dsh-config-manager 0.1.69 → 0.1.71

package/lib/foreign/read-zed.js

@@ -0,0 +1,473 @@
+/**
+ * Zed 用户数据目录的**读盘层** + 真值表路径。
+ *
+ * 真值表（read-vault §10.1 zed 行 + chat-import convert/zed.mjs:63-84；read-movein 附录 A、
+ * read-sessions-manager ⑧.4 第 24 行交叉核对一致）：
+ *  - win32  `%LOCALAPPDATA%/Zed/threads/threads.db`（**目录名大写 Zed**，且用 LOCALAPPDATA）
+ *  - darwin `~/Library/Application Support/Zed/threads/threads.db`
+ *  - linux  `$XDG_DATA_HOME/zed/threads/threads.db`（缺省 `~/.local/share/zed/threads/threads.db`）
+ *  - `--user-data-dir` 会**整体改写 data_dir**，那种安装本层探测不到 → 只能由用户显式给路径
+ *    （真值表同款说明；本层不假装能找到它，也不为此报码 —— 那是「不知道」，不是「读不到」）
+ *
+ * 表面（竞品 sources/zed.mjs:50-58 的签名判定）：**只有一张 `threads` 表**，
+ * 含 `{id, summary, updated_at, data_type, data, parent_id, folder_paths, folder_paths_order, created_at}`；
+ * `data` 是 BLOB，`data_type` 写入端**恒 zstd**（标准帧），`json` 只读兼容。
+ *
+ * 三种方言都要认（参考 lib/convert/zed.mjs）：
+ *  - **v0.3.0（当前）**：消息是 serde 外部标签 `{User:{content:[{Text}|{Mention}|{Image}]}}` /
+ *    `{Agent:{content:[{Text}|{Thinking}|{ToolUse}], tool_results:{<tool_use_id>:…}}}` /
+ *    `{Compaction:{Summary}}`（裸字符串 "Resume" 是 UI 续聊标记）—— 旧代码只匹配小写别名、
+ *    只取 content/message/segments/text/parts，整个线程都记 `message:unmapped`；
+ *  - **legacy**：`{role, segments, tool_uses, tool_results, is_visible}`；
+ *  - 工具结果按 `tool_use_id` 配对，v0.3.0 挂同一条 Agent 消息的对象里、legacy 是消息上的数组。
+ *
+ * zstd 走 `node:zlib` 的 `zstdDecompressSync`（Node ≥22.15 / ≥23.8；本仓库 engines ^22.19）。
+ * 该函数在运行时**能力探测**：缺它时该库整体降级为「一条 zstd-unavailable 的 source-unreadable」
+ * （逐条计数），绝不因为解不开就把线索引成空会话。
+ *
+ * 取证强度 = `fixture`（真机未验证；真实临时库夹具 + 单测端到端）。`parent_id` 非空 = 子代理线程
+ * （上游 UI/ThreadStore 都过滤）→ 本层同样跳过。
+ */
+import zlib from 'node:zlib';
+import { joinFor, normalizePlatform, xdgDataHome } from './platform-paths.js';
+import { isFile, labelForPath } from './session-read.js';
+import { openDetail, pickString, pickTime, prefixCounts, sqliteGate, sqliteOpen, sqliteSkip, splitToolResults, } from './read-opencode.js';
+import { irBump, irTextBlock, irToolCallBlock, irToolResultBlock } from './session-ir.js';
+import { firstUserText } from './session-source.js';
+import { isRecord } from '../utils/guards.js';
+/** Zed 的 data_dir（`XDG_DATA_HOME` 只对 linux 生效、且仅绝对路径） */
+export function zedDataDir(opts) {
+    const platform = normalizePlatform(opts.platform);
+    const product = platform === 'linux' ? 'zed' : 'Zed';
+    return joinFor(platform, xdgDataHome(platform, opts.homeDir, opts.env), product);
+}
+/** Zed 的库：`<data_dir>/threads/threads.db` */
+export function zedThreadsDbPath(opts) {
+    return joinFor(normalizePlatform(opts.platform), zedDataDir(opts), 'threads', 'threads.db');
+}
+/**
+ * `zstdDecompressSync` 走**窄化取值**而不是直接引用属性：`@types/node` 的小版本差异不会
+ * 让本文件编译失败，运行期缺它也只是「这一个库降级」。
+ */
+function zstdDecompressSync() {
+    const fn = zlib.zstdDecompressSync;
+    return typeof fn === 'function' ? fn : undefined;
+}
+/* ---------------- threads 表的解析 ---------------- */
+const ZED_ID_KEYS = ['id', 'thread_id', 'threadId'];
+const ZED_TITLE_KEYS = ['summary', 'title', 'name'];
+const ZED_TIME_KEYS = ['updated_at', 'updatedAt', 'created_at', 'createdAt'];
+const ZED_DATA_TYPE_KEYS = ['data_type', 'dataType', 'type'];
+const ZED_FOLDER_KEYS = ['folder_paths', 'folderPaths'];
+const ZED_FOLDER_ORDER_KEYS = ['folder_paths_order', 'folderPathsOrder'];
+const ZED_PARENT_KEYS = ['parent_id', 'parentId'];
+/** `data` BLOB → 线程 JSON（`data_type=json` 时直接解 utf8；其余恒 zstd） */
+function decodeThread(row) {
+    const data = row['data'] ?? row['thread_data'] ?? row['blob'];
+    if (data === undefined || data === null)
+        return { ok: false, problem: 'absent' };
+    const dataType = (pickString(row, ZED_DATA_TYPE_KEYS) ?? '').toLowerCase();
+    let text;
+    if (typeof data === 'string') {
+        if (dataType === 'zstd')
+            return { ok: false, problem: 'zstd-not-blob' };
+        text = data;
+    }
+    else if (data instanceof Uint8Array) {
+        if (dataType === 'json' || dataType === '') {
+            text = new TextDecoder().decode(data);
+        }
+        else {
+            const fn = zstdDecompressSync();
+            if (fn === undefined)
+                return { ok: false, problem: 'zstd-unavailable' };
+            try {
+                text = new TextDecoder().decode(fn(data));
+            }
+            catch {
+                return { ok: false, problem: 'decompress-error' };
+            }
+        }
+    }
+    else {
+        return { ok: false, problem: 'absent' };
+    }
+    try {
+        return { ok: true, value: JSON.parse(text) };
+    }
+    catch {
+        return { ok: false, problem: 'json-error' };
+    }
+}
+/* ---------------- folder_paths 编码（`\n` 连接 + `,` 连接的 order 索引） ---------------- */
+/**
+ * `folder_paths` 用 `\n` 连接路径、`folder_paths_order` 用 `,` 连接索引 —— 旧代码把它当
+ * JSON 数组解析（jsonCell），真实库上 cwd 永远拿不到。索引缺失/长度不符时退化为字典序。
+ */
+export function zedFolderPaths(folderPaths, order) {
+    const paths = typeof folderPaths === 'string' && folderPaths !== ''
+        ? folderPaths.split(String.fromCharCode(10)).filter((p) => p !== '')
+        : [];
+    if (paths.length <= 1)
+        return paths;
+    const orderText = typeof order === 'string' && order !== '' ? order : '';
+    if (orderText === '')
+        return paths;
+    const indexes = orderText.split(',').map((n) => Number.parseInt(n, 10));
+    if (indexes.length !== paths.length || indexes.some((i) => !Number.isInteger(i) || i < 0 || i >= paths.length)) {
+        return paths;
+    }
+    return indexes.map((i) => paths[i]);
+}
+/* ---------------- 内容块方言映射 ---------------- */
+/** `raw_input` / `input.value` 是原始 JSON 文本：能解析就解析，不能就原样 */
+function toolInputOfText(text) {
+    const trimmed = text.trim();
+    if (trimmed.startsWith('{') || trimmed.startsWith('[')) {
+        try {
+            return JSON.parse(trimmed);
+        }
+        catch {
+            return text;
+        }
+    }
+    return text;
+}
+/** v0.3.0 工具结果 content（`[{"Text":…}|{"Image":…}]`）→ 文本；图片只计数 */
+function v3ResultText(result, acc) {
+    const parts = [];
+    const nodes = Array.isArray(result['content']) ? result['content'] : [];
+    for (const item of nodes) {
+        if (isRecord(item) && typeof item['Text'] === 'string' && item['Text'] !== '')
+            parts.push(item['Text']);
+        else if (isRecord(item) && item['Image'] !== undefined)
+            irBump(acc.ignored, 'block:image');
+    }
+    if (parts.length === 0 && typeof result['output'] === 'string' && result['output'] !== '') {
+        parts.push(result['output']);
+    }
+    return parts.join(String.fromCharCode(10));
+}
+/** legacy 工具结果 content（形状不固定，容错取 text）→ 文本 */
+function legacyResultText(result) {
+    const parts = [];
+    if (Array.isArray(result['content'])) {
+        for (const item of result['content']) {
+            if (typeof item === 'string' && item !== '')
+                parts.push(item);
+            else if (isRecord(item) && typeof item['text'] === 'string' && item['text'] !== '')
+                parts.push(item['text']);
+        }
+    }
+    const output = result['output'];
+    if (typeof output === 'string' && output !== '')
+        parts.push(output);
+    else if (isRecord(output) && typeof output['text'] === 'string' && output['text'] !== '')
+        parts.push(output['text']);
+    return parts.join(String.fromCharCode(10));
+}
+/** v0.3.0 的 ToolUse → tool_call 块（input 是 `{type,value}`，raw_input 是原始 JSON 文本） */
+function v3ToolCall(toolUse) {
+    if (!isRecord(toolUse))
+        return null;
+    const id = pickString(toolUse, ['id']) ?? '';
+    const name = pickString(toolUse, ['name']) ?? '';
+    if (id === '' || name === '')
+        return null;
+    const rawInput = toolUse['raw_input'];
+    if (typeof rawInput === 'string' && rawInput !== '')
+        return irToolCallBlock(id, name, toolInputOfText(rawInput));
+    const wrapped = toolUse['input'];
+    if (isRecord(wrapped)) {
+        const value = wrapped['type'] === 'text' && typeof wrapped['value'] === 'string'
+            ? toolInputOfText(wrapped['value'])
+            : (wrapped['value'] ?? {});
+        return irToolCallBlock(id, name, value);
+    }
+    return irToolCallBlock(id, name, {});
+}
+/** legacy 的 ToolUse.input 是对象或字符串 */
+function legacyToolCall(toolUse) {
+    if (!isRecord(toolUse))
+        return null;
+    const id = pickString(toolUse, ['id']) ?? '';
+    const name = pickString(toolUse, ['name']) ?? '';
+    if (id === '' || name === '')
+        return null;
+    const input = toolUse['input'];
+    return irToolCallBlock(id, name, typeof input === 'string' ? toolInputOfText(input) : (input ?? {}));
+}
+/** v0.3.0 消息单元（serde 外部标签：{User:…} / {Agent:…} / {Compaction:…} / 裸字符串 "Resume"） */
+function zedV3Records(messages, acc) {
+    const out = [];
+    for (const msg of messages) {
+        if (typeof msg === 'string')
+            continue; // unit 变体 "Resume"：UI 续聊标记
+        if (!isRecord(msg)) {
+            acc.bad++;
+            continue;
+        }
+        if (isRecord(msg['User'])) {
+            const blocks = [];
+            const content = msg['User']['content'];
+            for (const block of Array.isArray(content) ? content : []) {
+                if (isRecord(block) && typeof block['Text'] === 'string' && block['Text'] !== '') {
+                    blocks.push(irTextBlock(block['Text']));
+                }
+                else if (isRecord(block) && block['Image'] !== undefined) {
+                    irBump(acc.ignored, 'block:image');
+                }
+                else if (isRecord(block) && block['Mention'] !== undefined) {
+                    irBump(acc.ignored, 'block:mention');
+                }
+                else {
+                    irBump(acc.ignored, 'block:unmapped');
+                }
+            }
+            if (blocks.length > 0)
+                out.push({ role: 'user', blocks });
+            continue;
+        }
+        if (isRecord(msg['Agent'])) {
+            const agent = msg['Agent'];
+            const blocks = [];
+            const content = agent['content'];
+            for (const block of Array.isArray(content) ? content : []) {
+                if (isRecord(block) && typeof block['Text'] === 'string' && block['Text'] !== '') {
+                    blocks.push(irTextBlock(block['Text']));
+                }
+                else if (isRecord(block) && isRecord(block['Thinking'])
+                    && typeof block['Thinking']['text'] === 'string' && block['Thinking']['text'] !== '') {
+                    // IR 只有 text/tool_call/tool_result（reasoning 需共享层改动才能保留）→ 逐类计数
+                    irBump(acc.ignored, 'block:thinking');
+                }
+                else if (isRecord(block) && block['ToolUse'] !== undefined) {
+                    const call = v3ToolCall(block['ToolUse']);
+                    if (call === null)
+                        irBump(acc.ignored, 'block:tool-use-invalid');
+                    else
+                        blocks.push(call);
+                }
+                else {
+                    irBump(acc.ignored, 'block:unmapped');
+                }
+            }
+            // 同一 Agent 消息上的 tool_results 对象（键 = tool_use_id）；合成器只把 tool_result 归入
+            // **已打开的 step** → 拆到用户侧记录后再挂回（splitToolResults）
+            const results = agent['tool_results'];
+            if (isRecord(results)) {
+                for (const [toolUseId, result] of Object.entries(results)) {
+                    if (!isRecord(result)) {
+                        irBump(acc.ignored, 'block:unmapped');
+                        continue;
+                    }
+                    blocks.push(irToolResultBlock(toolUseId, v3ResultText(result, acc), result['is_error'] === true));
+                }
+            }
+            if (blocks.length > 0)
+                out.push(...splitToolResults('assistant', blocks, {}));
+            continue;
+        }
+        if (isRecord(msg['Compaction'])) {
+            // 压缩落点：本地合成器没有原生压缩检查点（需共享层改动）→ 显式计数
+            irBump(acc.ignored, 'message:compaction');
+            continue;
+        }
+        irBump(acc.ignored, 'message:unmapped');
+    }
+    return out;
+}
+/** legacy 方言（role + segments/tool_uses/tool_results/is_visible） */
+function zedLegacyRecords(messages, acc) {
+    const out = [];
+    for (const msg of messages) {
+        if (!isRecord(msg)) {
+            acc.bad++;
+            continue;
+        }
+        if (msg['is_visible'] === false) {
+            irBump(acc.ignored, 'message:invisible');
+            continue;
+        }
+        const role = (pickString(msg, ['role']) ?? '').toLowerCase();
+        if (role === 'system') {
+            irBump(acc.ignored, 'role:system');
+            continue;
+        }
+        const blocks = [];
+        const segments = msg['segments'];
+        for (const seg of Array.isArray(segments) ? segments : []) {
+            if (isRecord(seg) && seg['type'] === 'text' && typeof seg['text'] === 'string' && seg['text'] !== '') {
+                blocks.push(irTextBlock(seg['text']));
+            }
+            else if (isRecord(seg) && (seg['type'] === 'thinking' || seg['type'] === 'RedactedThinking')) {
+                irBump(acc.ignored, 'block:' + String(seg['type']));
+            }
+            else if (isRecord(seg)) {
+                irBump(acc.ignored, 'block:unmapped');
+            }
+        }
+        if (role === 'assistant') {
+            for (const toolUse of Array.isArray(msg['tool_uses']) ? msg['tool_uses'] : []) {
+                const call = legacyToolCall(toolUse);
+                if (call === null)
+                    irBump(acc.ignored, 'block:tool-use-invalid');
+                else
+                    blocks.push(call);
+            }
+        }
+        if (Array.isArray(msg['tool_results'])) {
+            for (const result of msg['tool_results']) {
+                if (!isRecord(result))
+                    continue;
+                const id = pickString(result, ['tool_use_id', 'toolUseId']) ?? '';
+                if (id === '') {
+                    irBump(acc.ignored, 'block:tool-result-invalid');
+                    continue;
+                }
+                blocks.push(irToolResultBlock(id, legacyResultText(result), result['is_error'] === true));
+            }
+        }
+        if (role === 'user' || role === 'assistant') {
+            if (blocks.length === 0)
+                continue;
+            const pushed = splitToolResults(role, blocks, {});
+            // 用户消息同时带正文与结果：结果记录排在正文之前（同 read-goose.ts 的理由）
+            if (role === 'user' && pushed.length > 1)
+                pushed.reverse();
+            out.push(...pushed);
+        }
+        else if (role !== '') {
+            irBump(acc.ignored, 'role:' + role);
+        }
+    }
+    return out;
+}
+/** 线程 JSON → 归一记录（v0.3.0 外部标签方言优先；其余走 legacy） */
+function zedMessagesOf(thread, acc) {
+    if (!isRecord(thread)) {
+        acc.bad++;
+        return [];
+    }
+    const raw = thread['messages'];
+    if (!Array.isArray(raw)) {
+        acc.bad++;
+        return [];
+    }
+    const version = typeof thread['version'] === 'string' ? thread['version'] : '';
+    return version === '0.3.0' ? zedV3Records(raw, acc) : zedLegacyRecords(raw, acc);
+}
+/** cwd：行级 folder_paths（`\n` 分隔 + order 索引）的首项 → 线程 JSON 里的 cwd 字段（缺 → undefined） */
+function zedCwdOf(row, thread) {
+    const paths = zedFolderPaths(pickString(row, ZED_FOLDER_KEYS), pickString(row, ZED_FOLDER_ORDER_KEYS));
+    if (paths.length > 0)
+        return paths[0];
+    if (isRecord(thread)) {
+        return pickString(thread, ['cwd', 'working_dir', 'working_directory', 'directory', 'root', 'project_path']);
+    }
+    return undefined;
+}
+/**
+ * 打开着的库 → 每个 thread 一个归一文件；**返回 null ⇔ 读不到 / 不是 zed 库**。
+ *
+ * 签名判定：必须存在 `threads` 表，且它有 `id` 与 `data` 列（竞品同款「靠结构自证」）。
+ * `parent_id` 非空的行是子代理线程（上游 UI/ThreadStore 都过滤）→ 跳过并计数。
+ */
+export function readZedDatabase(db, label) {
+    const tables = db.tables();
+    if (tables === null)
+        return null;
+    if (!tables.includes('threads'))
+        return null;
+    const cols = db.columns('threads');
+    if (cols === null)
+        return null;
+    if (!cols.includes('id') || !cols.includes('data'))
+        return null;
+    const rows = db.all('SELECT * FROM "threads"');
+    if (rows === null)
+        return null;
+    const files = [];
+    const skipped = [];
+    let noId = 0;
+    let unreadable = 0;
+    let zstdMissing = 0;
+    let subagent = 0;
+    const hasParent = ZED_PARENT_KEYS.some((key) => cols.includes(key));
+    for (const row of rows) {
+        const id = pickString(row, ZED_ID_KEYS);
+        if (id === undefined) {
+            noId++;
+            continue;
+        }
+        if (hasParent && pickString(row, ZED_PARENT_KEYS) !== undefined) {
+            subagent++;
+            continue;
+        }
+        const decoded = decodeThread(row);
+        if (!decoded.ok) {
+            unreadable++;
+            if (decoded.problem === 'zstd-unavailable')
+                zstdMissing++;
+            continue;
+        }
+        const acc = { ignored: {}, bad: 0 };
+        const records = zedMessagesOf(decoded.value, acc);
+        const rowTitle = pickString(row, ZED_TITLE_KEYS) ?? '';
+        const blobTitle = isRecord(decoded.value) ? (pickString(decoded.value, ZED_TITLE_KEYS) ?? '') : '';
+        const title = rowTitle !== '' ? rowTitle : blobTitle;
+        const createdAt = pickTime(row, ZED_TIME_KEYS);
+        files.push({
+            id,
+            parsed: {
+                records,
+                cwd: zedCwdOf(row, decoded.value),
+                createdAt,
+                title: title !== '' ? title : firstUserText(records),
+                raw: records.length,
+                bad: acc.bad,
+                ignored: acc.ignored,
+            },
+        });
+    }
+    if (noId > 0)
+        skipped.push({ ...sqliteSkip(label, 'thread-without-id'), count: noId });
+    if (zstdMissing > 0)
+        skipped.push({ ...sqliteSkip(label, 'zstd-unavailable'), count: zstdMissing });
+    const other = unreadable - zstdMissing;
+    if (other > 0)
+        skipped.push({ ...sqliteSkip(label, 'thread-unreadable'), count: other });
+    return {
+        files,
+        skipped,
+        counts: { threads: files.length, 'threads.unreadable': unreadable, 'threads.subagent': subagent },
+    };
+}
+/** 读盘入口（wiring 的 `read`） */
+export async function readZed(opts) {
+    const dbFile = zedThreadsDbPath(opts);
+    const label = labelForPath(opts.homeDir, dbFile);
+    if (!(await isFile(dbFile)))
+        return { files: [], readFindings: [], extraCounts: {} };
+    const gate = await sqliteGate(label, opts.sqliteDeps);
+    if (gate !== undefined)
+        return { files: [], readFindings: [gate], extraCounts: {} };
+    const opened = await sqliteOpen(dbFile, opts.sqliteDeps);
+    if (opened.db === null)
+        return { files: [], readFindings: [sqliteSkip(label, openDetail(opened.problem))], extraCounts: {} };
+    const db = opened.db;
+    try {
+        const read = readZedDatabase(db, label);
+        if (read === null)
+            return { files: [], readFindings: [sqliteSkip(label, 'shape-mismatch')], extraCounts: {} };
+        return {
+            files: read.files,
+            readFindings: read.skipped,
+            extraCounts: { 'zed.sessions': read.files.length, ...prefixCounts('zed', read.counts) },
+        };
+    }
+    finally {
+        db.close();
+    }
+}
+//# sourceMappingURL=read-zed.js.map

package/lib/foreign/reasonix.d.ts

@@ -0,0 +1,6 @@
+import type { SessionSourceWiring } from './session-source.ts';
+import type { ReasonixSessionFile } from './read-reasonix.ts';
+import type { ForeignSource } from './registry.ts';
+export declare const reasonixWiring: SessionSourceWiring<ReasonixSessionFile>;
+export declare function createReasonixSource(): ForeignSource;
+export declare const reasonixSource: ForeignSource;

package/lib/foreign/reasonix.js

@@ -0,0 +1,20 @@
+/**
+ * Reasonix 来源的翻译层（wiring + draftOf）。
+ *
+ * evidence 与 truth-table.ts 的 reasonix 行一致（fixture = 单测按同一份布局建夹具端到端跑；
+ * 本机无 ~/.reasonix 安装，真机未验证）。
+ */
+import { draftFromTranscript, sessionSourceOf } from './session-source.js';
+import { readReasonix, reasonixSessionRoots } from './read-reasonix.js';
+export const reasonixWiring = {
+    id: 'reasonix',
+    evidence: 'fixture',
+    probePaths: (opts) => reasonixSessionRoots(opts),
+    read: (opts) => readReasonix(opts),
+    draftOf: (file) => draftFromTranscript(file.id, file, 'reasonix'),
+};
+export function createReasonixSource() {
+    return sessionSourceOf(reasonixWiring);
+}
+export const reasonixSource = createReasonixSource();
+//# sourceMappingURL=reasonix.js.map