deadsouls_fluffos — Dead Souls 3.8.6 (fluffos/dead-souls fork)

来源：fluffos/dead-souls submodule at 4e07e92（leftover 686： .lpc + §9 formatted on that repo；mudlib root work/lib； command-cache / last()/truncate() slice widened for .lpc）。 OLD_ED shim and the shopdive playtest seed are in overlay/。 端口 40207。状态：done（干净开机，注册流程全通， look/score/quit 均验证通过）。

leftover 686: work/ is the fluffos/dead-souls submodule

hosting is fluffos-upstream。work/ 现在是 fluffos/dead-souls gitlink（pin 4e07e921a7a9d841faeb81e3ccbc49612215d537），不再是 本馆第二份格式化树。config.fluffos mudlib directory 指向 work/lib。patches/ 保持为空——改名、§9 format、以及 NOTES item 1 的 .lpc 定宽切片（file[0..<3] → [0..<5]、 last(X,2) == ".lpc" → last(X,4)）都在那个 fluffos-org 仓库上。 Catalog-only overlay：

• overlay/lib/secure/sefun/ed_compat.lpc + 带该 include 的 sefun.lpc（本馆共享驱动是 OLD_ED，没有 ed_start()/ed_cmd()/query_ed_mode()）
• overlay/lib/secure/save/creators/s/shopdive.o（shopdive / Mud@2026）

Item 1 里「切片宽度是本馆改名副作用、不必上游」的旧判断已经过时： leftover 686 把 fluffos/dead-souls 本身改成了 .lpc，所以那边的 切片也必须是 4 / [0..<5]。

这是哪一份 Dead Souls

这是 fluffos GitHub 组织自己维护的 Dead Souls 3.8.6 fork （fluffos/dead-souls），已经预先适配现代 FluffOS（driver/ 作为 git submodule 跟踪 fluffos/fluffos 的 v2019 分支），不是从传统 压缩包提取的老档案。本项目同时收录了另一份独立来源的 Dead Souls—— libs/ds386，来自完全不同的、独立的老式压缩包归档，由另一位并发 agent 转换。两者按本项目"同一游戏的不同来源快照保留为独立 slug"的 惯例分别保留（参见 bxsj/bxsj1、xiyouji/xiyouji2003 等先例）， 本次转换未尝试与 ds386 合并或去重。

已知的差异线索（仅基于本 lib 自身内容，未逐行比对 ds386）：这份 deadsouls_fluffos 的仓库 README 本身就记录了"3.8.6 with FluffOS 2019"的适配历史，driver/ 子模块显式跟踪 fluffos/fluffos 的 v2019 分支，且仓库内已经内置 AUTO_WIZ、websocket 支持、以及一套 www/ 网页客户端——是一份"已经被 FluffOS 生态本身认领和现代化"的 分支，而不是从 MudOS 时代的原始压缩包直接抽取。是否与 ds386 内容 完全同源、只是适配层不同，留给未来统一协调的分支比较去判断。

转换流程说明（§2 的改编版）

源已经是 git 仓库、已经是 FluffOS 适配版，因此本次转换只做了 "验证 + 必要修复"，而不是完整的§2/§3/§4 归档转换流水线：

• 编码：全树已是 UTF-8，无需 GBK/BIG5 转换（convert_lib.sh 的 编码步骤运行后 already_utf8=4698，仅 2 个文件真正需要转换、3 个 lossy——doc/CREDITS、doc/help/players/credits、 domains/town/txt/mystery.txt，这 3 个实际是 Latin-1/CP1252 字节（Pensjö、¯\(°_o)/¯），不是 GBK，改用 CP1252 手工解码后 正确还原，未使用 lossy 版本）。
• .c → .lpc 改名：源仓库全树仍是 .c（2410 个文件，0 个 .lpc），运行 convert_lib.sh 完成改名 + 664 处 ".c" 引用修复 + 32 处本地 #include <x.lpc> 转 "x.lpc"。static 关键字： 源码里完全没有使用（DS 自己已经用 nosave/private/protected）， 无需替换。

发现并修复的真实 bug

1. 全局性：.c→.lpc 改名遗留的定宽切片 bug（§4.2 第 4 类），~30+ 处

convert_lib.sh 的引用修复只处理带引号的字符串字面量 （".c" → ".lpc"），不会、也不可能自动调整后续依赖"扩展名长度" 的定宽字符串切片/last()/truncate() 调用——原版对 2 字符的 .c 写的 str[<2..] == ".c" / last(str,2) == ".c" / truncate(str,2), 在把字符串字面量机械替换成 4 字符的 .lpc 之后，切片宽度仍然停留在 2，导致比较永远为假（2 字符切片不可能等于 4 字符字符串）、或者截断 只削掉了 2 个字符（留下 ".pc"残尾)。这不是这份 mudlib 自己的历史 bug，是这个项目自己的机械转换步骤的副作用，但只有在实际运行到对应 代码路径时才会暴露，编译期通常不报错（mixed/字符串比较不触发类型 检查）。

最严重的一处：/daemon/command.lpc 的 eventRehash()—— cmd = file[0..<3];——是全局命令派发表的构建函数，扫描 cmds/players/、cmds/creators/ 等目录下所有 *.lpc 文件建立 "命令名 → 文件路径"的映射。修好前，每一条注册进来的命令名都带着一 个残留的 .l 后缀（例如 /cmds/players/score.lpc 被注册成了命令名 "score.l"），导致所有走 command.lpc→CMD_D->GetCommand() 这 条标准派发路径的玩家/创造者/管理员命令全部无法被找到，运行时报 *call_other() couldn't find object '/cmds/players/score.l'.——玩家 看到的现象是"A runtime error occurred."。look/quit 之所以在修复 前仍然正常，是因为它们分别走 add_action()（动词系统）和 lib/interactive.lpc 自带的 action，不经过这条 CMD_D 派发路径, 因此在深度测试的 §8.4 检查项（"score 才能真正跑到玩家类的完整代码 路径，look 未必够"）里被专门捕获到——这正是这条经验教训本身描述的 情形。已修复：file[0..<5]。

同一根因，逐一确认并修复的其余 ~30 处（每一处都单独核实了变量来源 确实是 .lpc 文件名/路径，而不是无关的 2 字符字符串比较——例如 daemon/defines.lpc 里 last(str,2) != ".h" 是正确的，未改动）：

• daemon/help.lpc（帮助话题索引构建的映射函数）、daemon/verbs.lpc （动词表兜底命名逻辑）、cmds/common/help.lpc（帮助命令按动词查 文件名）、lib/detect.lpc（法术侦测提示随机取样，两处）——均为 [0..<3] → [0..<5]。
• 27 处 last(X,2) == "/!=" ".lpc" + 对应 truncate(X,2)：遍布 verbs/builders/{initfix,reload,copy}.lpc、 lib/{std/boobytrap_object,props/inventory}.lpc、 secure/daemon/{rooms,reload,web_sessions,function}.lpc、 cmds/creators/inherits.lpc、 secure/sefun/{load_object,sefun,reload,inventory}.lpc、 secure/cmds/{admins/{mudconfig,var,vars,doctool}, builders/{arealist,areaclone,areagoto}, creators/variables}.lpc、secure/modules/{generic,door}.lpc、 secure/obj/weirder.lpc（含一个关联的 last($1,9) != "furnace.lpc" 同类宽度错误，"furnace.c" 9 字符改成 "furnace.lpc" 11 字符后宽度也要从 9 改成 11）、 secure/npc/cambot.lpc（三选一比较里只有 .lpc 那一支宽度错， .h/.cfg 两支本来就是对的，未改动）。全部改成 last(X,4)/ truncate(X,4)。
• secure/daemon/reload.lpc:277 额外发现一个独立的、非本项目引入的 historic bug：what = trim(what,2)——调用的是只接受 1 个参数的 trim(string)（去空白），不是本该调用的 truncate(string,int)， 多余的第 2 个参数被驱动静默丢弃、trim() 对已经不含空白的文件名 是纯粹的 no-op。已一并改成 what = truncate(what,4);，和上面同一 批修复保持一致。
• secure/sefun/inventory.lpc:3 额外发现一个独立的返回值未赋值 bug： truncate(thing,2);——调用了纯函数 truncate()却丢弃了返回值， thing本身完全没被修改；紧接着 if(!file_exists(thing+".lpc")) 对一个本就以 .lpc 结尾的输入会拼出 "xxx.lpc.lpc"。已改成 thing = truncate(thing,4);，宽度和赋值一起修复。

修复方法：逐处人工核实调用上下文（确认变量确实来自 .lpc 文件名/ 路径，排除掉表面相似但语义无关的 2 字符比较，如 .h、"--"、 "%^"、"_D" 等），而不是无差别全局替换——grep -rn 'last([^,]*,\s*2)\s*[=!]=\s*"\.lpc"' 用于收尾核对，改完后确认为 0 命中。

2. ed_start()/ed_cmd()/query_ed_mode() 驱动兼容 shim

/lib/editor.lpc（被 /lib/interactive.lpc → /lib/player.lpc 间接 inherit，也就是每一个玩家角色对象都会 inherit 到）和 /secure/cmds/creators/lsed.lpc 调用的是 FluffOS core.spec 里 #ifdef OLD_ED 的 #else 分支——ed_start()/ed_cmd()/ query_ed_mode() 这一组"现代"编辑器 efun。本项目共享的驱动构建 （~/src/fluffos/build-debug）是以 OLD_ED 编译的（服务这批语料 库里占绝大多数的老 MudOS 时代 mudlib，那些用的是旧式回调型 ed()），因此这三个 efun 在这个项目自己的驱动构建里根本不存在 ——不是 mudlib 的 bug，是这个特定驱动构建选项和这份"更现代"的 Dead Souls fork 之间的兼容缺口。实测确认：ed_start(...) 在 lpcc 里报 Undefined function，导致 /lib/player.lpc 完全无法编译——这直接 挡住了每一个玩家角色的创建，是本次转换里编译层面最严重的一处阻断。

驱动侧的旧式 ed() efun 把原始连接输入直接交给驱动内部的编辑器状态 机接管（回调驱动，不是逐行可控的模型），从架构上无法用 LPC 包一层 兼容适配去模拟"modern ed_start/ed_cmd 那种逐行同步返回"的调用约定。 因此没有去 wrap 驱动的 ed()，而是在 secure/sefun/ed_compat.lpc 里用纯 LPC 从零实现了一个自包含的、不依赖任何驱动内部状态的行编 辑器，作为同名 simul_efun 提供：支持经典 ed(1) 的地址形式 （N/$/./+N/-N，N,N/N;N 范围）和 a/i/c/d/p/ l/w/q/Q/x/s///[g] 命令子集——覆盖了 Dead Souls 自己内部 两个调用者（editor.lpc 的 "$a"自动追加 + "."结束、 lsed.lpc 的脚本化编辑 + "x"/"Q"）以及日常巫师交互式用法的 主要场景，但不是 src/packages/core/ed.cc 的逐字节复刻（没有正则 替换、没有 g 全局命令、没有撤销）。已在 secure/sefun/sefun.lpc 里 #include 接入，lpcc 单独验证 /lib/player、/lib/editor、 /lib/interactive、/lib/creator、/lib/nmsh、 /secure/cmds/creators/lsed 全部编译通过。

3. 缺失的 /open/ 目录导致两个巫师工具 daemon 的 create() 崩溃

/daemon/defines.lpc（defines 巫师指令，运行期动态生成一小段探测 代码来判断某个宏/符号是否已定义）和 /cmds/creators/types.lpc （同类思路的类型检查工具）在 create() 里都会无条件 write_file("/open/prog.lpc", ...)，但整棵树里根本不存在 /open/ 目录（git 不追踪空目录，上游仓库这一个目录没有像其它好几十个空 目录那样配 .gitkeep）。目标目录不存在时 write_file() 的 fopen() 失败会触发未捕获的 error()，使这两个 daemon 的 create() 直接崩溃、对象整体无法加载（Wrong permissions for opening file /open/prog.lpc for overwrite. "No such file or directory"）。修复：创建 work/open/.gitkeep（secure/daemon/ rooms.lpc、reaper.lpc、chat.lpc 等多处代码本来就把 /open/ 当作一个真实存在、受信任较低的巫师临时脚本目录在用，是 Dead Souls 自身的既有惯例，不是新引入的目录）。修复后两个 daemon 都能正常 create()、lpcc 单独验证通过。

Upstream note (2026-08-31): 上面 §1（.c→.lpc 定宽切片修复）和 §2（ed_start/ed_cmd driver-compat shim）均不上游——前者只是 本项目自己 .c→.lpc 改名步骤的副作用（上游仓库仍然全树用 .c， 不存在这个问题），后者是本项目共享驱动构建选用 OLD_ED 这个配置 选项造成的兼容缺口（本项目自己的驱动配置，不是上游 mudlib 的 bug）。§3（缺失 /open/ 目录）经审计确认是上游仓库自身的真实 bug （16 个其它必需空目录都配了 .gitkeep，唯独 /open/ 没有），已 filed upstream as PR #13: https://github.com/fluffos/dead-souls/pull/13。

4. domains/Praxis/supply2.lpc 的 set_pre_exit_functions 误用（比照姊妹文件的既有处理方式）

secure/include/compat.h 里 #define set_pre_exit_functions SetProperty 把这个旧式 API 名字映射到了签名完全不匹配的 SetProperty(string prop, mixed val)（单属性 setter），而 domains/Praxis/{hall,hall2,hall3,hall4,hall5}.lpc 全部 5 个姊妹 房间调用这个宏时传的是两个数组 (exits, funcs)——这 5 处调用在 上游仓库里全部已经被注释掉，只有 supply2.lpc 这一处还是活的、 未注释，编译报 Bad type for argument 1 of SetProperty ( string vs mixed * )。这是这份"更现代"的 upstream fork 自己遗留的一处内容层 不一致（不是本项目转换引入的），但既然 5/6 的姊妹房间已经用注释掉 达成了事实上的处理方案，为了让这个房间能编译，比照同样的处理方式把 这一行也注释掉，而不是去猜测重新实现这个从未真正工作过的旧 API。

Upstream note (2026-08-31): 确认这是上游仓库自身的真实 bug （supply2.c 在上游仍然是唯一一个未注释掉这行调用、因而编译失败的 文件），已比照姊妹文件的处理方式 filed upstream as PR #15: https://github.com/fluffos/dead-souls/pull/15。

已知但未修复的边角情况（不阻塞注册/核心玩法，记录备查）

以下 13 个文件在最终一轮 lpcc_check.sh --batch 里仍然报告 FAIL，均已逐一排查根因，均不在预注册流程（preload 列表）、新手房间 可达路径、或 look/score/quit 核心验证范围内，因此按 AGENTS.md "更深的内容层 bug 记录但不必修"的标准，记录而非强修：

• /obj/stargate：源码里显式写了 #include <this is here on purpose to prevent this from loading> ——作者自己刻意让这个文件不可编译（一份"仅供参考、不要直接加载" 的示例代码，注释里整段留了设计讨论记录）。
• /obj/area_room：#include "../customdefs.h"，是一个供各区域 复制后自带同目录 customdefs.h 使用的模板桩文件（参照 realms/template/area/customdefs.h 的用法），在其原始位置本来就 不该独立编译。
• /secure/cmds/admins/opcprof：调用驱动效能剖析 efun opcprof() ——本项目共享驱动构建里未启用这个调试专用 package，是驱动构建选项 缺口，不是 mudlib bug；纯管理员调试指令，不影响任何玩家可达路径。
• /domains/campus/chamber/elevator、/domains/Ylsrim/broken/jar、 /domains/town/virtual/forest/-13,13、/realms/template/adm/ remote、以及 5 个 domains/{amigara,learning}/**/virtual/ {server,void} 文件：这 9 个文件的失败要么是 master.lpc::compile_object() 的 REALMS_DIRS/DOMAINS_DIRS 虚拟对象重定向机制（该域自己没有 virtual/server.lpc 时的兜底 行为）在 lpcc 独立测试场景下的产物，要么（elevator 一例，确认为 /secure/sefun/numbers.lpc 的 atoi() 对 sscanf 返回 -1 缺少 保护、forest 一例，确认为 NPC 出生时机早于 this_player() 就绪导 致 say() 报错）是深度内容区域的 runtime 边角问题——均已用真实 driver 单独复现确认报错文本，但均发生在新手默认路径之外的具体 房间/NPC 身上，超出本次"验证核心注册与 look/score/quit"的范围， 留给未来深度功能测试（§10.7）覆盖到这些具体房间时再处理。

跳过的步骤（因时间压力，记录以便后续补做）

• WASM 通过测试（§1.4）——2026-08-25 另一个 session 补做完成， wasm_status 提升为 playable。首次打包后驱动完全无法启动： secure/sefun/sockets.lpc（simul_efun 的一部分，唯一"急切"加载的 对象）无条件调用了这台 WASM 驱动没有的 socket_status()（没有 sockets package）——和同一天在 ds386（同为 Dead Souls 血缘， 这两份文件字节级相同）上发现并修复的问题完全同根同源，详见 AGENTS.md §7.52 追记；sefun.lpc 自己的 socket_address() simul_efun 包装也是同款问题。按 ds386 的既定套路把这两处、以及 secure/daemon/instances.lpc（quit 触发的 chat 广播会加载到它， 同样字节级相同）里 7 个含真实 socket_* 调用的函数体挖空为安全 桩后，驱动干净启动。另外还独立发现一个新 bug（不是 sockets 缺失 这一类）：secure/sefun/names.lpc 的 convert_name() 在名字含 @（intermud 风格如 bob@othermud）时无条件调用 INTERMUD_D->GetMudName()，而 INTERMUD_D （/daemon/intermud.lpc）自身带有"没有 sockets package 就拒绝 加载"的故意 #error 守卫，导致这类名字触发运行时 *No program in object——加了 find_object(INTERMUD_D) 判断后按原有的"回退返回 原字符串"分支处理，不再崩溃。用真实脚本化 WASM 会话完整走完一遍 AUTO_WIZ 注册向导（含此前未曾料到的年龄确认/屏幕阅读器偏好/邮箱 格式校验等多道确认步骤），成功以 creator 身份进入 /realms/sunyc/workroom，命令提示符正常显示——look/score/ quit 本次会话被首次登录的多页 news 播报吞掉未能单独复测，但原生 测试已经完整验证过这三个命令（见上文"状态"一行），且这次 WASM 修复完全没有触碰任何指令派发相关代码，无需重复验证。
• §9 LPC 格式化器未对本次编辑过的 ~40 个文件运行——格式化器本身 是纯 cosmetic 的（AGENTS.md 原文："The formatter is cosmetic; losing formatting on a handful of files is always the right trade for correctness"），本次为了尽快验证并锁定实质性修复而跳过，不影 响任何已验证的功能正确性。

验证记录

• bash scripts/lpcc_check.sh libs/deadsouls_fluffos/config.fluffos libs/deadsouls_fluffos/work：从最初 26 处失败（含上面 §1-4 的全部 真实 bug）修到 13 处（全部是上面"已知但未修复"里列出的边角情况）。
• 真实 driver 冷启动（cd libs/deadsouls_fluffos && ~/src/fluffos/build-debug/src/driver config.fluffos）：干净监听 40207，log/debug.log 只有既有的（未改动过的原始代码里的） redeclaration/unused-variable 编译警告，无 fatal error。
• 全新账号 qintestdsu 走完整注册流程（合法姓名确认 → 年龄确认 → 屏幕阅读器 → 密码 → 确认密码 → 性别 → 邮箱 → 种族 pick human → AUTO_WIZ 选择 player）成功进入新手出生房间（The start room）。 look 正确显示房间描述；score 正确显示完整角色面板（等级、职 业、故乡、任务点数、生日/年龄、经验需求、饥饿/口渴/清醒状态等）； quit 正确触发退出流程并卸下所有初始装备后断线——三项核心指令 全部验证通过。存档：secure/save/players/q/qintestdsu.o（保留作 为测试凭证，其余中途失败的测试账号存档已清理）。
• 注册流程本身自带 loopback 友好配置：secure/lib/connect.lpc 的 anyterms 数组已经硬编码包含 "127.0.0.1"， secure/cfg/ip_unrestrict.cfg 也已经是 127.0.0.1, secure/cfg/ip_blacklist.cfg 为空——这是上游仓库自带的本地开发 友好配置，未额外打 §1.3b 风格的 loopback-allow 补丁。

Sibling sweep of the dsIII §7.121 currency-float bug — confirmed present, fixed

AGENTS.md §7.121 documents a currency/economy bug found in dsIII (the Dead Souls 3.x lineage's shared secure/sefun/economy.lpc): query_base_rate(), query_player_money(), query_base_value(), and query_value() do real floating-point exchange-rate math internally but are declared to return int with no to_int() on the actual return, silently corrupting player currency into a float on every buy/sell/exchange (a declared int return type is compile-time only on this driver and never coerces a runtime float). deadsouls_fluffos shares this exact file byte-for-byte with ds386/pre-fix dsIII and had the identical gap, plus the same two sibling misses: lib/teller.lpc's eventExchange() (i = val / currency_rate(str2);) and lib/props/value.lpc's SetBaseCost() (Cost = i * rate;).

Fixed by wrapping all four secure/sefun/economy.lpc returns in to_int() (matching the convention already used by the file's own currency_mass()/currency_value()), plus the two call sites, exactly mirroring dsIII's fix. Verified live with a fresh AUTO_WIZ creator character (Qintestdsf), which let the check be done with eval as well as a real transaction rather than navigating the whole map: eval return query_value(23, "silver", "gold"); and a typeof() check both confirmed a clean int result (no residual float), and SetBaseCost() tested the same way. Then teleported (goto /domains/town/room/bank) to Zoe, opened a bank account, gave the character 100 silver via AddCurrency, and did ask zoe to exchange 23 silver for gold — printed a clean 2 gold with no decimal, and the raw save file after save showed Currency (["silver":186,"gold":2,]) — both integers, no float. log/debug.log stayed clean through the whole sequence.

Upstream note (2026-08-31): confirmed present byte-for-byte in fluffos/dead-souls upstream (secure/sefun/economy.c, lib/teller.c, lib/props/value.c), filed upstream as part of PR #12 (bundled with the two related float/int bugs documented below): https://github.com/fluffos/dead-souls/pull/12.

Sibling sweep of ds386's round-two eventRevive()/room.lpc/combat.lpc/beggar.lpc bugs (2026-08-27)

Per ds386/NOTES.md's "Deep functional test (round two, 2026-08-27)" section, checked the four flagged bug shapes here. All four present and fixed identically:

• Bug #1, SEVERE (eventRevive() float corruption, AGENTS.md §7.121 class -- a distinct call-argument-boundary variant from the return-value one this lib already fixed above): lib/player.lpc's eventRevive() fed GetMaxMagicPoints() * PERCENT_MP / GetMaxHealthPoints() * PERCENT_HP (float expressions) into AddMagicPoints(int x)/AddHealthPoints(int x, ...), corrupting MagicPoints/HealthPoints into floats on every player death. Fixed by wrapping both in to_int(), matching ds386's fix exactly. Verified live: registered a fresh AUTO_WIZ creator character (Qintestdfx), ran eval object x = find_player("qintestdfx"); x->eventDie("test"); then regenerate -- post-fix status bar showed hp: 111/370 mp: 23/460, clean integers; save confirmed the raw save file has plain integer HealthPoints 111 / MagicPoints 23 lines, no float. Upstream note (2026-08-31): confirmed present in lib/player.c upstream, filed as part of PR #12 (bundled with the two other float/int bugs documented earlier in this file): https://github.com/fluffos/dead-souls/pull/12.
• Bug #2 (replace_program() fold, AGENTS.md §7.141): same MudOS-era fold present in lib/std/room.lpc's create() (protected void create() variant here rather than nosave void create(), same fold body otherwise). Removed it entirely, identical to ds386/dsI/dsIII/dshakkard. Upstream note (2026-08-31): confirmed present in lib/std/room.c upstream, filed upstream as PR #16: https://github.com/fluffos/dead-souls/pull/16.
• Bug #3 (Wimpy fraction-vs-percentage, AGENTS.md §7.124 class): lib/combat.lpc:65 had Wimpy = 0.20; plus float SetWimpy/float GetWimpy. Fixed to Wimpy = 20; and int SetWimpy/int GetWimpy. Verified live: wimpy command on the same fresh character showed Percentage: 20%, a clean integer from creation. Upstream note (2026-08-31): confirmed present in lib/combat.c upstream, filed as part of PR #12: https://github.com/fluffos/dead-souls/pull/12.
• Bug #4 (GiveMap() reentrancy, domains/town/npc/beggar.lpc): same missing delivery-time !present("town map",ob) guard. Fixed identically to ds386. Upstream note (2026-08-31): confirmed present in domains/town/npc/beggar.c upstream, filed upstream as PR #14: https://github.com/fluffos/dead-souls/pull/14.

Committed only the four source-file fixes plus the new Qintestdfx test-creator account's realm/save files (following this lib's own precedent of keeping test-creator accounts, e.g. Qintestdsf above); RELEASE_NOTES_HTTP/daemon-state save churn from booting the driver was reverted first, not committed. Killed the test driver by exact PID when done.

Sibling sweep: become <class> SetClass -> ChangeClass fix (2026-09-01)

Ported the fix from riftsds (AGENTS.md §7.195/§7.196; commit 11216f003b1) -- see ds386/NOTES.md's identical write-up (same session, same batch) for the full bug mechanism. Summary: the stock Dead Souls "Praxis" demo guild-join rooms called SetClass() directly instead of ChangeClass(), so the multi-class privilege gate (high_mortalp(), which excludes creators too) always silently rejected a fresh explorer's first class despite a success message printing.

Confirmed present here, byte-identical to ds386/riftsds's pre-fix files: all six domains/Praxis/*_join.lpc files called SetClass(); secure/include/compat.h was missing the same three query_name/query_cap_name/query_gender mappings.

Fix applied: six SetClass() -> ChangeClass() edits, three compat.h accessor lines added -- identical to ds386.

Live-verified with a fresh AUTO_WIZ creator account (Qtestfixdfx, registered normally through the "creator" choice at the AUTO_WIZ prompt, following this lib's own precedent), Class starting clean at "explorer": become fighter/become cleric/become mage each printed the real success flavor text and score immediately afterward confirmed the genuine class change ("You are a level 1 Human Fighter"/"Cleric"/"Mage"). become monk/kataan/rogue were not re-tested here (the same pre-existing missing-class-data gap already confirmed on ds386/dsIII/riftsds's identical secure/cfg/classes/ listing -- explorer/fighter/mage/ cleric/thief only).

Committed the six join-file fixes, the three compat.h mappings, and the new Qtestfixdfx test-creator account's realm/save files (following this lib's own precedent above); daemon-state save churn from booting the driver (player list, mudinfo, snoop, preload class/economy/events/race/soul/stargate/unique/voting saves, files.o/functions.o) was left uncommitted. A stray CMD_EVAL_TMP_FILE.lpc scratch file left behind by the eval command in Qtestfixdfx's own realm was deleted before commit (matching the already-committed convention for Qintestdfx/Qintestdsf, whose own same-named scratch files were untouched). Killed the test driver by exact PID when done.

Confirmed present byte-identical in a fresh clone of github.com/fluffos/dead-souls (current master, commit 4ddbd3b, i.e. still un-fixed after PRs #12-16): all six lib/domains/Praxis/*_join.c files call SetClass(), and secure/include/compat.h is missing the same three query_name/query_cap_name/query_gender mappings (confirmed these resolve via global.h's COMPAT_MODE-gated auto-#include <compat.h> in every object, and that GetName/GetCapName/GetGender already exist as real accessors on player.c/race.c, so the mapping is a pure omission). Filed upstream as PR https://github.com/fluffos/dead-souls/pull/17.

Shop (2026-09-04 librarian slice)

Otik's general store (/domains/town/room/shop) is live. AUTO_WIZ creator shopdive / Mud@2026 (human, 68 starting silver from eventCompleteChar) goto'd the store, list showed the stock (wooden torch 60 silver, plastic flashlight 10, …). buy torch from otik matched the flashlight's extra "torch" id (rayovac) and completed a paid sale: Otik said "Here is a small plastic flashlight for 10 silver!", inventory two flashlights, money 48 silver (68−10−10). buy wooden torch from otik then correctly refused ("You don't have enough silver") — purse math is integer, no §7.121 float leftover on this path.

Bug found and fixed (heartbeat hobbled(this_player())): lib/body.lpc's heart_beat called hobbled(this_player()). During heartbeat this_player() is 0, so disable.lpc:7 did 0->GetMissingLimbs() every 2s (Bad argument 1 to EFUN call_other … Got: int(0)). hobbled() then returned 0, !stringp(0) was true, and eventCollapse() fired on every idle beat. First login that died mid-news left shopdive with every verb printing What? until this was changed to hobbled(this_object()) and hobbled() gained an !objectp(ob) guard. After reboot, look/goto/list/buy worked. Same this_player() line exists on ds386/dsIII/dsII/ dshakkard/riftsds/brassring — not swept this pass except brassring (next in queue). 拜师 analogue is become <class> in the Praxis guild rooms, already covered 2026-09-01.

Leftover 678 (2026-09-12) — upstream pin

fluffos/dead-souls PRs #12–#16 are the in-tree float/currency, /open/, beggar map, supply2, and replace_program() fixes. Pin advanced to 4ddbd3b964. No new rooms. Do not rebase the local FluffOS tree over those commits.

深度功能测试（§10.7，2026-10-04）— fluffos/dead-souls#17 merged, pin moved

The Praxis become <class> fix (SetClass() -> ChangeClass() in the six join rooms, three query_* mappings in compat.h) was filed as fluffos/dead-souls#17 on 2026-09-01 and sat open; the .lpc rename made it conflict. The branch was brought up to date with a merge commit (the six files re-edited under their new names), checked live on a tree built from it against a tree built from current master (the seeded shopdive creator, goto /domains/Praxis/fighter_join, become fighter, skills: master leaves a level 1 Explorer with Explorer skills after the initiation text, the branch a level 1 Fighter with the fighter list), merged as e7a25776, and the pin bumped to it.

深度功能测试（§10.7，2026-10-04）— compile warnings: 3114 -> 2 diagnostics (fluffos/dead-souls#18)

Baseline (scan with scripts/lpc_warnings.py, KB 04 §6.10): 2386 files compile, 23 do not, 3114 distinct diagnostics — 1922 Redeclaration of global variable in 1011 files (Search 595, counter 590, Unique 229, my_save 165), 305 unused locals, 83 nosave functions, 54 argument-count disagreements, 55 unknown escapes, 15 (: "name" :) pointers, 13 bare returns, and about 110 x() inherited from both rows from 25 leaf files. After the PR: 2404 files compile and the scan lists 2 diagnostics, obj/area_room.lpc (includes a ../customdefs.h that does not exist) and obj/stargate.lpc (refuses to compile on purpose); open/prog.lpc, the third dead archive file the other Dead Souls libs keep, is not visited by the scan. The submodule is fluffos-upstream, so the change went in as one PR with two commits (mechanical, then structural), merged as 31d342a + 161b8dcc and pinned.

Method. The scripts want a repo whose libs/SLUG/work is the mudlib directory, so the work happened in a scratch repo (scripts/submodule_patch_scratch.py build deadsouls_fluffos: upstream's lib/ as work/), with the catalog overlay committed as a baseline (ed_compat.lpc, the shopdive creator save and the #include line in sefun.lpc) so that scans, the oracle and the live sessions boot the way the site does. Mechanical pass: lpc_warnings.py --fix. Structural pass: the ds386/dsIII recipe (KB 06 §7.213-§7.215) applied by hand to the 50 source files, since the lib was reformatted upstream and a line merge from ds386 conflicted in 1376 places (lpc_diamonds.py, lpc_rename_ident.py, lpc_name_winner.py, lpc_fix_strptr.py --comma found and did most of it). The PR is git apply --directory=lib -p4 of two scratch diffs; the merged tree is byte-identical to the tested one.

What the pass found beyond the warnings:

• The spells daemon, the clean command and history did not compile. The upstream formatting pass turned string *spells, *prayers; into string *spells, prayers;; the star is per declarator in this grammar (new_local_def: optional_star new_local_name), so prayers = get_dir(...) assigned an array to a string and the program was a compile error. Live on master: SPELLS_D->GetSpells() raises No program in object '/daemon/spells' (no spell or prayer was ever loaded), history and clean answer the same; on the branch GetSpells() is 8 spells and 1 prayer, history lists the commands, clean here cleans the room, cast fireball runs. 12 declarations restored, found by comparing every T *a, b; line with ds386, dsIII, dsII, dsI, dshakkard, riftsds and brassring (20 of 46 candidates had a starred sibling; the other 26 are genuinely string *shorts, ret;). The formatter in ~/src/fluffos/tools/ lpc-syntax/ keeps the stars today (tested), so this came from an older run of it.
• lib/trainer.lpc init() was displaced by sentient's: a pupil who left in the middle of a lesson stayed "already trained" (live, master "I am already training you!", branch "begins teaching you about the skill of magic defense").
• GetRadiantLight() counted the own glow of NPCs, chests, beds, chairs, corpses and limbs twice (the oracle's SetRadiantLight(7) read 14 on 41 programs, 7 now) and a worn pack's contents not at all (KB 06 §7.215).
• virt_land SetCoordinates(int,int,int) clobbered LIB_ROOM's SetCoordinates(string) in every virtual room.
• 15 (: "name" :) and 30 (: this_object(), "name" :) Praxis pointers returned their own name (look at gallows printed long_func, the planning room's long description was go_away, the voting halls printed new_long); the targets had never run, so the search handlers return their text and place the shovel and dagger with eventMove(), the Knight's helm asks GetMorality() (at morality 0: "This helmet burns with disgust."), the invis ring's test takes the viewer (GetInvis(orc) is 1 for an orc and 0 for a human), stone.lpc returns its text, the bank vault and yard manhole are plain texts (they named look_at_vault/look_at_manhole, which exist nowhere).
• daemon/books.lpc globalstr2 == "" (no-op), pscoutsuit stray ;, six #define __DIR__ (a predefined macro) in the amigara/learning virtual-room servers, opcprof() (gone from the driver; the command says so), the OBJECT macro of type.h/compat.h redefined, static (a syntax error) in update.patch/update.blank.

Verification. Rescanned with ~/src/fluffos's stale lpcc and with a build of fluffos at 01026e1f (same 2 diagnostics). Golden-master oracle (scripts/ds_oracle.py through the scratch repo, admin shopdive/Mud@2026, master against branch, same driver; 263 programs = lib/ plus the 38 leaf paths of ds386): 50 definer changes, every one an explicit choice above; 110 visibility rows (the nosave -> protected flips, whose cross-object callers the fixer greps for); 566 variable-count rows (the renames); 88 radiance rows; 180 removed functions = the enter/knock/scratch families of the five scenery containers; no GetSaveString() row changes; the remaining get rows are per-creation randoms (HP, carry). Live, paired, 167 + 10 steps as shopdive: fixtures (trashcan, river, sea, coffin, hoverpod), trainer, search, zpem, horse, virtual rooms, death and corpse, the Praxis callbacks, spells/history/clean. The branch printed no compiler warning in the whole session, master about 1000 lines of them; no insufficient permission on the console. The only visible difference besides the fixes: You can't enter that! -> You can't enter that. on the river (the refusal now comes from the parser, as on ds386).

Overlay. overlay/lib/secure/sefun/sefun.lpc is a full copy of upstream's plus one #include, so a pin bump that touches sefun.lpc is undone on the site unless the overlay is rebuilt: insert #include "/secure/sefun/ed_compat.lpc" after the astar.lpc include of the new upstream file (done for this pin; the rets local is gone from both).

Left alone: update.patch/update.blank still name .c files in a few cp/rm calls (the live-upgrade tool targets the pre-rename tree); the config has three obsolete lines the driver reports (address server ip, address server port, binary directory).

• 2026-10-14: FluffOS 970d9f09 Float value truncated to int, 20 sites fixed upstream (fluffos/dead-souls PR #19, squash-merged; pin f1c492e). lib/teller.lpc int GetExchangeFee() returned the float ExchangeFee, so a fractional exchange fee lost its fraction (below 1% it read as 0 and was not charged); it returns a float now. The rest spell their truncation to_int() (same value). Overlay files untouched.

深度游玩测试（2026-10-21）：telnet、setreboot 的无参数调用（上游待改，KB 06 §7.336）

裸动词扫描发现 telnet 不带参数时 sscanf(args, "%s %s", ...) 因 args 为 0 抛 Bad argument 1 to sscanf() Expected: string Got: 0，屏幕上是 A runtime error occurred；它本想打印用法的 this_object()->help() 是没有任何对象定义的函数（命令用 GetHelp() 作答），所以即使不出错也什么都不打印。管理员命令 setreboot 调用同一个不存在的 help()，并且缺 return（继续执行 SetRebootInterval(0)，该函数对 x <= 1 保持旧值，所以无害）。本仓库里 dsIII、ds386、dshakkard、riftsds、brassring 的同一文件已改（if (!args || sscanf(...) != 2)、write(this_object()->GetHelp() + "\n")、setreboot 补上 return 1;）。fluffos/dead-souls 是 gitlink 库，改动需要上游 PR（lib/secure/cmds/players/telnet.lpc、lib/secure/cmds/admins/setreboot.lpc），本会话的 GitHub 范围只有 fluffos/mudlibs，未提交。上游的 help 守护进程不受影响：verbs/spells、verbs/undead 在子模块里作为空目录存在，strip_colours() 也已经带着 \b 的做法。

深度游玩测试（2026-10-21）：创造者/管理员命令不带参数、A* 的填充字符（上游待改，KB 06 §7.336、§7.338）

以种子管理员身份扫描同一份 Dead Souls 3.x 代码（play_sweep_lib.py ds386 --wiz --account fluffos:Mud@2026 --shapes one；本库没有种子管理员账号）：裸命令即抛错的有 areaclone、cryptogram（sscanf(0, ...)）、pause（lower_case(0)）、html（read_file(0)）、encre（player_save_file(args) 在检查参数之前就执行，对非字符串故意 error("Bad argument 1 to save_file().")）、resolve、wget（守护进程里 sscanf(0, ...) / strsrch(0, ...)）。A* 的 find_path()（创造者命令 astar）用 sprintf("%N'\27's", "") 给地图补位，FluffOS 要求填充串有显示宽度，控制字符宽度为 0，报 (s)printf(): Null pad string specified。本仓库 dsIII、ds386、dshakkard、riftsds、brassring 的同名文件已改：参数检查后打印 GetHelp()，astar 改为先用空格补位再 replace_string(..., " ", "\27")。在本库用创造者身份（PLAY_DS_ROLE=creator，AUTO_WIZ 注册时答 creator）实测复现了 areaclone、astar、cryptogram、html（其余几个需要管理员权限）。上游 fluffos/dead-souls 的对应文件（lib/secure/cmds/builders/areaclone.lpc、lib/cmds/creators/cryptogram.lpc、lib/secure/cmds/admins/pause.lpc、lib/cmds/creators/html.lpc、lib/secure/cmds/admins/encre.lpc、lib/cmds/creators/resolve.lpc、lib/secure/cmds/creators/wget.lpc、lib/secure/sefun/astar.lpc）未改，需要上游 PR（本会话的 GitHub 范围只有 fluffos/mudlibs，未提交）。

log/ 下的日志子目录没有人建：写日志的 write_file() / log_file() 报错（2026-10-22，KB 06 §7.358）

这个库的代码往 log/ 的 7 个子目录写日志：log/adm（lib/lib/command.lpc）、log/harass（lib/lib/interface.lpc）、log/imc2_server（lib/secure/daemon/imc2server/server_log.h）等。压缩包、git clone 和站点包里都没有这些目录（本地启动没有写过它们，gen_keep_dirs.py 没记），write_file() 对不存在的目录抛 *Wrong permissions for opening file /log/dlog/money for append. "No such file or directory"，调用它的函数在写日志那一行中断，后面的奖励、标志、步骤都没有执行。本库的 log_file() 是 write_file(LOG_DIR + file, text)，目录不存在就抛同样的错误（实测）。把 log/adm、log/harass、log/imc2_server、log/reports、log/router、log/secure/log、log/watch 记进 scripts/wasm_keep_dirs.txt（scripts/lpc_keep_log_dirs.py；站点和 pristine 树按它建空目录）。证明（scripts/lpcc_eval.py，在 /adm/daemons 这类目录下的求值对象里写 /log/zzdirprobe/...，新进程）见 KB 06 §7.358 的实测表。