bixiecanyang — 碧血残阳 之「豪侠晚歌」

Archive: 碧血残阳之豪侠晚歌.rar (archive #71). Port: 40065. Status: done (boots clean, full registration flow verified end-to-end including a real Chinese name reaching the actual game world).

What this is

"碧血残阳" ("Blood on the Setting Sun"), subtitle "「豪侠晚歌」" ("Ballad of the Gallant Hero"), domain akai.isme.net, dual-port config (ZMUD 6666 / TOMUD 8888 in the raw config; reassigned to a single port 40065 here). adm/obj/{master,simul_efun} layout. The bundled 碧血残阳说明.txt (author's own readme) explicitly states this lib is built on top of "夕阳再现－江湖风云2－风云再起" ("Sunset Reappears - Jianghu Fengyun 2

• Storm Rises Again") — i.e. this is a member of the same **夕阳再现

(Xiyangzaixian)** lineage as archives #46 (xyzxfk), #47 (xyzxfy2), #55 (wmkj), and #59 (jhfy), not a standalone codebase.

Lineage confirmed via md5sum, not just the readme's claim:

• adm/simul_efun/chinese.c: byte-identical (md5 961d77af057bb93db320af05be5883fc) across all 5 archives (#46/#47/#55/#59 and this one).
• adm/daemons/securityd.c: byte-identical (md5 032452f57e844ff416d4747b12b85d27) specifically to xyzxfy2 (#47)'s copy — closest sibling in this family.
• adm/obj/master.c and adm/daemons/logind.c are each distinct from every sibling (own md5sums, different line counts) — same pattern as every other pair of forks in this family: shared engine utility files, divergent site-specific daemons.

Cross-check against dtsl2 (#49) — required by task instructions

libs/dtsl2/NOTES.md documents that its raw config file's own name field contained the exact string "碧血残阳 之 豪侠晚歌" — this archive's own title — as a confirmed-stale copy-paste leftover, not a sign that dtsl2 is secretly a duplicate of this archive. Verified the reverse direction too, as instructed, via direct md5sum/diff on both raw archives (not work copies):

• adm/simul_efun/chinese.c: differs — this archive's md5 is 961d77af... (the 夕阳再现-family shape: strlen(str)>=2 && str[0] > 160 && str[0] < 255, 14 lines); dtsl2's is a619b926... (a different GBK-range formula: str[0] < 176 || str[0] > 247 + str[1] checks, 19 lines). Not the same file at all.
• adm/obj/master.c: differs substantially — this archive's is 342 lines with a TOMUD_PORT/get_player_fname() backup-login-object fallback mechanism and a "change by Fan@wanxia" credit header, none of which appear in dtsl2's (302-line, no TOMUD/backup-login logic, no Fan@wanxia credit) — full diff is 646 lines of changes out of ~342, i.e. almost entirely different content past the shared connect()/compile_object() skeleton wording.
• adm/daemons/logind.c: 1007 lines here vs 917 in dtsl2 — different.
• Interesting second piece of evidence for the copy-paste theory: dtsl2's raw archive's own config file is also literally named config.bxcy (verified: libs/dtsl2/raw/DTSL/config.bxcy exists) — the same filename as this archive's config.bxcy. This means not just the name field's *text* but the config file's *filename convention itself* was copy-pasted from a bixiecanyang-lineage template into whatever base dtsl2 forked from, while the actual mudlib code (chinese.c/master.c/logind.c) diverged completely.

Conclusion: confirmed NOT related in either direction. This archive is squarely 夕阳再现/Xiyangzaixian-lineage (shares chinese.c/securityd.c with #46/#47/#55/#59); dtsl2 is squarely dtsl(#8)/大唐双龙-lineage (shares chinese.c/F_UNIQUE-macro-gap/diamond-inherit-bug with #8/#18). The coincidence is purely a copy-pasted config template/field, confirmed twice over (stale name field + stale filename convention), never a sign of shared mudlib content.

Fixes applied (ported/cross-checked against #47's proven fix list where lineage matches, applied proactively before first boot)

1. AGENTS.md §15h (adm/simul_efun/chinese.lpc's is_chinese()): GBK lead-byte range check (strlen(str)>=2 && str[0] > 160 && str[0] < 255) → CJK Unicode codepoint range check (strlen(str)>=1 && str[0] >= 0x4e00 && str[0] <= 0x9fff). 2. AGENTS.md §15h (adm/daemons/logind.lpc's check_legal_name()): byte-count bound strlen(name) < 2 || > 10 → character-count bound < 1 || > 5 (matches the message's own stated "1 到 5 个中文字"); the sliding-window gate i%2==0 && !is_chinese(name[i..<0]) had its i%2==0 && dropped so every character position is checked (the name[i..<0] slice itself needed no change since is_chinese() only ever inspects index 0 of whatever's passed to it — a full-length suffix slice starting at i is functionally equivalent to a single-character check either way). 3. AGENTS.md §15z (5 files: include/globals.h, adm/obj/master.lpc, adm/obj/simul_efun.lpc, adm/obj/simul_efun_bak.lpc, u/isle/ToMud/globals.h): each carried an #ifndef __SENSIBLE_MODIFIERS__ / #define nosave static / #define protected static compatibility shim (this driver never defines __SENSIBLE_MODIFIERS__, so the shim is always active). convert_lib.sh's blanket static→nosave sed rewrote the shim's own lines too, turning #define protected static into #define protected nosave — silently aliasing protected to nosave instead of leaving it as its own distinct, correctly-behaving keyword. Fixed to #define protected protected (self-map, matching the already-correct #define nosave nosave sibling line) in all 5 files. 4. AGENTS.md §3 counterexample (22 files): the same blanket sed also hit log_file("static/XXX", ...) string literals (a real on-disk log subdirectory name, unrelated to the static/nosave keyword), rewriting them to "nosave/XXX" and orphaning the archive's real seed data at log/static/ (confirmed present on disk: archive, banned_sites, CALL_PLAYER, CRASHES, EAT_DAN, etc.). Reverted all 22 occurrences (clone/misc/dan.lpc, cmds/{arch,adm}/{call,purge,setskill}.lpc, cmds/wiz/call.lpc, cmds/std/suicide.lpc, adm/simul_efun/override.lpc ×5, adm/daemons/securityd.lpc, adm/obj/master.lpc ×5) back to "static/XXX". 5. AGENTS.md §14 (adm/obj/master.lpc's valid_override): upgraded from the 2-arg signature to the documented 3-arg (file, name, main_file) form, checking main_file == SIMUL_EFUN_OB || main_file == MASTER_OB in addition to the original file check — this lib's simul_efun.lpc is a pure #include composer (pulls in object.lpc/message.lpc/wizard.lpc/etc., several of which contain efun:: overrides like object.lpc's destruct()), so file alone (the physically-containing fragment) never equals SIMUL_EFUN_OB for those. No live boot/registration failure was observed from the 2-arg gap (consistent with AGENTS.md's note that this may never surface on a real boot, only under isolated lpcc checks) but applied as free insurance per the catalog's standing guidance. 6. AGENTS.md §15p: /adm/daemons/network/dns_master was present (live, uncommented) in adm/etc/preload — commented out proactively before the first boot attempt. Checked every other preload entry's own source (securityd, virtuald, logind, rankd, preload, commandd, chinesed, emoted, aliasd, fingerd, autosaved, channeld, natured, monitord, storyd) for socket_create/resolve() calls — none found, confirmed safe to leave active. httpd was already commented out in the shipped preload file. 7. AGENTS.md §15s (adm/simul_efun/message.lpc's tell_room()): the common 2-arg call form (tell_room(ob, str), confirmed via grep to be the overwhelmingly dominant call shape across ~155 files in this archive) leaves the varargs object *exclude parameter as its default int 0, which then flows straight into message("tell_room", str, ob, exclude) — this driver's message() efun declares its 4th argument void | object | object * (a real type union, not "any"), so a literal int 0 is rejected at runtime (Bad argument 4 to message()) the first time any room's heartbeat/emote broadcast calls the wrapper with only 2 args. Fixed at the single shared root: message("tell_room", str, ob, exclude || ({}));. (File is CRLF-terminated — used a byte-exact Python replace rather than a line-oriented sed/Edit to avoid the line-ending pitfalls documented elsewhere in AGENTS.md.)

Confirmed NOT needed (verified by reading source, not by hitting a crash)

• §4 (lazy security-daemon load_object recursion in valid_read/valid_write): master.lpc's valid_write only ever find_object(SECURITY_D) (returns 0 if absent, never lazily loads it); valid_read does the same plus an early return 1 for callers with no resolvable previous_object(). No load_object() call anywhere in either apply — no reentrancy guard needed.
• §7 (get_root_uid/get_bb_uid): both present and correct (return ROOT_UID; / return BACKBONE_UID;).
• §15l (destruct(SIMUL_EFUN_OB) in master's create()): create() is a one-line write("master: 成功加载.\n"); — no destruct/reload dance present at all.
• §15n (custom securityd.lpc ACL denying "(player)"-status reads under /adm//cmds, blocking mid-connection lazy compiles): this lib's securityd.lpc (byte-identical to #47's, confirmed above) IS a genuinely custom ACL with real exclude_read/trusted_read tables, but — read directly, not assumed from the sibling's writeup — exclude_read never lists "(player)" in any of its directory entries (adm/d/data/userdata/feature/cmds/u/clone/include/ inherit/kungfu/quest/log all key off wizard-tier statuses only), and valid_read itself returns 1 early whenever the caller has no resolvable euid/status (if( !euid ) return 1;) and again whenever the file's author matches root (author_file(...)==ROOT_UID). The §15n failure mode genuinely cannot trigger here. Confirmed further by the clean interactive test — zero "Read access denied" anywhere in debug.log.
• §15u (dormant "phone-home license check" self-destruct function): grepped securityd.lpc/master.lpc for checking_status/mass deletion/shutdown()-gated-on-opaque-check shapes — none found.
• §8d/§15o (get_include_path() / local <header.h>-next-to-its-user angle-bracket includes): convert_lib.sh's automatic local-include-to- quotes pass already converted 57 such includes to quoted form during conversion (quotes resolve regardless of VM context, unconditionally fixing this whole class) — no master.lpc get_include_path() addition needed on top of that.
• Nitan-family simul_efun-dbase architecture bug (§15): not present. adm/obj/simul_efun.lpc is a pure #include composer with no bare set/query/delete definitions anywhere in its included fragments; feature/dbase.lpc (F_DBASE, widely inherited) implements real local set/query/delete methods directly. Correct inherit-based architecture.
• No named.lpc daemon exists in this lib at all (grep came up empty) — the deep §15h PATH()/sliding-window fix doesn't apply; chinesed.lpc only handles chinese_number()/dictionary translation, not name validation.
• §8e (tail() missing-efun bug): adm/simul_efun/file.c has no tail() call anywhere.
• §15x (hardcoded MUD_PORT mismatch silently rejecting every connection): master.lpc's connect(int port) does have a hardcoded #define TOMUD_PORT 8888 constant, but it's only used to conditionally set an optional "tomud" temp flag (if (port == TOMUD_PORT) login_ob->set_temp("tomud", 1);) — it never rejects a connection on a port mismatch. Since this project's assigned port (40065) never equals 8888, the flag is simply never set; confirmed no rejection occurs via the successful end-to-end interactive test.
• A related §15q-shaped hidden gate DOES exist but is provably inert here: logind.lpc's logon() has if (ob->query_temp("tomud")) { ... input_to("get_version", ob, ip_cnt); } — get_version() checks the input against a hardcoded literal "2060" (a TOMUD client-handshake code) and disconnects on any mismatch, exactly like the gate documented in §15q on xyzx3. But this branch is itself gated behind the same TOMUD_PORT-only flag described above, which (per the point just above) never gets set on this project's assigned port — so the gate is present in the source but structurally unreachable for our single-port setup. Confirmed by the transcript: no "2060" prompt appeared at any point, get_id was reached directly after the banner as expected.

Interactive test result — full registration flow

Read logind.lpc's actual input_to callback chain before testing (not assumed from a sibling lib). Flow: logon() → (no get_version, per above) → English id (get_id, 3-14 bytes / a-z only) → y/n create-confirmation (confirm_id) → Chinese name (get_name, 1-5 characters, NO separate y/n confirmation for a typed name — the y/n loop in get_resp only fires for the auto-generated random_name() path, triggered by submitting an empty name) → 管理密码/wizpwd (new_wizpwd, ≥5 chars) → confirm wizpwd (confirm_wizpwd) → real login password (new_password, ≥5 chars) → confirm password (confirm_password) → attribute-gift selection (select_gift, 0-4; 0 = system random via get_gift's y/n reroll loop) → email (get_email, [email protected] format required) → gender (get_gender, m/f) → enter_world().

banned_name list: 你/你的/我/我的/自己/屁/他/他的/她/她的/它/它的/鬼王/ 风云/流氓/爸/妈/祖先/祖宗/爷/奶/毛泽东/邓小平/江泽民/【/】/马克思/恩格斯/ 列宁/抄/手淫/老百姓 — checked before testing, "秦风" doesn't match any entry (note "风云" alone is banned, but "秦风" isn't a substring match for it in either direction as tested — the real driver accepted it cleanly).

Verified the complete registration path in one continuous mudclient.py connection:

• "" → correctly rejected ("必须是 3 到 10 个英文字母"), confirming check_legal_id is live and the empty-input retry loop works.
• qinfeng → accepted, "您确定吗(y/n)？" prompt.
• y → confirms new character, shows the "give yourself a name" banner.
• 秦风 (real Chinese name) → accepted silently, no rejection, proceeds straight to "请设定您的管理密码：" — confirms the §15h fix works correctly end-to-end, not just at the prompt-rendering level.
• pass1234 / pass1234 (wizpwd + confirm) → accepted.
• pass5678 / pass5678 (real password + confirm) → accepted, shows the attribute-gift explanation banner (膂力/悟性/根骨/身法/福缘/容貌).
• 0 → system-random gift roll (rolled 膂力18/悟性22/根骨22/身法18).
• y → accepts the rolled gift.
• [email protected] → accepted as email.
• m → male gender.
• Character actually entered the game world, landing in /d/city/wumiao ("武庙"/Wumiao — Yue Fei Temple, one of the 4 configured start_rooms), saw NPCs render correctly (紫衣神女小玲儿, 千里眼, 阿凯哥哥) and the bulletin board object. CHANNEL_D broadcast "听说又来了一位叫做秦风的 少年侠士" (new-player channel announcement) with the correct Chinese name embedded, confirming the character record itself carries the real name, not just the prompt accepting it.
• look → re-displayed the room correctly.
• quit → dropped a starting item ("布衣") and disconnected gracefully with "欢迎下次再来！" ("welcome back next time").

Zero real error: lines, zero "Read access denied"/"Bad argument"/"Too deep recursion"/segfault signatures anywhere in debug.log across the whole session — only the same cosmetic 编译时段错误：... warning: Unknown #pragma, ignored lines seen in xyzxfy2 (this lib's own error_handler()/log_error() mislabels compile-time *warnings* as "compile-time errors" in its Chinese message text; harmless, matches every other lib in this lineage family).

Second independent registration-flow verification (different session, different name/gender)

Rebooted the driver fresh (new PID) after the lpcc-sweep-driven fixes below (several of which touched adm/simul_efun/message.lpc, a preload-critical file) to make sure nothing regressed. Ran a second, independent mudclient.py test with a different id/Chinese name/gender combination in one continuous connection:

• id linfeng → confirm y → real Chinese name 林风 → accepted silently, straight to wizpwd prompt.
• wizpass12/wizpass12 (wizpwd+confirm) → accepted.
• realpass34/realpass34 (password+confirm) → accepted.
• 0 → random gift roll, y → accepted.
• [email protected] → accepted.
• f → female gender this time (previous test used male) → correctly set "gender" to 女性.
• Character entered the game world in a different start room this time, /d/xingxiu/beijiang ("北疆小镇"/Beijiang Town — one of the same 4 configured start_rooms, picked by the lib's own random selection logic), saw 2 "维吾尔族妇女" (Uyghur woman) NPCs, one of which reacted interactively ("维吾尔族妇女看着你笑了起来" / "the Uyghur woman looks at you and smiles"). CHANNEL_D broadcast "听说又来了一位叫做林风的妙龄 少女" (note: correctly used the female-appropriate honorific "妙龄少女" /"young maiden", not the male "少年侠士" from the first test — confirms gender-conditional text is also working correctly).
• look → re-displayed the room.
• quit → dropped a starting item ("粉红绸衫") and disconnected gracefully.

Zero real error:/"Read access denied"/"Bad argument"/"Too deep recursion"/"Undefined function" lines in debug.log for this second run either — confirms the message.lpc/simul_efun.lpc changes made during the lpcc-sweep triage (below) didn't regress anything.

lpcc sweep

13,560 files initially (.lpc/.c count after rename — matching xyzxfy2's file count almost exactly, same-generation snapshot of a similar world size); 13,559 after reclassifying 2 non-LPC files (below).

First full sweep (after only the proactive pre-boot fixes above, before any sweep-driven triage): 13,355 / 13,560 pass (98.5%), 205 failures.

Triaged the 205 failures and fixed every genuine driver-compat/typo bug found (leaving only true content-gap/context-dependent noise, per AGENTS.md §6b's "verify against the real boot log before trusting an lpcc-only failure" guidance and its "fix genuine one-off typos, don't chase call_other/couldn't-find-object noise" triage method):

1. 3× missing-opening-quote typos (d/quanzhen_old/hudi4.lpc, hudi6.lpc) and 1× missing-both-quotes typo (hudi5.lpc) in set("long", <bare Chinese text>) calls — same §10 shape documented project-wide. Confirmed pre-existing in the raw archive (not a conversion artifact) via direct byte inspection. 2. 1× missing-closing-quote typo (kungfu/skill/qingyi-jian.lpc): "skill_name" : "大悲式, → "skill_name" : "大悲式", (file has unusual \r\r\r\n line endings — fixed via exact byte-level Python replace, not a line-oriented sed/Edit). 3. 1× dropped-closing-quote typo mid-string-concat (kungfu/skill/kunlun-zhang.lpc): a Chinese sentence ending in the opening bracket "「" with no closing quote before the following comma — added the missing " (left the "「" as-is rather than guess at the sentence's real missing continuation, per AGENTS.md's "don't fabricate content" principle). 4. 1× missing-semicolon typo (kungfu/skill/huang.lpc): set_skill("yuxiao-jian", 550) (no ;) merged into the next statement, causing "unexpected map_skill". Now compiles clean but still shows "Fail to load object" for an unrelated context-dependent reason (NPC create() logic) — moved from a real compile bug to the same benign "Fail to load object" bucket as everything else below. 5. 1× stray leading character (kungfu/skill/xixi-unarmed.lpc): a bare c on line 1 (before the file's own header comment) made inherit SKILL; a few lines later illegal ("inherit after defining global variables"-shaped error, AGENTS.md §15t#3's failure mode from a different root cause). Removed the stray character. 6. 1× broken for loop + 1× same-file duplicate function (clone/misc/objmaker.lpc): for(i=0; i if(...) continue; was missing its condition/increment clause and opening brace (confirmed pre-existing in the raw archive) — restored to for(i=0; i<sizeof(name1); i++) { matching this codebase's universal idiom; separately, a duplicate int query_autoload() { return 1; } later in the same file (harmless copy-paste, same shape as objmaker's two query_autoload originally, see #9 below) was removed. 7. 1× missing comma between mapping-array elements (quest/murong/qlist3.lpc): a ]) closing one quest entry wasn't followed by a comma before the next ([ entry. 8. 1× stray backtick typo (kungfu/skill/hanbing-mianzhang2.lpc): ` random(...) ` — a leading backtick before a random() call (confirmed pre-existing in the raw archive) caused "End of file in template literal" since the lexer treated it as an unterminated literal. Removed the stray backtick. 9. Same-file/base-class query_autoload() duplicate-definition cluster (5 files: d/obj/quest/{chai,chai1,chai3}.lpc, clone/book/{lbook5,pixie_book}.lpc): chai.lpc/chai1.lpc/ chai3.lpc each #include "task.h" (a per-directory local header that ALREADY defines query_autoload()) and then redundantly redefined the identical function themselves; lbook5.lpc/pixie_book.lpc each had the exact same one-liner defined twice within their own file. Removed the redundant copy in each of the 5. 10. 1× duplicate create() function (d/npc/m_weapon/weapon/m_bian.lpc): a full, content-rich create() (name/weight/messages/init_whip) was followed later in the same file by a second, generic stub create() (missing all the specific weapon details) — clearly a copy-paste template leftover. Removed the redundant stub, kept the real one. 11. 2× lossy-conversion-shaped "End of file in text block" (d/quanzhen_old/manglin1.lpc, manglin2.lpc): the @LONG ... LONG text block's closing LONG tag was merged onto the end of the preceding text line with no newline — confirmed via the raw pre-conversion archive bytes that this is a PRE-EXISTING authoring typo, not an iconv/lossy-conversion artifact (unlike the similar- looking cases documented elsewhere in AGENTS.md) — the original GBK bytes already had "...耳际。LONG\r\n" with zero separator. Inserted the missing newline before LONG in both files. 12. message_combatd() — a 3rd message_vision-family alias gap (called live by kungfu/skill/yunv-xinfa/roar.lpc, 2 more call sites commented out elsewhere): never defined anywhere in this archive, same "§15b-style called-everywhere-defined-nowhere" gap already fixed on xzyx/wmkj for the identical function name. Added as a thin alias to message_vision() in adm/simul_efun/message.lpc. Hit the exact §15aa/§8b trap while adding it: an earlier attempt placed the alias BEFORE message_vision's own textual definition in the same file, which made adm/obj/simul_efun.lpc itself fail to load ("Undefined function message_vision") — moved the alias to AFTER message_vision's closing brace, which fixed it immediately. Re-ran the FULL boot + registration test after this specific change (since it touches a preload-critical file) to confirm no regression — see below. 13. 2 non-LPC files caught by the blind .c→.lpc rename (AGENTS.md §12): doc/mudlib/efuns.lpc (a plain columnar list of efun names — pure documentation, not compilable code, nothing references it) and d/huashan/map.lpc (a pure ASCII-art room-layout map, not a single line of LPC, nothing references it either). Both renamed to .txt to keep the sweep's pass/fail signal meaningful. 14. 1 genuine typo, unrelated context (doc/mudlib/efuns/shiwu.lpc, an example/tutorial room under doc/): call_out(do_close,0); — do_close used as a bare unquoted identifier instead of the string literal call_out() requires ("do_close"`). Fixed; the file now compiles (still shows "Fail to load object" for the same context-dependent reason as ordinary room files below, not a compile bug anymore).

Second full sweep (after all fixes above, and after reclassifying the 2 non-LPC files so the total file count is 13,559): 13,374 / 13,559 pass (98.6%).

Remaining ~185 failures, triaged by category, confirmed as expected noise (not fixed further):

• ~49× generic "Fail to load object" with no compile-error text at all (d/migong/lev*/dong* — 35 deep-dungeon-maze rooms; clone/gem/ring* — 11 loot rings whose setup()/init_sockets() chain calls this_player(), which is null in lpcc's bare find_object() load context; several individual kungfu/skill/*, kungfu/class/*, clone/npc/*, clone/misc/*, quest/* items/NPCs). Confirmed via the real boot+interactive-test debug.log that NONE of these paths are ever touched (grep -i "migong\|clone/gem" on the log came up empty) — none are on the preload/registration path, matching the exact category documented in xyzxfy2's own NOTES.md for the same shape of failure. Standalone lpcc re-checks of a sample (e.g. /clone/gem/ring4, whose inherit/gems/ring.lpc base class itself compiles clean standalone) confirm this is a bare-test-context limitation (no real player/room graph), not a genuine bug.
• ~8× /u/snow/wudujiao/... Cannot #include globals.h: the exact same personal-wizard-scratch-directory VM-context-dependent include gap already documented (and left unfixed, for the same reason) in xyzxfy2's own NOTES.md — this world content is shared between the two forks. Not on the registration/preload path.
• feature/vendor_sale.lpc: significantly corrupted in the RAW archive itself (confirmed via direct byte comparison, not a conversion artifact) — dozens of missing semicolons/parens and what looks like a fragment of a DIFFERENT function's if clause spliced into the middle of a string literal, spanning ~60 lines. This goes well beyond a single-line typo fix; reconstructing the intended shop-transaction logic would mean fabricating business logic, not restoring a known-correct original — left as a documented, unfixed content- corruption gap (matches AGENTS.md §13's "missing/broken content is a real archive gap" philosophy). Confirmed unreachable: nothing in the tree inherits or otherwise references this file at all (dead code as far as this archive goes).
• d/player/ltsh/npc/saveme.lpc: exert_function(10) passes an int literal where every other ~60 call sites in this archive pass a function-name STRING (e.g. exert_function("recover"), (: exert_function, "recover" :)), per inherit/char/npc.lpc's int exert_function(string func) signature. Genuine isolated typo, but the "correct" intended string can't be recovered without guessing — a single obscure healer-NPC's chat_msg config, not on any critical path. Left as a documented, unfixed typo.
• clone/weapon/panguanbi.lpc: inherit PEN; — PEN is never #defined anywhere (include/weapon.h defines every other weapon-type macro — AXE/BLADE/DAGGER/FORK/HAMMER/SWORD/STAFF/CLUB/THROWING/WHIP/RING — but never PEN), and no /inherit/weapon/pen file exists either. Exact same shape as the already-documented CLUB/WEAPON_COMBO/ BBS_BOARD gaps on dtsl2(#49) — a missing macro with no clear matching target file to point it at, so guessing a mapping (e.g. to STICK) would be fabricating behavior rather than restoring a documented gap. Left unfixed, single file affected.
• 3× d/heimuya/{shenggu,tang}.lpc, d/heimuya/npc/tang.lpc: "End of file in text block" — all 3 were also flagged LOSSY conversion during the encoding pass, matching the genuine iconv-eaten-newline variant documented in AGENTS.md's Encoding section (as opposed to #11 above, which was confirmed pre-existing in the raw bytes) — not chased further given the small scope (3 files, none on the registration path).
• Remainder (~20 files) is the usual long tail of scattered one-off undefined variables/functions, a Bad type for argument on set_name() (string vs string*, 2 files), a Redeclaration, and a couple more isolated syntax errors in unrelated content files (quest NPCs, weapon templates) — consistent with every other lib in this project's own long tail, not triaged file-by-file per AGENTS.md §6b.

Memory stayed healthy throughout every sweep pass on this 23GB host, though two of the three sweep runs briefly hit low-availability windows (<1.2GB) caused by OTHER concurrently-running agents' own lpcc sweeps on different libs (confirmed via ps aux — not this lib's own process) — the affected runs were killed and cleanly retried once the concurrent load eased, per AGENTS.md §6b's monitoring guidance.

How to run

cd libs/bixiecanyang
~/src/fluffos/build-debug/src/driver config.fluffos
python3 ../../scripts/mudclient.py 127.0.0.1 40065 --timeout 25 --idle 1.5 \
  --send "" --send "<newid>" --send "y" --send "<中文名字>" \
  --send "<wizpwd>" --send "<wizpwd>" \
  --send "<password>" --send "<password>" \
  --send "0" --send "y" \
  --send "[email protected]" --send "m" \
  --send "look" --send "quit"

Retroactive fix (QA re-verification pass, 2026-07-23): log_error() compile-warning spam (AGENTS.md §15af)

Found (proactively, before the first re-verification boot — same shape as several sibling libs in this project's ES II/夕阳再现-adjacent families) during a routine re-verification pass: adm/obj/master.lpc's log_error() showed the raw compiler diagnostic text to ANY connected player, not just wizards — no wizardp() gate at all — for every compile diagnostic funneled through APPLY_LOG_ERROR, including harmless "Unused local variable" warnings triggered by the first lazy compile of an ordinary room/command file. Fixed proactively before booting (same pattern as dtsl/dtsl2/dfgs2/wuhanzhan/shenzhou): only show the full diagnostic to a wizard; only alarm an ordinary player with the generic default error message for a genuine compile error (gated on absence of "warning:" in the message) — #include <runtime_config.h> was already present, so no additional include was needed here.

Re-verified with a fresh full registration (real name 秦湖, following the documented wizpwd→password→gift→email→gender flow) reaching the actual game world (铁枪庙 starting room), then look/score both producing correct, clean output with zero compiler-warning spam and zero real error:/denied/too deep recursion lines in debug.log.

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

• Formatter: ran format-corpus.mjs over all of work/ (13,558 files, 13,459 written/reformatted, 37 already-clean, 62 refused with an error — expected on legacy code, not chased individually).
• Native retest against rebuilt driver (build-debug/src/driver, rebuilt from latest upstream master): clean boot, zero fatal errors in debug.log. Full registration re-verified end-to-end on the now-reformatted source with a fresh real Chinese name (秦岳), reaching the actual game world (武庙 starting room); look/score/quit all produced correct output (score's full character-sheet box rendered correctly), zero real errors in debug.log. No regressions from the reformat or the fresh driver build.
• WASM build: preload completes and prints Initializations complete. cleanly. Login is blocked by two issues, neither of which is a mudlib bug and both already anticipated by this project's known-limitations list: 1. adm/daemons/logind.lpc's logon() enforces an intentional uptime() < 30 post-boot grace period (an anti-crash-loop safeguard, same pattern this lib already needed natively — see the 30s config-startup message players see for real). The stock scripts/wasm_client.js calls fluffos_connect() immediately after boot with no delay, so it always trips this gate before any --send is even processed (logon() runs at connect time, before the first line of input). Confirmed this is purely a harness-timing artifact, not a WASM defect, by testing with a scratch variant of the harness that adds a real 31s delay before calling fluffos_connect() — after that delay, login proceeds past this check. 2. Past that point, login hits adm/daemons/band.lpc's is_banned(), which does sscanf(site, "%s.%s.%s.%s", ...) != 4 on query_ip_number(ob) and treats a non-matching result as banned — this is exactly the documented query_ip_number() WASM-mode limitation (AGENTS.md's post-conversion-tooling section): the wasm build's loopback connection doesn't format as a real dotted-quad, so the sscanf never matches 4 parts and every login is rejected as "banned" (你的地址在本 MUD 不受欢迎...). Not a mudlib bug — do not patch; this lib is a second concrete example of the same driver-side gap bxsj's sited.lpc demonstrates.

- Verdict: boots cleanly under WASM; login cannot complete due to the driver's documented query_ip_number() limitation (once the unrelated 30s harness-timing artifact is worked around). Not a regression, not a mudlib bug.

WASM-enablement pass (2026-07 standard: loopback-allow, uptime bypass, admin seed)

Gates patched:

• adm/daemons/logind.lpc logon() (~line 80): the uptime() < 30 startup-grace destruct now applies only to real remote dotted-quad addresses; loopback and malformed/empty IPs (current WASM driver behavior) are exempt. (Previously documented as a WASM blocker; now bypassed per the standing §1.3e policy.)
• adm/daemons/band.lpc is_banned() (~line 108): loopback / 127. prefix / empty / non-string / non-dotted-quad → return 0. This also supersedes the original "malformed IP => banned" behavior that was the second documented WASM blocker for this lib.
• The per-IP cap in get_id() (ip_cnt > 8) counts users sharing the connecting IP; under loopback-only local play this could in theory trigger with 9+ tabs, but ip_cnt is computed in logon() before the gate and the limit is generous; left unpatched (wiz_level>=2 and welcome-listed users are already exempt; fluffos is admin). Note if 9-tab local testing is ever needed, add a loopback exemption here.

Admin account: id fluffos, login password Mud@2026, wizpwd Wiz@2026 (this lineage's separate "管理密码" asked during registration), display name 浮浮, (admin) via adm/etc/wizlist. Because wizlist was seeded before registration, the account came out of registration already with wizard view; verified update /d/quanzhou/tieqiang → 成功 on re-login. Save files for the orchestrator to force-add: libs/bixiecanyang/work/data/user/f/fluffos.o and libs/bixiecanyang/work/data/login/f/fluffos.o (untracked dirs).

Retest: fresh normal registration (id ceshisan, name 秦风, female) end-to-end into the world, look/score/quit correct, 0 new errors in debug.log; test char saves removed.

Retrofit (2026-07-24): fail-closed loopback check (security correction)

The loopback-allow gate patched above originally also treated a non-string/empty/malformed query_ip_number() result as loopback (a defensive stand-in for the WASM driver bug). That driver bug is now fixed upstream, so this was tightened to fail-closed: only an exact "127.0.0.1" / "127."-prefix / "::1" match bypasses the gate; a malformed or non-string address now falls through to the original gate logic (treated as untrusted/remote) instead of being auto-allowed. Re-verified fluffos login still works after tightening.

WASM 修复摘要（迁移自 meta.json 的 group_note）

基于"夕阳再现"衍生引擎的独立游戏。状态已从过时的 limited 修正——这份档案自己的 README 里从未记录过任何缺陷说明，本轮重新测试也没有发现：一次完整的 WASM 注册流程（英文 id 仅限小写字母 → 确认 → 中文名字 → 管理密码+确认 → 登录密码+确认 → 天赋摇点，0 为随机 → 接受 → 电子邮件，需 id@address 格式 → 性别）全程无错误地进入"武庙"起始区域（"你连线进入这个世界，开始了自己的江湖生涯"），预先播种的管理员账号（fluffos/Mud@2026）也能正常登录（"您目前权限：(admin)"）。

深度功能测试（第二轮，2026-08-04）

按 §10.7 方法论用原生 driver（端口 40065）跑了一遍超出注册流程的完整游玩 ——移动、战斗、死亡/复活体系，并在此过程中主动核对了 AGENTS.md 已归档 的几类高发 bug。

主动检查命中 3 处，全部在 adm/daemons/logind.lpc 里：

1. printf("%O\n", ob)（§7.34）——随机接受姓名的分支，紧挨在 ob->set("name", Rname);（"请设定您的管理密码："提示之前）。 2. printf("%O\n", ob)（§7.34）——手动输入姓名的分支，同样紧挨在 ob->set("name", arg); 之前，与上一条是同一 bug 的两条平行路径 （加入 AGENTS.md §7.34 的"两条平行路径"分组，与 hc/yxjh/ xkyx3b/mnhf 同类）。 3. if (ob->query("age") == 14) { ... }（§8.9）——user->setup() 之后 的食物/饮水初始化判断，用的是登录桩对象 ob 而不是玩家本体 user 的 age（只有 user 会在 update_age() 里被设成 14），错的 对象上永远读不到 14，食物/饮水永远不会走到这段初始化逻辑里。这是 AGENTS.md §8.9 的第五个独立血统实例（前四个是 cctx/niaoren 共祖对、 yxjh、ldtxii/ldtx 共祖对），用的是 yxjh 那种精简写法（裸 if，没有额外的 !user->query("food") 门槛）。

三处一次性改完，跑了 formatter（首次接触触发全文件重排版，用 git diff | grep -B2 -A2 "printf\|query(\"age\")" 核实过只有这三处语义 改动，其余全是空白/大括号风格规整）。修完后注册新角色食物/饮水条显示 满格（16/16），确认 §8.9 修复生效。

注册流程本身有几个容易搞反的细节（本轮踩过坑，供后续参考）： 双密码系统里，第一组输的是"管理密码"（wizpwd），第二组才是真正登录 用的"普通密码"——用第一组去登录会被强制要求重设普通密码 （"您输入的是管理密码，请重新设定您的普通密码"）；天赋分配菜单要走 "0"（系统随机）再 "y"（确认接受），漏掉任一步都会卡在原地重问。

游玩内容：注册成功的测试角色（沐豪侠 / id bxcyfu）从"北疆小镇" 出发，逛了赛马场（"姑娘追"维吾尔族特色内容）、草原、峡谷——沿途遇到 的 NPC 全是 attitude "friendly"，fight 指令一律被判定不同意开战。 确认 kill 指令能绕开这层 friendly 同意门槛，对赛马场的"阿拉木罕" （combat_exp 5000）直接 kill，触发了一整套战斗流程并以测试角色死亡 告终。

死亡后发现死亡/复活系统与 bmxkx2001/yszz 同宗：死亡后被送到 "鬼门关"，NPC 是"白无常"，判词、场景描述、death_stage() 的多阶段 call_out 结构与 d/death/npc/{wgargoyle,bgargoyle}.lpc 几乎逐字一致 ——这是 AGENTS.md §7.68 已归档的复活软锁死 bug 类的又一实例。核对 wgargoyle.lpc/bgargoyle.lpc，两个文件都是同样的 if (!ob || !present(ob)) return;——只要复活序列进行中的任意一瞬间 present(ob) 判假（哪怕只是暂时不在场，不是真的走了），整个复活流程 就会被无声永久放弃，角色卡成鬼魂再也无法复活，没有任何报错提示。

按 §7.68 已验证过的修法，把守卫拆成两段：!ob（对象真的已销毁）才 永久放弃；!present(ob)（此刻只是不在场）改为 5 秒后重试同一阶段，不 再放弃。两个文件应用完全相同的改法，formatter 跑完确认 diff 干净（各 9 ++++++++-，非全文件重排）。

验证：重启 driver 让改动生效后，用新一轮 kill guo jing（郭靖， 北大街的另一个高战力 NPC）复现了一次死亡，角色被送到鬼门关的白无常 处。这次没有主动制造"复活序列进行中被挪走"的中断场景（不像 bmxkx2001 当时是被一个无关 NPC 的强制移动撞上），而是让连接在等待 过程中断线重连——call_out 链本来就不依赖网络连接是否在线，所以这个 测试严格来说只验证了"正常、未被打断"路径在改动后仍然完整走完：断线 期间复活序列在服务端自行跑完，重新连线时角色已经复活、站在起始房间 "武庙"，score 显示"你共死亡：1 次"、精/气条降到一半（符合死亡惩罚）， 食物/饮水满格，没有任何异常。也就是说，这个具体实例上的修复是照搬 已证实的 bug 类模式应用的，没有在 bixiecanyang 里独立复现出"中途被 打断导致永久卡死"的原始故障场景——诚实记录：如果以后有更方便强制 挪动鬼魂 NPC 的手段，值得回来专门验证一次中断场景。

git status --short libs/bixiecanyang/ 复查干净：只有三个源文件的修改 是跟踪变更，测试角色的新存档（data/{user,login}/b/）保持未跟踪状态， 未纳入提交（沿用本项目约定，测试角色存档不提交，只有预置的 fluffos 管理员账号存档例外）。

更正（2026-08-05）：§7.68 复活软锁"修复"已撤销

上面提到的"鬼魂离开/不在场时被永久放弃复活流程"曾被当作 AGENTS.md §7.68 记录的一类 bug 修复（把单次判定改成每 5 秒重试）。经用户指出并 重新审视：这更可能是有意的游戏设计，不是 bug——大多数这类档案里 鬼魂根本无法自行移动，所以"不在场"要么从未真正发生，要么是"离开去 在阴间游荡，想回来时再走回这个房间、流程会通过 init() 重新从头开始" 这种有意为之的宽松机制，而不是需要强制追上玩家的错误。强行重试还可能 引入新问题：如果鬼魂之后又走回这个房间，旧的重试和 init() 重新触发的 新一轮流程可能同时运行，导致对话重叠错乱。已把这处改动撤销，恢复成 原始的 if (!ob || !present(ob)) return; 单次判定写法（bmxkx2001 除外——那份档案里这确实是一个真实存在、经过实际复现验证的 bug：鬼魂 本身完全无法移动，是另一个不相关的 NPC 强行把鬼魂拖走导致的）。详见 AGENTS.md §7.68 顶部的撤销说明。

§7.86 跨库扫描修复（留言板 post 崩溃）

• BULLETIN_BOARD inherit + 多余 replace_program() 致命形状（AGENTS.md §7.86，post 命令崩溃）：全档案 100 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（2026-08-14，round two，新驱动重测）

针对驱动升级（quest_times/win_times %-operator 修复 + Warning/warning 大小写回退兼容）做的重测，逐项独立核对了本 session 新确立的检查 项（不只是信任 NOTES.md 已有记录）：log_error() 确认已经有更精 细的巫师/玩家分流+严重度判断（§15af 记录的修复）；§8.9、printf 泄 漏、§7.68 撤销均确认代码状态和记录一致；feature/dbase.lpc 未发 现密码写保护，不适用 tybxjh/wlhd 那一类 bug；win_times 的 %-operator 也已用 to_int(query("win_times")) % 5 （d/city2/npc/refereew.lpc:146）。

本轮新发现并修复的 PROGRAMMING bug

1. log_file()（adm/simul_efun/file.lpc）本身缺少 assure_file() 保护（AGENTS.md §7.11-class）：已加上前向声明 + assure_file(LOG_DIR + file);。 2. cat()（同一文件）对不存在文件的空指针式崩溃，主动加固： 未在本档案现场触发，属主动加固，改成 write(read_file(file) || "");。

实测过程

管理员 fluffos/Mud@2026（此前已注册，(admin) 权限此前已确 认）真实重连两次：第一次专门用 update /adm/simul_efun/file 确认 写权限仍然生效（"重新编译 /adm/simul_efun/file.lpc：成功！"），第 二次单独验证密码重连本身。均成功登录，存档数据一致。全程 debug.log 无运行时错误。驱动按精确 PID 结束；测试期间产生的存 档时间戳增量已 git checkout -- 还原。

Round three deep functional test (2026-08-18)

Registered a fresh character (killguoj) and ran it through a full 死亡/复活 playthrough — combat kill by a d/death/npc/{bgargoyle, wgargoyle}.lpc (鬼门关) death sequence — with a deliberate mid-chain disconnect/reconnect, mirroring the reconnect-race scenario documented in AGENTS.md §7.112.

Bug found and fixed: AGENTS.md §7.112 (duplicate death_stage() call_out chain on reconnect)

Both d/death/npc/bgargoyle.lpc and d/death/npc/wgargoyle.lpc had the textbook vulnerable shape: init() unconditionally scheduled a 5-stage death_stage() call_out chain with no re-entry guard. Because enable_commands() re-broadcasts init() to every object in the room and this lib's clone/user/user.lpc::reconnect() calls enable_commands() on every reconnect, any ghost sitting in 鬼门关 who reconnects even once mid-sequence would get a second, independent copy of the whole chain stacked on top of the first — the two chains racing to move/reincarnate the player, potentially double-applying the death penalty or misrouting the revival.

Fix (matches the reference shape from dtsl/sj/yueyingqiyuan etc. in §7.112): added a per-victim set_temp("death_stage_active", 1) / query_temp(...) guard around the call_out() scheduling in init(), cleared at every exit point of death_stage() (early return on !present(ob), the aggressive-living-player branch, and the final reincarnation branch).

Live-verified: registered killguoj, walked to 铁枪庙, picked a fight with 乌鸦 to die, landed in 鬼门关 with 白无常 (wgargoyle.lpc). Deliberately dropped the connection ~2s after the death-stage chain had already started, then reconnected as the same character mid-chain (which re-triggers init() via enable_commands()). Post-fix behavior: each of the 5 death_msg[] lines appeared exactly once, in order, 5 seconds apart, with no interleaving or duplication; the final stage reincarnated the character exactly once (moved to 武庙, startroom reset, inventory dropped) and score afterward showed "你共死亡：1 次" — a single, correctly-counted death, not a doubled one. Confirms the guard suppresses the duplicate chain without breaking the normal single-chain path.

Standing round-three checklist — other items checked

• §7.111 (master.lpc standard_trace() crash on error["object"] == 0): already fixed in this lib — adm/obj/master.lpc:221 already guards with objectp(error["object"]) ? file_name(error["object"]) : "<none>". Clean, no action needed.
• §7.113 (netdead reconnect never restores heart_beat): checked the real reconnect path (adm/daemons/logind.lpc::reconnect() → clone/user/user.lpc::reconnect()). Unlike shzs, this lib's actual reconnect function correctly calls set_heart_beat(1) (clone/user/user.lpc:198), matching its net_dead()'s set_heart_beat(0). Clean, not applicable.
• §7.90 (eval-cost too low for this lib's NPC-creation cost): config.fluffos already sets maximum evaluation cost : 50000000, far above the 700000 template default that triggers this class of bug. No cost limit reached aborts observed anywhere in this session's testing (combat, room movement, death, reincarnation, two full reconnects). Clean, not applicable.
• §7.11 nosave-dir pattern / logind.lpc save path: this lib's log_file() simul_efun was already hardened with assure_file() in the round-two pass above, so every log_file() call site in logind.lpc (registration id-creation log, etc.) is covered. Separately checked adm/daemons/toptend.lpc::topten_save()'s unguarded write_file(f_name, ...) (the nitan_ceshi-class non-fatal §7.11 variant) — its target directory /topten/ ships with the archive and is present (rich.txt, exp.txt, etc. all exist), so this write path is not broken here. logind.lpc's other write_file() calls (users, maxonline, iduser counters) target its own adm/daemons/ directory, which obviously exists. Registration itself was live-verified working end-to-end (new character reached score successfully). Clean, not applicable.

Test hygiene

Driver ended by exact PID. Test-character save files (killguoj, deathtst, deathwu and their .o twins under data/{login,user}/) were left untracked and deleted, not committed — only the two source fixes (bgargoyle.lpc, wgargoyle.lpc) are part of this commit, per this project's convention that test-character saves never get committed.

§7.100 跨库扫描修复（ROOM 冗余 replace_program() 关闭包炸弹，2026-08-19）

同一形状覆盖到几乎所有房间基类（机制详见 AGENTS.md §7.100）。本库属 于该扫描已知最大规模的 10 个库之一。二进制模式脚本机械删除了 5241 处独立、未注释的 replace_program(ROOM); 整行。另外手工清理了造房工 具代码生成模板里内嵌的同一形状，本库有两份完全同形状的拷贝 （clone/misc/roommaker.lpc、u/fyue/misc/roommaker.lpc，各 1 处 字符串拼接，和同一 XYZX 血统的 longyunmeng/xyzxfk/xyzx 完全一样的形 状）。删除总计 5243 行，与本次扫描 FINDINGS.md 记录的 bixiecanyang 存活命中数完全一致。

验证：干净启动一次真实调试驱动，端口 40065 正常监听， work/log/debug.log 全程无新增内容。用已播种的 fluffos/Mud@2026 管理员账号连线，在北疆小镇/巴依家庭院/巴依家客厅之间往返走了十余个 房间（含 NPC 互动），look/score/who 均正常（score 首次惰性编译 adm/daemons/combatd.lpc 打印了一条无害的 Unknown #pragma 警告，与 本次修复无关），未见任何 "cannot replace"/"cannot bind" 或崩溃迹 象。测试产生的 data/{login,user}/f/fluffos.o 存档时间戳 diff 已 git checkout 撤销，不提交。驱动按精确 PID kill。

§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): 4 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.

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

新角度：醉仙楼购物 + 丐帮左全拜师。2026-08-04 / 2026-08-14 / 2026-08-18 三轮覆盖了注册、战斗、死亡/复活和 log_file，没有买东西、也没有拜师。 这是真正的夕阳再现/江湖风云血统（d/city/sj.lpc 与 xyzx 逐字节相同）， 不是天涯家族；拜师路线仍是扬州树洞左全，不是华山。TOMUD 的 2060 握手 只在 query_temp("tomud") 时走 get_version()，标准端口 40065 第一 输入就是「请输入您的英文名字：」，不要发 2060。

实测过程

管理员 fluffos / Mud@2026（wizpwd Wiz@2026 不是登录密码）。落地 北疆小镇 /d/xingxiu/beijiang。游戏内时间是「盛夏的清晨」，醉仙楼 F_DEALER 在 night/midnight/dawn 会打烊，本轮赶上可买窗口。

goto /d/city/zuixianlou，list 烤鸡腿八十文钱 / 牛皮酒袋一两白银 / 包子五十文钱（jitui value 80，不是 xyzxfk 那种八十两黄金标价）。 clone /clone/money/gold 后 buy jitui 成功（「你向店小二买下一根烤 鸡腿」）。当场 i 就是九十九两银子 + 二十个铜板 + 烤鸡腿，黄金条目 已消失（找零 10000−80 = 9920）。F_DEALER 对丐帮拒绝购买（穷叫化）， 必须先买再拜。

goto /d/gaibang/inhole，左全源码是 kungfu/class/gaibang/zuo-qu.lpc （文件名少一个 n），apprentice zuo 一次成功：左全收徒，score 「丐 帮第二十代弟子」、师傅左全。cmds/usr/save.lpc 真正调用两个 save() （60 秒内再 save 会假「档案储存完毕」不写盘，本轮只 save 一次）。 user.o 立刻带上 family_name":"丐帮" / master_name":"左全" / generation":20。断线后再连（「重新连线完毕」），称谓/师傅/银子还在。 左全只收男性。烤鸡腿未进 autoload 列表（身上还在是因为这次是 linkdead 重连，不是冷启动读档）。

本轮没有新的 programming bug。巫师账号登录时 score 刷了一条 combatd.lpc Unknown #pragma，是 2026-07-23 §15af 已记录的「巫师看全 部诊断、玩家只看真正错误」分流，不是漏网。live debug.log 是 libs/bixiecanyang/log/debug.log（Boot Time Fri Sep 4 02:01:32 2026）， 无 error: / Too deep recursion。error_handler 把轨迹交回驱动 debug.log。work/log/log 只有编译期 pragma/unused 警告；move.bug 停在 2026-07-22。管理员存档未提交。