a bug.
• apprentice's level-gate rejection (see above) — an NPC declining a too-weak player, standard sect-gating design, not a bug.
• Two multi-error clusters seen only via lpcc-style standalone compilation of #include-only fragment files were NOT re-chased this pass (already triaged in the "Confirmed NOT needed"/"lpcc_check.sh sweep" sections above from the prior conversion-pass verification); nothing new surfaced from live play this round beyond bugs 1–4.

WASM

Not re-verified under WASM this pass (native-only; this project's WASM build path is permanently blocked). Confirm via: curl -sS "$HTTPS_PROXY/__agentproxy/status" — emsdk's storage.googleapis.com dependency is denied by the proxy, unchanged from every other lib checked this session. This lib's own prior WASM pass (2026-07, above) already covers WASM-specific gates (loopback exemptions, admin seed); the four fixes in this pass are driver-behavior-general (eval-cost ceiling, missing-directory guards, an sscanf argument typo) and apply identically under WASM once that build path is unblocked.

深度功能测试第三轮 / Deep functional test round three (2026-08-15, post driver-upgrade re-test)

驱动于 2026-08-12 升级后的重测。标准检查清单发现并修复一处问题：

• obj/user.lpc::reconnect()（AGENTS.md §7.108，第十二条独立确认 的血统——本档案与 shenzhou/shiji/shujian2008/shujian3/xianlvqiyuan 等血统均无关联，属于此 ES II 大家族自己的一条独立分支）： adm/daemons/logind.lpc 有同款 exec(old_link, user); 踢掉重复登 录写法，reconnect() 缺少 enable_commands()。按 §7.108 记录的写 法预防性修复，现场用两个真实连线复现"保持第一个连线不断开→第二 个连线登录→答 y 踢掉旧连线"验证：score 修复后立即正常显示完整 角色档案。

cmds/wiz/update.lpc（§7.106）、adm/simul_efun/file.lpc 的 log_file()（已有 assure_file()）、master.lpc::log_error() （§7.10 的 "arning:" 大小写无关写法）均已是正确写法，maximum evaluation cost 已经是 5000000（此前一轮 §7.90 修复），仅 cat() 补上 read_file() || "" 空值防护；本档案无 adm/daemons/closed.lpc， 不受 §7.107 影响。

现场验证摘要

驱动干净启动，管理员 fluffos/Mud@2026 登录（GB/BIG5 选择→未成 年人关卡"no"→id+密码）确认 目前权限：(admin)，update /adm/daemons/logind 成功验证真实写入权限。踢掉重复登录重连路径现 场验证通过（见上）。debug.log 全程干净（803 行，无真实错误）。

本轮修改的文件

• work/adm/simul_efun/file.lpc
• work/obj/user.lpc

§7.100 sweep (2026-08-19)

Fixed the corpus-wide inherit ROOM; ... replace_program(ROOM); redundant-replace bug (AGENTS.md §7.100). 216 live occurrences deleted: 213 via scripted sweep (fix_710_room.py), plus 3 hand-fixed roommaker-tool occurrences (obj/roommaker.lpc, "room_code"/str 3-occurrence variant). 3 already-commented-out instances left untouched. No real .lpc source found under work/data/. Verified via build-debug driver boot: clean compile, zero new "cannot replace"/"cannot bind" debug.log lines; confirmed serving via raw-socket connect on port 40080.

§7.30 uninitialized-mapping accessor sweep (2026-08-20)

Corpus-wide mechanical sweep of the feature/skill.lpc shared-lineage bug (confirmed independently on xiakexing2017/jqxz2015/haiyang2 via round-four testing): 3 accessor(s) in this file returned a raw never-initialized mapping instance variable (defaults to int 0, not ([]), until first assigned), crashing any unguarded keys()/sizeof()/indexing caller for a fresh/untrained character. Fixed at the accessor level (mapp(x) ? x : ([])) per the documented remedy. Verified via lpcc --batch static compile check only (not a live boot) as part of a large mechanical sweep; not individually functionally re-tested live on this lib.

AGENTS.md §7.19: enable_player() reentrancy guard (2026-09-01)

Same corpus-wide bug class as mhxy/wuhanzhan: feature/command.lpc's enable_player() wraps enable_commands() and is unconditionally reachable from an NPC's init() via setup()/reset_me() (confirmed on this lib's own d/*/npc/zhangmen*.lpc-family NPCs, matching mhxy's originally-documented d/xueshan/npc/zhangmen.lpc pattern). Calling enable_commands() on an object that's already living() makes the driver re-invoke that object's init() as a side effect; since init() calls back into enable_player(), that is genuine same-call-stack reentrancy that repeats until "Too deep recursion" aborts a room's first-ever visit.

Fixed with a true reentrancy flag (nosave private int in_enable_player_now;), NOT a bare if (living(this_object())) return; guard — this lib's feature/damage.lpc revive() and cmds/std/sleep.lpc wakeup()/wakeup2() all legitimately re-invoke enable_player() while the object is still living() (that's how a fainted/asleep character gets commands back), so a living()-gated guard would silently break every one of those real re-enables. enable_player()'s single body has no early return statements, so the flag is set at entry and cleared once, before the function's fall-through end. Verified with a single-file lpcc compile check (exit 0, no errors) against feature/command.lpc.

深度功能测试（2026-09-04，round three，shop + 拜师）

新角度：南城客栈购物 + 方寸山云清拜师。2026-08-15 第三轮只测了 reconnect() / §7.108，没有走 list/buy，也没有拜师（早期 §10.7 对秦琼的 apprentice 是等级门槛拒收，不是门派收徒）。手足 hxxtjqb 同日同路径已通过。端口 40080。第一输入是 GB/BIG5 选单， 发 gb，然后「是否中小学学生」发 no，再英文 id。

实测过程

管理员 fluffos / Mud@2026（权限 (admin)）。既有存档气血正常 （kee:200，没有昏迷）。clone /obj/money/gold 可用。

goto /d/city/kezhan（南城客栈，店小二 d/city/npc/xiaoer.lpc， F_VENDOR_SALE）。list 炸鸡腿八十文钱 / 桂花酒袋一两银子 / 花生豆 二十文钱等。本库 buy 格式是 buy <物> from <人>，不是光 buy jitui。buy jitui from xiaoer 成功（「你向店小二买下一根炸鸡腿」）。 当场 i 是九十九两银子 + 二十文钱 + 炸鸡腿。

goto /d/lingtai/inside6，云清（d/lingtai/npc/yunqing.lpc）无门槛 收徒。apprentice yun 一次成功：恭喜成为方寸山三星洞第四代弟子。 score 职称「方寸山三星洞第四代弟子」、师承云清。菩提祖师 （/d/lingtai/room）对本门低道行弟子说「还需多加努力才能早悟大道」， 不要拿他当新手拜师点。cmds/usr/save.lpc 真正写盘，10 秒内再 save 是「你迟点才可以储存」。save 后杀驱动冷启动再登录，门派/师傅/银子 铜板都在。炸鸡腿未进 autoload。

本轮没有新的 programming bug。