Daniel's World

WASM 修复摘要（迁移自 meta.json 的 group_note）

逸俠之世界（Daniel's World），ES II 血统，台湾出品。是 libs/yxsj 关系很近的手足档案（同样的标题、同样的 [email protected] 联系人、同样的 is_killing/check_legal_name bug 指纹、同样的文件布局），但确认不是逐字节相同的重复档案——即使做完空白/CRLF 归一化，内容依然有差异（比如 adm/daemons/combatd.lpc 有大约 80 行不同），而且这份快照的 securityd.lpc 的 save_object() 路径本来就已经修好了（query_save_file()+__SAVE_EXTENSION__），yxsj 那份才需要应用这处修复——是同一血统里不同时间点的构建版本，不是同一套文件的重新打包。先发现并修复了和 yxsj 相同的重大编码 bug：这份档案同样是 BIG5 编码，但最初的 convert_lib.sh 用了整个语料库通用的 GB18030 默认值，产生了静默乱码（合法的 UTF-8，但是真实存在却错误的中文码点），而不是报出转换错误。已用 BIG5 重新跑了 convert_lib.sh（见处理 yxsj 时做的工具扩展提交），并在重新转换出来的代码树上重做了每一处修复。应用的 WASM 专属修复（和 yxsj 完全一致的模式）：（1）经典的 §8.1 is_chinese()/check_legal_name() 字节区间 bug，改成逐码点 0x4e00-0x9fff 检查，界限从 2/12 减半为 1/6。（2）§7.12 类的 tell_room() 裸 exclude 导致的 message() 崩溃，用 exclude || ({}) 保护修复。（3）master.lpc 的 valid_write() 缺少 'user == this_object()' 保护（valid_read() 本来就无条件回传 1；这份快照里 securityd.lpc 的 save_object() 检查本来就是对的，和 yxsj 不同）。（4）§7.50 类的 is_killing(object) 对 is_killing(string id) 修复了和 yxsj 完全相同的 9 处呼叫点（kungfu 类的 roar/nuke 档案、u/d/daniel/ 下的 NPC 档案、cmds/std/surrender.lpc）。（5）data/ 下 45 个 CRLF 换行的 .o 存档档案统一转换成 LF。adm/daemons/ftpd.lpc 和 adm/daemons/network/dns_master.lpc 都处于休眠状态（在 preload 里被注释掉，只有带保护的巫师指令呼叫者），保持原样。和 yxsj 相同的已知非阻断性问题，没有进一步深挖：chinesed.lpc/emoted.lpc 在启动时都会抛出一个被捕获的'Illegal mapping format while restoring dict/emote'；登录过程中会出现一次针对 obj/user 的 command_hook 的短暂'apply() with insufficient permission... needs: private, has: hidden'，但不会阻挡任何功能。管理员账号播种：fluffos (admin) 加入 adm/etc/wizlist。注册流程在格式化前后都用一个真实的中文名字完整验证过；管理员权限已通过'目前權限﹕(admin)'确认。LPC 格式化工具对全部 949 个档案运行（写入 919 个，21 个转档之前就存在的错误，9 个未改动）——数字和 yxsj 完全一样，符合这两份是同一代码库不同时间点快照的判断。

§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 11 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。

§7.100 sweep (2026-08-19)

Fixed the corpus-wide inherit ROOM; ... replace_program(ROOM); redundant-replace bug (AGENTS.md §7.100). 185 live occurrences deleted: 184 via scripted sweep (fix_710_room.py), plus 1 hand-fixed roommaker-tool template (obj/roommaker.lpc, simple string-builder variant — a second, already-commented-out copy of the same template lower in the same file was correctly left untouched). 2 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 40170.

§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.

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

Full live continuous playthrough (build-debug driver, port 40170, raw Python socket client): registered a real character (譚雲/Tanyun), apprenticed to a sect NPC (柳淳風, 封山劍派/swordsman), learned a skill via learn ... from ..., fought a wild NPC with the safe fight command (ended in unconsciousness with a scheduled revive() call_out, not death — confirmed working, not a soft-lock), did two separate reconnect cycles (one abrupt-disconnect-then-reconnect while a revive() timer was still pending, one explicit quit+fresh relogin after a real wall-clock gap), and cross-checked find_player() + tell + the admin call command between two simultaneous connections. Registered and confirmed the seeded admin account (fluffos/Mud@2026) live for the first time (previously wizlist-only, no save file existed) — 目前權限﹕(admin) confirmed, and score correctly showed the admin-only extended attribute block.

Systematically grepped this lib's source for every standing cross-cutting bug pattern in AGENTS.md (§4.3, §4.4, §7.11, §7.80/118, §7.86/100, §7.103, §7.112, §7.121–§7.153) rather than spot-checking. Found and fixed 6 real bugs, all from the already-documented pattern list — same lineage, same bugs as yxsj's own round-two pass, since these two libs share the ES2/daniel codebase at different points in its history:

• data/chinese.o (§4.4, BIG5 0x5C-collision): this snapshot's BIG5→UTF-8 conversion had the exact same collision as yxsj, but WORSE — not just the one trailing dict entry, but 10 separate string values ended in a 0x5C-final BIG5 character (功/許) immediately before their closing quote (e.g. "move":"...輕功\", "celestial":"...神功\", plus 3 more mid-string cosmetic instances like "force":"內功\心法"). Confirmed via boot: before the fix, chinesed's dict restore silently zeroed (same Illegal mapping format while restoring dict class as yxsj, though this boot didn't even surface the warning — the corruption ran past EOF differently given the extra breaks), and learn/skills crash-risked on any entry past the first broken string. Re-derived the whole file from raw/mudos/es2lib/data/chinese.o with a custom escape-aware BIG5 decoder (unescape \-doubling byte-for-byte, THEN BIG5-decode, THEN re-escape for LPC string syntax) rather than a blind iconv — diffed byte-for-byte against the naive conversion and confirmed the ONLY changes were removal of the 13 spurious backslashes, nothing else. Verified live: fresh boot showed no chinesed restore warning at all, and learn sword from master / skills both worked crash-free on the test character.
• §4.3 static→nosave collision, 7 call sites / 4 files (cmds/wiz/call.lpc's CALL_PLAYER, adm/obj/master.lpc's CRASHES ×3, cmds/arch/purge.lpc's PURGE ×2, adm/daemons/securityd.lpc's promotion) — identical shape and identical files to yxsj's fix, confirmed against raw/'s "static/..." literals. Reverted all 7, and hardened adm/simul_efun/file.lpc's shared log_file() with an assure_file() call before write_file(), matching yxsj. Verified live: logged in as admin (fluffos), ran call tanyun->query("name") and call tanyun->query("class") against the live test character — both executed cleanly and appended correctly to the pre-existing (real 2000-era) log/static/CALL_PLAYER history file with no crash.
• §7.103 log_error() broadcasting raw compile warnings (adm/obj/master.lpc): added the strsrch(message,"warning:") == -1 guard. Verified live: a lazy first-compile of cmds/usr/quit.lpc during testing threw multiple Unknown escape sequence warnings (the ~129-instance cosmetic tail of the same §4.4 BIG5 collision, left unfixed here exactly as documented for yxsj — none of these are crash-causing) and none of them leaked to the connected test player's screen.
• §7.86/§7.100 redundant replace_program() (obj/board/wizard_j.lpc): inherit "/std/jboard" + replace_program("/std/jboard") in create() — same file, same shape as yxsj's fix (missed by the earlier macro-based corpus sweep since this board uses a literal path). Deleted the redundant replace_program() call. Verified live as the admin test character: navigated to /d/wiz/jobroom, the board loaded without the "cannot bind an lfun fp to an object with a pending replace_program()" crash, and project/read new both worked end-to-end.
• §7.153 std/jboard.lpc do_read(): identical missing-else shape to yxsj — read new/read next computed the correct index, then fell through into the unconditional sscanf(arg, "%d", num) numeric parse, which always failed on the literal string "new"/"next" and returned "你要讀第幾個計畫的簡報﹖" instead of showing the note. Fixed by chaining it into the existing if/else if instead of a bare trailing if. Verified live: project <title> followed by read new now correctly displays the just-posted note instead of erroring.

No other cross-cutting pattern from the AGENTS.md checklist matched live source in this lib (checked explicitly, not just assumed clean): §7.121 (no float-typed economy function found), §7.122/§7.132/§7.134/ §7.135/§7.136/§7.139/§7.144/§7.145/§7.146/§7.147/§7.148/§7.149/§7.151 (no matching shape found by grep), §7.129 (tell_room() wrapper already has the exclude || ({}) guard from the earlier §7.12-class onboarding fix), §7.133 (net_dead() IS defined on obj/user.lpc), §7.150 (the login-verification object restores directly onto itself, no throwaway clone; the leftover post-exec() shell object is destructed lazily on the player's *next* disconnect/quit/wizard update, confirmed via code read of obj/user.lpc's own net_dead() and cmds/usr/quit.lpc/cmds/wiz/update.lpc — no active heart_beat on the orphan in the meantime, so no resave-corruption path), §7.152 (no remove_living_name()/unset_living_name call exists anywhere in this codebase, so the living-name registration from initial login persists across reconnect() without needing to be reset — confirmed empirically: cross-connection tell tanyun ... from the admin account worked immediately after tanyun had already gone through one full netdead→revive→reconnect cycle).

Two harmless, pre-existing oddities noted but explicitly NOT touched (design/leftover, not crashes): d/snow/schoolhall.lpc and d/wiz/jobroom.lpc both end their create() with a "obj/board/xxx"->foo()/->???() call to an undefined function on a board object — apparently a preload idiom from the original codebase; call_other to an undefined function is a silent no-op on this driver (confirmed no error in either boot output or the caught-error handler across this whole session), not a crash. Also: cmds/usr/recall.lpc sends players to /d/wiz/entrance while cmds/usr/recall2.lpc (a distinct, separately-registered command) sends players to /d/snow/inn — both work correctly, left as-is since choosing which should be "the" recall command is a content decision, not a bug.

This lib's actual playable world is a much smaller custom build (a "school" starting zone plus a 雪亭鎮/snow-town zone with the ES2 sect system) than doc/help/newbie's text describes (which lists many more locations/sects than are reachable from the actual map) — a content/documentation mismatch, not a programming bug, left unfixed per the scope boundary.

Admin account fluffos/Mud@2026 confirmed seeded and working (kept, with its live-registered save file). Test character 譚雲 (tanyun) and its test bulletin-board post removed before commit.

深度功能测试（2026-09-04，shop）

Live pass, port 40170. 2026-08-27 already apprenticed 譚雲 to 柳淳風 (封山劍派); this round is shop only. First send is the English id (no GB/BIG5 menu). Admin fluffos / Mud@2026 lands at /d/school/gate.

Shop: goto /d/school/store (逸俠中學福利社). list is a static price board; keys are English (dumpling, not 包子). clone /obj/money/gold then buy dumpling succeeded: “你買下了一個包子”, inventory gained 包子 plus change (銀/銅). std/room/store.lpc treats any non-zero can_afford() as success, so the gold-only “零錢” return-2 path does not block this counter (unlike cmds/std/buy.lpc / es2's waiter). No finance patch needed for this purchase.

/d/snow/npc/inn (飲風客棧) is not on the walkable map (no room exits point here). goto hits make_inventory() → call_other(0, …) because objects names /d/snow/npc/waiter and that file is absent from both work/ and raw/ (also absent from sibling yxsj). Not restored from es2 — this snapshot never shipped the NPC. Walkable /d/snow/inn (義守客棧) has an empty objects map.

No code change. log_error() already gates on "warning:" (2026-08-27); this login did not broadcast compile warnings.