用户追加三条需求,全部落地: 1) 打开方式 = 宿主探测应用 + 插件自研 open-with(并存) - 逐字恢复 src/client/open-with.ts / open-with-settings.tsx / plugin-settings.ts / src/open-external.ts / open.external 路由 / api.openExternal / editor descriptor 的 settings.render,以及 20 个 openWith* 词条(19 份词典从 main 取回原文) - 恢复 menu-flip.ts + layout.css 子菜单翻转规则(子菜单回来了) - FileTree 菜单段(无标题行):宿主行 → 分隔线 → 插件 pinned + 「在应用中打开 >」 子菜单(图钉 / SSH 后缀原样)→ 分隔线 → 宿主 reveal;宿主可用时过滤插件 explorer 目标,宿主不可用时插件行照旧(SSH 远端场景保持可用) - 死键清理:菜单去掉标题行后 openInApp 词条无引用,已从 21 份词典删除 2) 文件页多选右键「压缩并打包下载」 - 新增 src/zip.ts(自研 ZIP:方法 0/8 按收益二选一、UTF-8 bit 11、目录条目、 自实现 CRC32、无第三方依赖)+ src/archive-route.ts(名净化 + 递归收集)+ GET /sidebar/archive(同一道 workspace fence、目录递归跳过符号链接、 10000 条目 / 256 MiB 上限)+ api.archiveUrl - FileTree:多选 ≥2 项或单选目录时出现;fetch → blob → objectURL 下载, 路由错误(超限 / fence / 非 GET)以 zipFailed 显示在错误条 - 修掉真实缺陷:central directory 只写 MS-DOS 位而 version-made-by 声称 Unix, Info-ZIP 会把每个文件解成 mode 000;改为 Unix 语义(0o100644 * 0x10000, << 16 会溢出 int32)。系统 unzip 复验:mode 644/755、unzip -t 无错 3) 文件变动页层级树 + 视觉打磨 - 新增 src/client/changes/change-tree.ts:单子目录链压缩成一行、目录先于文件、 大小写不敏感排序、文件带 GitFileStatus、目录带下级变更数 - GitLens 两组各自渲染树:折叠状态按「组+路径」独立、默认展开;分组头吸顶 + 计数胶囊;Stage all/Unstage all 改为带 aria-label 的 IconButton;选中 = interactive-bg-active + 2px 内嵌强调条;12px/级缩进 + l1 参考线; 历史行 hash 徽标 / ref 胶囊 / 作者·时间右对齐 验证:typecheck / lint / test(134 文件 1422 通过)/ build 全绿; DSH_CMD='npx -y --package @deepseek-ai/dsh@0.1.7-rc.1 dsh' pnpm test:mount 7/7 通过; 真实宿主截图确认三条需求(多选+双源菜单+zip 项、变动页树形层级)
296 lines
13 KiB
TypeScript
296 lines
13 KiB
TypeScript
/**
|
|
* Changes tab data layer: event-log folding into file operations, the
|
|
* directory tree the Git lens renders its groups through, and a sanity pass
|
|
* over the shared diff/highlight engines (their full behavior suites live in
|
|
* the standalone dsh-file-trace plugin).
|
|
*/
|
|
import { describe, expect, it } from 'vitest'
|
|
import { createToolResultMessage, ToolCallId } from '@deepseek-ai/dsh-llm'
|
|
import { extractFileOps, groupByFile, knownContentBefore, parseReadContent, parseReadLines } from '../src/client/changes/ops.ts'
|
|
import { buildChangeTree, type ChangeNode } from '../src/client/changes/change-tree.ts'
|
|
import { diffLines, buildDiffSegments, coalesceInline, diffInline } from '../src/client/diff/rows.ts'
|
|
import { langOfPath, scanLine } from '../src/client/diff/highlight.ts'
|
|
import type { GitStatusEntry } from '../src/client/api.ts'
|
|
import type { SidebarSessionEvent } from '../src/context-types.ts'
|
|
|
|
/** One synthetic session event. */
|
|
function ev(type: string, seq: number, time: number, data: Record<string, unknown>): SidebarSessionEvent {
|
|
return { type, seq, time, data }
|
|
}
|
|
|
|
/** A tool/call event. */
|
|
function call(seq: number, name: string, callId: string, args: unknown, time = seq): SidebarSessionEvent {
|
|
return ev('tool/call', seq, time, { name, callId, arguments: JSON.stringify(args) })
|
|
}
|
|
|
|
/** A tool/result event carrying one first-class tool-role message (the 0.1.7
|
|
* shape: result blocks and `isError` at the message's own top level). */
|
|
function result(seq: number, callId: string, text: string, isError = false, time = seq): SidebarSessionEvent {
|
|
return ev('tool/result', seq, time, {
|
|
message: {
|
|
role: 'tool',
|
|
source: { kind: 'tool', callId },
|
|
toolCallId: callId,
|
|
isError,
|
|
content: [{ type: 'text', text }],
|
|
},
|
|
})
|
|
}
|
|
|
|
/** The retired 0.1.6 shape: a role-'user' message wrapping the result in one
|
|
* `tool-result` content block. Historical logs still carry it. */
|
|
function legacyResult(seq: number, callId: string, text: string, isError = false, time = seq): SidebarSessionEvent {
|
|
return ev('tool/result', seq, time, {
|
|
message: {
|
|
role: 'user',
|
|
source: { kind: 'tool', callId },
|
|
content: [{ type: 'tool-result', isError, content: [{ type: 'text', text }] }],
|
|
},
|
|
})
|
|
}
|
|
|
|
describe('parseReadContent', () => {
|
|
it('strips the envelope and line-number prefixes but keeps blank lines', () => {
|
|
const raw = [
|
|
'<path>notes.md</path>',
|
|
'<type>file</type>',
|
|
'(Showing lines 1 to 4)',
|
|
'<content>',
|
|
'1: # Title',
|
|
'2: ',
|
|
'3: para one',
|
|
'4: ',
|
|
'</content>',
|
|
].join('\n')
|
|
expect(parseReadContent(raw)).toBe('# Title\n\npara one')
|
|
})
|
|
|
|
it('returns the raw text unchanged when no <content> section is present', () => {
|
|
expect(parseReadContent('plain text')).toBe('plain text')
|
|
})
|
|
})
|
|
|
|
describe('extractFileOps', () => {
|
|
it('seeds running ops from tool/call and settles them from tool/result', () => {
|
|
const ops = extractFileOps([
|
|
call(1, 'write', 'w1', { file_path: 'a.ts', content: 'body' }),
|
|
result(2, 'w1', '<path>a.ts</path>\n<type>file</type>\n<content>\nbody\n</content>'),
|
|
])
|
|
expect(ops).toHaveLength(1)
|
|
expect(ops[0]!.running).toBe(false)
|
|
expect(ops[0]!.isError).toBe(false)
|
|
expect(ops[0]!.content).toBe('body')
|
|
})
|
|
|
|
it('captures the error text of an errored result for every kind', () => {
|
|
const ops = extractFileOps([
|
|
call(1, 'write', 'w', { file_path: 'a', content: 'x' }),
|
|
result(2, 'w', '写入失败:目标只读', true),
|
|
call(3, 'read', 'r', { file_path: 'b' }),
|
|
result(4, 'r', 'cannot read "b": not found', true),
|
|
])
|
|
expect(ops.every(op => op.isError && op.errorText !== undefined)).toBe(true)
|
|
expect(ops.find(op => op.kind === 'read')?.errorText).toBe('cannot read "b": not found')
|
|
})
|
|
|
|
it('keeps a call without result as running, and reads capture their content', () => {
|
|
const ops = extractFileOps([
|
|
call(1, 'read', 'r1', { file_path: 'a.ts' }),
|
|
result(2, 'r1', '<path>a.ts</path>\n<type>file</type>\n<content>\n1: hello\n\n(End of file - total 1 lines)\n</content>'),
|
|
call(3, 'edit', 'e1', { file_path: 'c.ts', old_string: 'x', new_string: 'y' }),
|
|
])
|
|
const read = ops.find(op => op.callId === 'r1')
|
|
expect(read?.running).toBe(false)
|
|
expect(read?.read).toContain('1: hello')
|
|
const edit = ops.find(op => op.callId === 'e1')
|
|
expect(edit?.running).toBe(true)
|
|
expect(edit?.edit).toEqual({ oldString: 'x', newString: 'y' })
|
|
})
|
|
|
|
it('ignores non-file tools and unrelated results, newest first', () => {
|
|
const ops = extractFileOps([
|
|
call(1, 'pwsh', 'p', { command: 'ls' }),
|
|
result(2, 'unknown-call', 'orphan'),
|
|
call(3, 'read', 'r', { file_path: 'z.ts' }),
|
|
])
|
|
expect(ops.map(op => op.callId)).toEqual(['r'])
|
|
})
|
|
|
|
it('reads the tool/result message dsh-llm actually produces (producer round-trip)', () => {
|
|
const message = createToolResultMessage({
|
|
callId: ToolCallId('r'),
|
|
content: [{ type: 'text', text: 'produced file body' }],
|
|
isError: false,
|
|
})
|
|
const ops = extractFileOps([
|
|
call(1, 'read', 'r', { file_path: 'a.ts' }),
|
|
ev('tool/result', 2, 2, { message: message as unknown as Record<string, unknown> }),
|
|
])
|
|
expect(ops[0]!.running).toBe(false)
|
|
expect(ops[0]!.read).toBe('produced file body')
|
|
})
|
|
|
|
it('reads a LEGACY 0.1.6-era tool-result wrapper (historical logs keep that shape)', () => {
|
|
const ops = extractFileOps([
|
|
// No content in the model's arguments, so the settled result supplies it.
|
|
call(1, 'write', 'w', { file_path: 'a' }),
|
|
legacyResult(2, 'w', 'legacy write detail'),
|
|
call(3, 'read', 'r', { file_path: 'b' }),
|
|
legacyResult(4, 'r', 'cannot read "b": not found', true),
|
|
])
|
|
expect(ops.find(op => op.kind === 'write')?.content).toBe('legacy write detail')
|
|
expect(ops.find(op => op.kind === 'read')?.errorText).toBe('cannot read "b": not found')
|
|
})
|
|
|
|
it('groups by file newest-first and recovers prior write content', () => {
|
|
const ops = extractFileOps([
|
|
call(1, 'write', 'w1', { file_path: 'a.ts', content: 'first' }),
|
|
call(2, 'write', 'w2', { file_path: 'a.ts', content: 'second' }),
|
|
call(3, 'write', 'w3', { file_path: 'b.ts', content: 'x' }),
|
|
])
|
|
expect([...groupByFile(ops).keys()]).toEqual(['b.ts', 'a.ts'])
|
|
expect(knownContentBefore(ops, 'a.ts', ops.find(op => op.callId === 'w2')!)).toBe('first')
|
|
})
|
|
})
|
|
|
|
describe('parseReadLines', () => {
|
|
it('recovers real line numbers from the read envelope', () => {
|
|
const lines = parseReadLines('<path>a.ts</path>\n<type>file</type>\n<content>\n10: x\n11: y\n\n(Showing lines 10-11 of 20. Use offset=12 to continue.)\n</content>')
|
|
expect(lines).toEqual([{ line: 10, text: 'x' }, { line: 11, text: 'y' }])
|
|
})
|
|
})
|
|
|
|
describe('change tree (the Git lens reads its groups through it)', () => {
|
|
/** One `git status` row from a path and its porcelain code. */
|
|
const entry = (path: string, xy = ' M'): GitStatusEntry => ({ path, xy })
|
|
|
|
/** A compact, assertion-friendly projection of one tree. */
|
|
const shape = (nodes: readonly ChangeNode[]): unknown[] => nodes.map(node => node.kind === 'dir'
|
|
? { dir: node.name, path: node.path, changes: node.changes, children: shape(node.children) }
|
|
: { file: node.name, path: node.path, letter: node.status.letter, tone: node.status.tone })
|
|
|
|
it('keeps a single root-level file as one leaf', () => {
|
|
expect(shape(buildChangeTree([entry('readme.md')]))).toEqual([
|
|
{ file: 'readme.md', path: 'readme.md', letter: 'M', tone: 'modified' },
|
|
])
|
|
})
|
|
|
|
it('nests a deep path and compresses its single-child chain into one row', () => {
|
|
const tree = buildChangeTree([entry('src/client/changes/a.ts')])
|
|
expect(shape(tree)).toEqual([
|
|
{
|
|
dir: 'src/client/changes',
|
|
path: 'src/client/changes',
|
|
changes: 1,
|
|
children: [{ file: 'a.ts', path: 'src/client/changes/a.ts', letter: 'M', tone: 'modified' }],
|
|
},
|
|
])
|
|
})
|
|
|
|
it('stops compressing where the path branches or a directory holds its own file', () => {
|
|
// Two child directories: 'src' keeps its own row.
|
|
expect(shape(buildChangeTree([entry('src/a/x.ts'), entry('src/b/y.ts')]))).toEqual([
|
|
{
|
|
dir: 'src',
|
|
path: 'src',
|
|
changes: 2,
|
|
children: [
|
|
{ dir: 'a', path: 'src/a', changes: 1, children: [{ file: 'x.ts', path: 'src/a/x.ts', letter: 'M', tone: 'modified' }] },
|
|
{ dir: 'b', path: 'src/b', changes: 1, children: [{ file: 'y.ts', path: 'src/b/y.ts', letter: 'M', tone: 'modified' }] },
|
|
],
|
|
},
|
|
])
|
|
// A file AT this level: 'a' keeps its own row (and its file leads the child).
|
|
expect(shape(buildChangeTree([entry('a/x.ts'), entry('a/b/y.ts')]))).toEqual([
|
|
{
|
|
dir: 'a',
|
|
path: 'a',
|
|
changes: 2,
|
|
children: [
|
|
{ dir: 'b', path: 'a/b', changes: 1, children: [{ file: 'y.ts', path: 'a/b/y.ts', letter: 'M', tone: 'modified' }] },
|
|
{ file: 'x.ts', path: 'a/x.ts', letter: 'M', tone: 'modified' },
|
|
],
|
|
},
|
|
])
|
|
})
|
|
|
|
it('orders directories before files, each by name case-insensitively', () => {
|
|
const tree = buildChangeTree([
|
|
entry('Zed.ts'), entry('B.ts'), entry('a.ts'), entry('zdir/f.ts'), entry('Adir/f.ts'),
|
|
])
|
|
expect(shape(tree).map(node => (node as { dir?: string; file?: string }).dir ?? (node as { file: string }).file))
|
|
.toEqual(['Adir', 'zdir', 'a.ts', 'B.ts', 'Zed.ts'])
|
|
})
|
|
|
|
it('mixes root-level files with nested paths and keeps same-named files apart', () => {
|
|
expect(shape(buildChangeTree([
|
|
entry('b/index.ts'), entry('index.ts'), entry('a/index.ts'),
|
|
]))).toEqual([
|
|
{ dir: 'a', path: 'a', changes: 1, children: [{ file: 'index.ts', path: 'a/index.ts', letter: 'M', tone: 'modified' }] },
|
|
{ dir: 'b', path: 'b', changes: 1, children: [{ file: 'index.ts', path: 'b/index.ts', letter: 'M', tone: 'modified' }] },
|
|
{ file: 'index.ts', path: 'index.ts', letter: 'M', tone: 'modified' },
|
|
])
|
|
})
|
|
|
|
it('carries every file porcelain status and counts the files under each row', () => {
|
|
const tree = buildChangeTree([
|
|
entry('src/a.ts', ' M'), entry('src/new.ts', '??'), entry('src/gone.ts', ' D'), entry('src/kept.ts', ' '),
|
|
])
|
|
expect(shape(tree)).toEqual([
|
|
{
|
|
dir: 'src',
|
|
path: 'src',
|
|
changes: 3,
|
|
children: [
|
|
{ file: 'a.ts', path: 'src/a.ts', letter: 'M', tone: 'modified' },
|
|
{ file: 'gone.ts', path: 'src/gone.ts', letter: 'D', tone: 'deleted' },
|
|
{ file: 'new.ts', path: 'src/new.ts', letter: 'U', tone: 'untracked' },
|
|
],
|
|
},
|
|
])
|
|
})
|
|
|
|
it('returns no nodes for an empty or all-clean list', () => {
|
|
expect(buildChangeTree([])).toEqual([])
|
|
expect(buildChangeTree([entry('a.ts', ' '), entry('b.ts', '!!')])).toEqual([])
|
|
})
|
|
|
|
it('counts a repeated path once', () => {
|
|
expect(shape(buildChangeTree([entry('a.ts'), entry('a.ts', '??')]))).toEqual([
|
|
{ file: 'a.ts', path: 'a.ts', letter: 'M', tone: 'modified' },
|
|
])
|
|
})
|
|
})
|
|
|
|
describe('ported diff/highlight engines (sanity)', () => {
|
|
it('diffLines pairs a rewrite as mod and folds context into hunks', () => {
|
|
const rows = diffLines('hello world', 'hello dsh')
|
|
expect(rows.map(r => r.kind)).toEqual(['mod', 'mod'])
|
|
const before = Array.from({ length: 6 }, (_, i) => ({ kind: 'context' as const, oldLine: i + 1, newLine: i + 1, text: 'c' + String(i) }))
|
|
const change = [{ kind: 'mod' as const, oldLine: 7, newLine: 7, text: 'X' }]
|
|
const after = Array.from({ length: 8 }, (_, i) => ({ kind: 'context' as const, oldLine: i + 8, newLine: i + 8, text: 'c' + String(i + 8) }))
|
|
const segments = buildDiffSegments([...before, ...change, ...after], 3)
|
|
expect(segments.map(s => s.kind)).toEqual(['fold', 'hunk', 'fold'])
|
|
})
|
|
|
|
it('coalesces the ported prefix/suffix inline diff into runs', () => {
|
|
const { old: oldSide } = diffInline('hello world', 'hello dsh')
|
|
expect(coalesceInline(oldSide)).toEqual([
|
|
{ text: 'hello ', changed: false },
|
|
{ text: 'world', changed: true },
|
|
])
|
|
})
|
|
|
|
it('highlights keywords and threads block-comment state', () => {
|
|
expect(langOfPath('a.cpp')).toBe('cpp')
|
|
const open = scanLine('/* header', 'cpp')
|
|
expect(open.inBlock).toBe(true)
|
|
const mid = scanLine(' * interior note', 'cpp', true)
|
|
expect(mid.tokens).toEqual([{ text: ' * interior note', type: 'comment' }])
|
|
const code = scanLine('int main() { return 42; } // done', 'cpp')
|
|
expect(code.tokens).toContainEqual({ text: 'int', type: 'keyword' })
|
|
expect(code.tokens).toContainEqual({ text: '42', type: 'number' })
|
|
expect(code.tokens).toContainEqual({ text: '// done', type: 'comment' })
|
|
})
|
|
})
|