✨ feat(vscode): 代码阅读支持 Dockerfile:解析指令、每个 FROM 一个阶段(阶段名能跳过去)、RUN 交给 shell 按命令拆、复制、构建参数 / 环境变量、设置、标签、健康检查、启动,连着写的 ENV / ARG / COPY / LABEL 并成一块加一张表;语法包取自 npm 包 tree-sitter-wasm,打包时复制进扩展

This commit is contained in:
cruldra committed 2026-09-28 11:54:01 +08:00
1 parent 0665f35821
commit d4ec0e7de4
11 files changed
+1619 -4

No files matched your search

+6
View File
@@ -4,6 +4,7 @@ import { createRequire } from 'node:module'
import path from 'node:path'
import process from 'node:process'
import * as esbuild from 'esbuild'
import { getWasmPath } from 'tree-sitter-wasm'
const production = process.argv.includes('--production')
const watch = process.argv.includes('--watch')
@@ -33,12 +34,17 @@ const TREE_SITTER_WASM = [
'tree-sitter-bash.wasm',
]
// @vscode/tree-sitter-wasm 没有的语法包,从 tree-sitter-wasm(预编译、签过名的各家语法包)拿:语言名 → 它的 wasm
const EXTRA_GRAMMARS = ['dockerfile']
function copyTreeSitterWasm() {
const from = path.dirname(createRequire(import.meta.url).resolve('@vscode/tree-sitter-wasm'))
const to = 'dist/tree-sitter'
mkdirSync(to, { recursive: true })
for (const file of TREE_SITTER_WASM)
copyFileSync(path.join(from, file), path.join(to, file))
for (const language of EXTRA_GRAMMARS)
copyFileSync(getWasmPath(language), path.join(to, `tree-sitter-${language}.wasm`))
}
async function main() {
+3 -2
View File
@@ -320,7 +320,7 @@
"editor/title": [
{
"command": "superpowers.codeReader.showPreview",
"when": "activeEditor == workbench.editors.files.textFileEditor && (resourceLangId =~ /^(typescript|typescriptreact|javascript|javascriptreact|python|shellscript)$/ || resourceLangId in superpowers.codeReader.languageIds)",
"when": "activeEditor == workbench.editors.files.textFileEditor && (resourceLangId =~ /^(typescript|typescriptreact|javascript|javascriptreact|python|shellscript|dockerfile)$/ || resourceLangId in superpowers.codeReader.languageIds)",
"group": "navigation"
},
{
@@ -340,7 +340,7 @@
},
{
"command": "superpowers.codeReader.showPreview",
"when": "activeEditor == workbench.editors.files.textFileEditor && (resourceLangId =~ /^(typescript|typescriptreact|javascript|javascriptreact|python|shellscript)$/ || resourceLangId in superpowers.codeReader.languageIds)"
"when": "activeEditor == workbench.editors.files.textFileEditor && (resourceLangId =~ /^(typescript|typescriptreact|javascript|javascriptreact|python|shellscript|dockerfile)$/ || resourceLangId in superpowers.codeReader.languageIds)"
},
{
"command": "superpowers.codeReader.showSource",
@@ -414,6 +414,7 @@
"esbuild": "^0.25.0",
"eslint": "^9.39.2",
"ovsx": "^0.10.7",
"tree-sitter-wasm": "^2.0.2",
"typescript": "^5.9.3",
"vitest": "^4.0.16"
}
+8
View File
@@ -59,6 +59,9 @@ importers:
ovsx:
specifier: ^0.10.7
version: 0.10.7
tree-sitter-wasm:
specifier: ^2.0.2
version: 2.0.2
typescript:
specifier: ^5.9.3
version: 5.9.3
@@ -3394,6 +3397,9 @@ packages:
resolution: {integrity: sha512-L0Orpi8qGpRG//Nd+H90vFB+3iHnue1zSSGmNOOCh1GLJ7rUKVwV2HvijphGQS2UmhUZewS9VgvxYIdgr+fG1A==}
hasBin: true
tree-sitter-wasm@2.0.2:
resolution: {integrity: sha512-zxC/EcugWmdytml6hZoU7vn1olencvNDSluO3BOyYsPWPXkBfbjmvNriJCHght+vXMYsLJZhK9v9e+YBggV4kQ==}
ts-api-utils@2.4.0:
resolution: {integrity: sha512-3TaVTaAv2gTiMB35i3FiGJaRfwb3Pyn/j3m/bfAvGe8FB7CF6u+LMYqYlDh7reQf7UNvoTvdfAqHGmPGOSsPmA==}
engines: {node: '>=18.12'}
@@ -7205,6 +7211,8 @@ snapshots:
tree-kill@1.2.2: {}
tree-sitter-wasm@2.0.2: {}
ts-api-utils@2.4.0(typescript@5.9.3):
dependencies:
typescript: 5.9.3
@@ -1,15 +1,18 @@
import type { Memento } from 'vscode'
import type { LibraryRule } from '../contract'
import { describe, expect, it } from 'vitest'
import { packageCommandRules } from '../commands/packageCommands'
import { shellCommandRules } from '../commands/shellCommands'
import { createBashAdapter } from '../languages/bash'
import { createDockerfileAdapter } from '../languages/dockerfile'
import { createPythonAdapter } from '../languages/python'
import { createTypeScriptAdapter } from '../languages/typescript'
import { CodeReaderRegistry } from '../registry'
import { fastapiRule } from '../rules/fastapi'
import { sqlalchemyRule } from '../rules/sqlalchemy'
import { runtime, WASM_DIR } from '../testSupport'
import { DOCKERFILE_WASM_DIR, runtime, WASM_DIR } from '../testSupport'
import { bashZhTranslator } from '../translators/bashZh'
import { dockerfileZhTranslator } from '../translators/dockerfileZh'
import { pythonZhTranslator } from '../translators/pythonZh'
import { typescriptZhTranslator } from '../translators/typescriptZh'
import { documentMessage } from './documentMessage'
@@ -168,6 +171,24 @@ describe('预览的 document 消息', () => {
expect(symbols.map(symbol => symbol.occurrences.map(range => source.slice(range.startIndex, range.endIndex)))).toEqual([['dest', 'dest']])
})
it('dockerfile:RUN 里的 shell 交给注册的 shell 语言插件、翻译插件和命令规则读,名字连同 shell 里的一起交给宿主跳转', async () => {
const registry = new CodeReaderRegistry({ get: () => undefined, update: async () => {} } as Pick<Memento, 'get' | 'update'>)
registry.registerLanguage(createBashAdapter(WASM_DIR))
registry.registerLanguage(createDockerfileAdapter(DOCKERFILE_WASM_DIR))
registry.registerTranslator(bashZhTranslator)
registry.registerTranslator(dockerfileZhTranslator)
for (const rule of [...shellCommandRules, ...packageCommandRules])
registry.registerCommandRule(rule)
const source = 'FROM node:22 AS deps\nRUN x=1 && echo "$x" && apt-get update\nCMD ["node", "server.js"]\n'
const { message, symbols } = await documentMessage(doc('dockerfile', source), registry, runtime, ignore)
const [stage] = message.blocks
expect(stage.kind).toBe('stage')
expect(stage.children.map(child => [child.kind, child.children.map(step => step.kind)])).toEqual([['run', ['variable', 'command', 'command']], ['start', []]])
expect(stage.children[0].children[2]).toMatchObject({ translation: [{ text: '更新软件包列表' }] })
expect(message.attributes.find(attribute => attribute.id === 'kind')?.values.map(value => value.name)).toEqual(['阶段', '执行', '变量', '命令', '启动'])
expect(symbols.map(symbol => symbol.occurrences.map(range => source.slice(range.startIndex, range.endIndex)))).toEqual([['deps'], ['x', 'x']])
})
it('解析失败:把原因放进 error', async () => {
const failing = { parse: async () => Promise.reject(new Error('语法包加载失败')) }
const { message } = await documentMessage(doc('typescript', 'const a = 1'), registryWithTypeScript(), failing, ignore)
@@ -0,0 +1,110 @@
# ── 统一 Backend 镜像 ─────────────────────────────────────────────────
# 一个镜像同时承载 api / worker / beat / flower 四种角色;
# 由 docker-compose 传入不同 command 区分角色,避免重复构建/传输。
FROM python:3.12-slim AS base
LABEL maintainer="Cruldra" \
description="luffy-agent-platform - 统一 Backend 镜像 (API/Worker/Beat/Flower)"
# git: deployment 模块创建 bare repo;postgresql-client: alembic / psql 运维;curl: healthcheck
# postgresql-16-repack: lap checkpoint repack 的 CLI (与线上 PG16 大版本对齐).
# 扩展本体在目标库执行 CREATE EXTENSION pg_repack (运维一次性, 不在镜像构建时做).
# 官方 slim 默认源只有当前大版本 repack, 故加 PGDG 装 16 客户端.
RUN apt-get update && \
apt-get install -y --no-install-recommends \
ca-certificates curl gnupg postgresql-client git && \
install -d /usr/share/keyrings && \
curl -fsSL https://www.postgresql.org/media/keys/ACCC4CF8.asc \
| gpg --dearmor -o /usr/share/keyrings/postgresql.gpg && \
. /etc/os-release && \
echo "deb [signed-by=/usr/share/keyrings/postgresql.gpg] http://apt.postgresql.org/pub/repos/apt ${VERSION_CODENAME}-pgdg main" \
> /etc/apt/sources.list.d/pgdg.list && \
apt-get update && \
apt-get install -y --no-install-recommends postgresql-16-repack && \
pg_repack --version && \
rm -rf /var/lib/apt/lists/*
# Node.js:agent MCP 节点 (如 context7) 用 stdio + npx 拉子进程, api 容器必须有 npx。
# 不走 apt install nodejs (拉 ~250 MB Recommends), 从官方 node:22-slim 拷贝必要
# 文件; node:22-slim 同基于 debian:bookworm-slim, glibc 兼容, 二进制可直接跑。
# npm / npx 在官方镜像里是指向 node_modules/npm/bin/{npm-cli,npx-cli}.js 的 symlink,
# 拷贝 node_modules 后手建 symlink (COPY 单文件会 dereference, 不能保留 symlink).
COPY --from=node:22-slim /usr/local/bin/node /usr/local/bin/node
COPY --from=node:22-slim /usr/local/lib/node_modules /usr/local/lib/node_modules
RUN ln -sf /usr/local/lib/node_modules/npm/bin/npm-cli.js /usr/local/bin/npm && \
ln -sf /usr/local/lib/node_modules/npm/bin/npx-cli.js /usr/local/bin/npx
COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
# appuser 先建、/app /lib 先归属:.venv 由 appuser 直接创建,末尾无需 chown -R
# (chown -R 会在 overlayfs 里把 ~2GB venv 完整复制进新层:镜像翻倍,
# 且该层排在源码层后,每次代码变更都重建,打爆 registry 增量分发)
RUN useradd -m -s /bin/false appuser && \
mkdir -p /app /lib && \
chown appuser:appuser /app /lib
USER appuser
ENV UV_CACHE_DIR=/home/appuser/.cache/uv
WORKDIR /app
# ── 依赖层(仅随 pyproject/uv.lock 变化重建)───────────────────────────
# 只放元数据:backend 的 pyproject+lock,以及 ../lib/* 路径依赖的 pyproject
# (uv 解析 [tool.uv.sources] 需要路径存在)。源码后置,改代码不打爆本层,
# registry 分发时该 ~2GB 层可被远端复用,日常部署只传源码层。
COPY --chown=appuser:appuser backend/pyproject.toml backend/uv.lock ./
COPY --chown=appuser:appuser lib/agent-dsl/pyproject.toml /lib/agent-dsl/
COPY --chown=appuser:appuser lib/agent-compiler-deepagents/pyproject.toml /lib/agent-compiler-deepagents/
COPY --chown=appuser:appuser lib/vendor-tools/pyproject.toml /lib/vendor-tools/
COPY --chown=appuser:appuser lib/code-sandbox-core/pyproject.toml /lib/code-sandbox-core/
COPY --chown=appuser:appuser lib/code-sandbox-daytona/pyproject.toml /lib/code-sandbox-daytona/
COPY --chown=appuser:appuser lib/code-sandbox-e2b/pyproject.toml /lib/code-sandbox-e2b/
COPY --chown=appuser:appuser lib/code-sandbox-local/pyproject.toml /lib/code-sandbox-local/
COPY --chown=appuser:appuser lib/code-publish-core/pyproject.toml /lib/code-publish-core/
COPY --chown=appuser:appuser lib/code-publish-compose-runner/pyproject.toml /lib/code-publish-compose-runner/
COPY --chown=appuser:appuser lib/luffy-billing-sdk/pyproject.toml /lib/luffy-billing-sdk/
COPY --chown=appuser:appuser lib/luffy-observability/pyproject.toml /lib/luffy-observability/
# 三方依赖先装;本项目与 lib editable 包此时无源码,跳过安装留给第二次 sync
RUN uv sync --frozen --no-dev --no-install-project \
--no-install-package agent-dsl \
--no-install-package agent-compiler-deepagents \
--no-install-package vendor-tools \
--no-install-package code-sandbox-core \
--no-install-package code-sandbox-daytona \
--no-install-package code-sandbox-e2b \
--no-install-package code-sandbox-local \
--no-install-package code-publish-core \
--no-install-package code-publish-compose-runner \
--no-install-package luffy-billing-sdk \
--no-install-package luffy-observability
# ── 源码层(每次代码变更只重建这里往下)────────────────────────────────
COPY --chown=appuser:appuser lib /lib
# agent 定义 (vendor/agents/*/agent.yaml + nodes/*.py) 与代码是同一个版本, 必须随镜像原子切换;
# 路径与 settings 默认值 <_BASE_DIR>/../vendor/agents 对齐, 容器内即 /vendor/agents.
COPY --chown=appuser:appuser vendor /vendor
COPY --chown=appuser:appuser backend/src/ ./src/
COPY --chown=appuser:appuser backend/alembic/ ./alembic/
COPY --chown=appuser:appuser backend/alembic.ini ./
COPY --chown=appuser:appuser backend/scripts/ ./scripts/
# 补装本项目 + lib editable 包(三方依赖已就位,秒级)
RUN uv sync --frozen --no-dev
# 远程调试支持:装入 uv 创建的 venv,按需通过 DEBUGPY_ENABLE=1 开启。
# 必须在最后一次 uv sync 之后:sync 会精确对齐 lock,把 lock 外的包卸掉。
RUN uv pip install debugpy
ENV UV_NO_SYNC=1
# 部署 ID (Jenkins IMAGE_TAG): /health 回显、日志定位后端出自哪次构建. 选择性发布时 api 与 client 镜像 tag 可不同, 客户端是否过期以 client 镜像 (浏览器包 vs BFF) 为准.
ARG DEPLOYMENT_ID
ENV DEPLOYMENT_ID=$DEPLOYMENT_ID
EXPOSE 8000
# 默认 CMD 走 api,celery worker/beat/flower 由 compose 覆盖 command
HEALTHCHECK --interval=30s --timeout=5s --start-period=10s --retries=3 \
CMD curl -f http://localhost:8000/health || exit 1
CMD ["uv", "run", "uvicorn", "luffy_agent.main:app", "--host", "0.0.0.0", "--port", "8000"]
@@ -0,0 +1,41 @@
# syntax=docker/dockerfile:1.7
FROM node:22-alpine AS deps
WORKDIR /app
RUN corepack enable && corepack prepare pnpm@latest --activate
COPY package.json pnpm-lock.yaml ./
RUN pnpm install --frozen-lockfile --ignore-scripts
FROM node:22-alpine AS builder
WORKDIR /app
RUN corepack enable && corepack prepare pnpm@latest --activate
COPY --from=deps /app/node_modules ./node_modules
COPY . .
ENV NEXT_TELEMETRY_DISABLED=1
# 部署 ID 烙进本次构建: Next 据此给静态资源加 ?dpl=、页内跳转时比对版本;
# 浏览器包与服务端包都能读 process.env.NEXT_DEPLOYMENT_ID (见 src/lib/version-skew).
ARG DEPLOYMENT_ID
ENV NEXT_DEPLOYMENT_ID=$DEPLOYMENT_ID
RUN pnpm build
FROM node:22-alpine AS runner
WORKDIR /app
ENV NODE_ENV=production
ENV NEXT_TELEMETRY_DISABLED=1
ENV PORT=3000
ENV HOSTNAME=0.0.0.0
ARG DEPLOYMENT_ID
ENV NEXT_DEPLOYMENT_ID=$DEPLOYMENT_ID
RUN addgroup --system --gid 1001 nodejs && \
adduser --system --uid 1001 nextjs
COPY --from=builder --chown=nextjs:nodejs /app/.next/standalone ./
COPY --from=builder --chown=nextjs:nodejs /app/.next/static ./.next/static
# standalone 输出不含 public/, 必须显式拷贝 (字体等静态资源)
COPY --from=builder --chown=nextjs:nodejs /app/public ./public
USER nextjs
EXPOSE 3000
CMD ["node", "server.js"]
+7 -1
View File
@@ -3,7 +3,9 @@ import type { CodeReaderApi, RuleLocation, RuleReference } from './contract'
import type { IndexedName } from './nameResolution'
import { commands, window, workspace } from 'vscode'
import { logger } from '../logging/logger'
import { packageCommandRules } from './commands/packageCommands'
import { shellCommandRules } from './commands/shellCommands'
import { systemCommandRules } from './commands/systemCommands'
import { BlockFilter } from './editor/blockFilter'
import { NEXT_OCCURRENCE, PREVIOUS_OCCURRENCE } from './editor/occurrenceNavigation'
import { PREVIEW_VIEW_TYPE, PreviewEditorProvider } from './editor/previewEditor'
@@ -11,6 +13,7 @@ import { showPreview, showSource } from './editor/switchEditor'
import { registerCopyCommands } from './explorer/copyCommands'
import { FileTreeProvider } from './explorer/fileTree'
import { createBashAdapter } from './languages/bash'
import { createDockerfileAdapter } from './languages/dockerfile'
import { createPythonAdapter } from './languages/python'
import { createTypeScriptAdapter } from './languages/typescript'
import { resolveReference } from './libraryRules'
@@ -27,6 +30,7 @@ import { fastapiRule } from './rules/fastapi'
import { sqlalchemyRule } from './rules/sqlalchemy'
import { bundledWasmDir, TreeSitterRuntime } from './runtime'
import { bashZhTranslator } from './translators/bashZh'
import { dockerfileZhTranslator } from './translators/dockerfileZh'
import { pythonZhTranslator } from './translators/pythonZh'
import { typescriptZhTranslator } from './translators/typescriptZh'
import { workspaceIndexHost } from './workspaceFiles'
@@ -115,10 +119,12 @@ export function registerCodeReader(context: ExtensionContext): CodeReaderApi {
api.registerLanguage(createTypeScriptAdapter(wasmDir)),
api.registerLanguage(createPythonAdapter(wasmDir)),
api.registerLanguage(createBashAdapter(wasmDir)),
api.registerLanguage(createDockerfileAdapter(wasmDir)),
api.registerTranslator(typescriptZhTranslator),
api.registerTranslator(pythonZhTranslator),
api.registerTranslator(bashZhTranslator),
...shellCommandRules.map(rule => api.registerCommandRule(rule)),
api.registerTranslator(dockerfileZhTranslator),
...[...shellCommandRules, ...systemCommandRules, ...packageCommandRules].map(rule => api.registerCommandRule(rule)),
api.registerLibraryRule(sqlalchemyRule),
api.registerLibraryRule(celeryRule),
api.registerLibraryRule(enumRule),
@@ -0,0 +1,240 @@
import type { Outline } from '../testSupport'
import { describe, expect, it } from 'vitest'
import { DOCKERFILE_WASM_DIR, factsByLine, nodeFacts, outline, readFixture, runtime, withBlocks } from '../testSupport'
import { createDockerfileAdapter } from './dockerfile'
const adapter = createDockerfileAdapter(DOCKERFILE_WASM_DIR)
const NEXT = readFixture('clientNextjs.Dockerfile')
const BACKEND = readFixture('backend.Dockerfile')
function tree(source: string): Promise<Outline[]> {
return withBlocks(adapter, 'dockerfile', source, outline)
}
function facts(source: string): Promise<Record<number, unknown>> {
return withBlocks(adapter, 'dockerfile', source, factsByLine)
}
/** luffy 前端 frontend/client-nextjs/Dockerfile 的代码块树(同原型 NEXT_DOCKERFILE)。 */
const NEXT_TREE: Outline[] = [
{ at: 1, kind: 'directive' },
{ at: 3, kind: 'stage', kids: [
{ at: 4, kind: 'setting' },
{ at: 5, kind: 'run' },
{ at: 6, kind: 'copy' },
{ at: 7, kind: 'run' },
] },
{ at: 9, kind: 'stage', kids: [
{ at: 10, kind: 'setting' },
{ at: 11, kind: 'run' },
{ at: 12, kind: 'copy' },
{ at: 13, kind: 'copy' },
{ at: 14, kind: 'variable' },
{ at: [15, 16], kind: 'comment' },
{ at: 17, kind: 'variable' },
{ at: 18, kind: 'variable' },
{ at: 19, kind: 'run' },
] },
{ at: 21, kind: 'stage', kids: [
{ at: 22, kind: 'setting' },
{ at: [23, 26], kind: 'variable' },
{ at: 27, kind: 'variable' },
{ at: 28, kind: 'variable' },
{ at: [30, 31], kind: 'run' },
{ at: [33, 34], kind: 'copy' },
{ at: 35, kind: 'comment' },
{ at: 36, kind: 'copy' },
{ at: 38, kind: 'setting' },
{ at: 39, kind: 'setting' },
{ at: 41, kind: 'start' },
] },
]
/** luffy 后端 backend/docker/Dockerfile.backend 的代码块树:FROM 上面紧贴的三行注释是阶段的说明,算在阶段的头部里。 */
const BACKEND_TREE: Outline[] = [
{ at: [1, 4], kind: 'stage', kids: [
{ at: [6, 7], kind: 'label' },
{ at: [9, 12], kind: 'comment' },
{ at: [13, 25], kind: 'run' },
{ at: [27, 31], kind: 'comment' },
{ at: [32, 33], kind: 'copy' },
{ at: [34, 35], kind: 'run' },
{ at: 37, kind: 'copy' },
{ at: [39, 41], kind: 'comment' },
{ at: [42, 44], kind: 'run' },
{ at: 45, kind: 'setting' },
{ at: 46, kind: 'variable' },
{ at: 48, kind: 'setting' },
{ at: [50, 53], kind: 'comment' },
{ at: [54, 65], kind: 'copy' },
{ at: 67, kind: 'comment' },
{ at: [68, 79], kind: 'run' },
{ at: 81, kind: 'comment' },
{ at: 82, kind: 'copy' },
{ at: [83, 84], kind: 'comment' },
{ at: [85, 89], kind: 'copy' },
{ at: 91, kind: 'comment' },
{ at: 92, kind: 'run' },
{ at: [94, 95], kind: 'comment' },
{ at: 96, kind: 'run' },
{ at: 98, kind: 'variable' },
{ at: 100, kind: 'comment' },
{ at: 101, kind: 'variable' },
{ at: 102, kind: 'variable' },
{ at: 104, kind: 'setting' },
{ at: 106, kind: 'comment' },
{ at: [107, 108], kind: 'health' },
{ at: 110, kind: 'start' },
] },
]
describe('dockerfile 代码块树', () => {
it('两份 luffy 的 Dockerfile:解析指令、每个 FROM 一个阶段(后面的指令是子块)、连着写的 ENV / COPY / LABEL 并成一块', async () => {
expect(await tree(NEXT)).toEqual(NEXT_TREE)
expect(await tree(BACKEND)).toEqual(BACKEND_TREE)
})
it('并块:同种指令紧挨着才并,空行、注释、别的指令隔开;COPY 要从同一个地方复制;ADD、带 heredoc 的 COPY 不并;第一个 FROM 之前的照常是顶层的块', async () => {
const source = [
'ARG A=1',
'ARG B',
'FROM x',
'ENV C=1',
'',
'ENV D=2',
'COPY --from=a /x /x',
'COPY --from=a /y /y',
'COPY /z /z',
'ADD a.tar.gz /a/',
'ADD b.tar.gz /b/',
'COPY <<EOF /f',
'hi',
'EOF',
'COPY g /g',
'LABEL a=1',
'LABEL b=2',
].join('\n')
expect(await tree(source)).toEqual([
{ at: [1, 2], kind: 'variable' },
{ at: 3, kind: 'stage', kids: [
{ at: 4, kind: 'variable' },
{ at: 6, kind: 'variable' },
{ at: [7, 8], kind: 'copy' },
{ at: 9, kind: 'copy' },
{ at: 10, kind: 'copy' },
{ at: 11, kind: 'copy' },
{ at: [12, 14], kind: 'copy' },
{ at: 15, kind: 'copy' },
{ at: [16, 17], kind: 'label' },
] },
])
})
it('onbuild 是一条设置,包着的指令是它的子块;别的指令(WORKDIR、USER、EXPOSE、VOLUME、STOPSIGNAL、SHELL)各是一行设置', async () => {
const source = 'FROM x\nONBUILD RUN make\nWORKDIR /a\nUSER u\nEXPOSE 80\nVOLUME /v\nSTOPSIGNAL SIGTERM\nSHELL ["/bin/bash", "-c"]\n'
expect(await tree(source)).toEqual([{ at: 1, kind: 'stage', kids: [
{ at: 2, kind: 'setting', kids: [{ at: 2, kind: 'run' }] },
...[3, 4, 5, 6, 7, 8].map(at => ({ at, kind: 'setting' })),
] }])
})
})
describe('dockerfile 属性表信息(facts)', () => {
it('解析指令、阶段(名称、第几个、说明、基础镜像、标签 / 摘要、接着哪个阶段、选项)', async () => {
expect((await facts(NEXT))[1]).toEqual({ directives: [{ name: 'syntax', value: 'docker/dockerfile:1.7' }] })
expect((await facts(NEXT))[9]).toEqual({ name: 'builder', index: 1, image: 'node', stage: false, tag: ':22-alpine', flags: [] })
expect((await facts(BACKEND))[1]).toMatchObject({ name: 'base', index: 0, doc: ['# ── 统一 Backend 镜像 ─────────────────────────────────────────────────', expect.any(String), expect.any(String)] })
const byLine = await facts('FROM --platform=$BUILDPLATFORM golang@sha256:abc AS build\nFROM build\n')
expect(byLine[1]).toMatchObject({ image: 'golang', digest: '@sha256:abc', flags: [{ node: '--platform=$BUILDPLATFORM', name: 'platform', value: '$BUILDPLATFORM', valueStart: 16 }] })
expect(byLine[2]).toEqual({ index: 1, image: 'build', stage: true, flags: [] })
})
it('变量:ARG 是构建参数、ENV 是环境变量,连着写的并成一块(各条的变量都在里面)', async () => {
const byLine = await facts(NEXT)
expect(byLine[17]).toEqual({ declarators: [{ name: 'DEPLOYMENT_ID' }], scope: 'build' })
expect(byLine[23]).toEqual({ scope: 'environment', declarators: [
{ name: 'NODE_ENV', value: 'production' },
{ name: 'NEXT_TELEMETRY_DISABLED', value: '1' },
{ name: 'PORT', value: '3000' },
{ name: 'HOSTNAME', value: '0.0.0.0' },
] })
})
it('健康检查(选项、检查命令两种写法、NONE),启动(CMD / ENTRYPOINT,数组写法和交给 shell 的一段)', async () => {
const byLine = await facts(BACKEND)
expect(byLine[107]).toMatchObject({
none: false,
flags: [{ name: 'interval', value: '30s' }, { name: 'timeout', value: '5s' }, { name: 'start-period', value: '10s' }, { name: 'retries', value: '3' }],
command: { form: 'shell', code: { languageId: 'shellscript' } },
})
expect(byLine[110]).toMatchObject({ instruction: 'cmd', command: { form: 'exec', node: '["uv", "run", "uvicorn", "luffy_agent.main:app", "--host", "0.0.0.0", "--port", "8000"]' } })
const other = await facts('FROM x\nHEALTHCHECK NONE\nENTRYPOINT node a.js\n')
expect(other[2]).toEqual({ none: true, flags: [] })
expect(other[3]).toMatchObject({ instruction: 'entrypoint', command: { form: 'shell' } })
})
})
describe('dockerfile 嵌着的 shell、命令、名字', () => {
it('嵌着的 shell:各个片段去掉续行的 \\ 和缩进;只写了 <<EOT 的是 heredoc 的正文', async () => {
const source = 'RUN useradd u && \\\n mkdir -p /a\nRUN <<EOT\nset -e\napt-get update\nEOT\n'
const parsed = await runtime.parse(adapter, 'dockerfile', source)
try {
const codes = adapter.embedded?.(parsed.rootNode, source, 'dockerfile') ?? []
expect(codes.map(code => ({ node: code.node.type, languageId: code.languageId, text: code.ranges.map(range => source.slice(range.startIndex, range.endIndex)) }))).toEqual([
{ node: 'shell_command', languageId: 'shellscript', text: ['useradd u && ', 'mkdir -p /a'] },
{ node: 'shell_command', languageId: 'shellscript', text: ['set -e\napt-get update'] },
])
}
finally {
parsed.delete()
}
})
it('数组写法的命令(RUN、CMD、ENTRYPOINT、HEALTHCHECK CMD)交给命令规则:第一项是命令名,各项去掉引号;VOLUME、SHELL 的数组不是命令', async () => {
const commands = await nodeFacts(adapter, 'dockerfile', 'CMD ["node", "server.js"]\nVOLUME ["/data"]\nHEALTHCHECK CMD ["curl", "-f", "http://x/"]\n', node => adapter.command?.(node))
expect(commands).toEqual([
{ name: { node: '"node"', literal: 'node' }, arguments: [{ node: '"server.js"', literal: 'server.js' }], redirects: [] },
{ name: { node: '"curl"', literal: 'curl' }, arguments: [{ node: '"-f"', literal: '-f' }, { node: '"http://x/"', literal: 'http://x/' }], redirects: [] },
])
})
it('同一文件里的名字:阶段(AS 后面的名字、COPY --from=、--mount=…,from=、FROM 阶段名),变量(ARG、ENV 的名字,$X、{X} 这种展开)', async () => {
const source = `FROM node AS deps\nARG V\nFROM deps AS build\nENV W=$V\nCOPY --from=deps /a /a\nRUN --mount=type=bind,from=deps,target=/x ls\nWORKDIR \${W}\n`
const parsed = await runtime.parse(adapter, 'dockerfile', source)
try {
const symbols = adapter.symbols?.(parsed.rootNode, source, 'dockerfile') ?? []
const texts = symbols.map(symbol => ({
definitions: symbol.definitions.map(range => `${source.slice(range.startIndex, range.endIndex)}@${range.startIndex}`),
occurrences: symbol.occurrences.map(range => source.slice(range.startIndex, range.endIndex)),
}))
expect(texts).toEqual([
{ definitions: [`deps@${source.indexOf('deps')}`], occurrences: ['deps', 'deps', 'deps', 'deps'] },
{ definitions: [`V@${source.indexOf('V\n')}`], occurrences: ['V', 'V'] },
{ definitions: [`build@${source.indexOf('build')}`], occurrences: ['build'] },
{ definitions: [`W@${source.indexOf('W=')}`], occurrences: ['W', 'W'] },
])
}
finally {
parsed.delete()
}
})
it('中文名:紧挨在 ARG、ENV(一个变量的)、FROM … AS 上面、只写了 # @Comment 的注释;这行注释算在那条指令的块里', async () => {
const source = '# @Comment 部署 ID\nARG DEPLOYMENT_ID\n# @Comment 依赖阶段\nFROM node AS deps\n'
const parsed = await runtime.parse(adapter, 'dockerfile', source)
try {
const found = adapter.displayNames?.find(parsed.rootNode, source, 'dockerfile')
expect(found?.names.map(({ name, text }) => [source.slice(name.startIndex, name.endIndex), text])).toEqual([['DEPLOYMENT_ID', '部署 ID'], ['deps', '依赖阶段']])
expect(found?.markers.map(range => source.slice(range.startIndex, range.endIndex))).toEqual(['@Comment 部署 ID', '@Comment 依赖阶段'])
}
finally {
parsed.delete()
}
expect(await tree(source)).toEqual([{ at: [1, 2], kind: 'variable' }, { at: [3, 4], kind: 'stage' }])
})
it('着色:指令关键字、镜像名、标签、阶段名、路径、变量', async () => {
const roles = await nodeFacts(adapter, 'dockerfile', 'FROM node:22 AS deps\nCOPY a /b\nENV PORT=$X\n', node => node.childCount === 0 ? `${node.text}:${adapter.tokenRole(node)}` : undefined)
expect(roles).toEqual(expect.arrayContaining(['FROM:keyword', 'node:type', 'AS:keyword', 'deps:variable', 'COPY:keyword', 'a:string', '/b:string', 'PORT:constant', 'X:constant']))
})
})
@@ -0,0 +1,435 @@
import type { CodeBlock, CommandFacts, CommandForm, CommandWord, DeclaratorFacts, DirectiveFacts, DisplayName, DisplayNameReader, EmbeddedCode, InstructionFlag, LanguageAdapter, LocalSymbol, Node, ParserDirective, SourceRange, SourceSpan, StageFacts, TextBlock, TokenRole } from '../contract'
import * as path from 'node:path'
import { COMMENT_MARK, commentTag } from './displayNames'
import { children, leafBlock, leafShape, named } from './shared'
/** RUN、shell 写法的 CMD / ENTRYPOINT、HEALTHCHECK 的检查命令交给 shell 的语言插件读。 */
const SHELL = 'shellscript'
/** BuildKit 认的解析指令,只在文件最开头连着写的几行注释里算。 */
const DIRECTIVE = /^#\s*(syntax|escape|check)\s*=(.*)$/i
/** 一条指令一行「设置」。 */
const SETTINGS = new Set(['workdir_instruction', 'user_instruction', 'expose_instruction', 'volume_instruction', 'stopsignal_instruction', 'shell_instruction', 'maintainer_instruction'])
/** 连着写(中间没空行、没注释)就并成一块的指令;ADD 各条做的事不一样(下载、解压、复制),不并。 */
const MERGED = new Set(['env_instruction', 'arg_instruction', 'copy_instruction', 'label_instruction'])
/** 执行命令的指令:数组写法的命令交给命令规则读。 */
const COMMAND_INSTRUCTIONS = new Set(['run_instruction', 'cmd_instruction', 'entrypoint_instruction'])
/** 按习惯当常量用的名字:全大写(可带数字、下划线)。 */
const CONSTANT_NAME = /^[A-Z][A-Z0-9_]*$/
const NAME = /^[A-Z_]\w*/i
function spanOf(first: Node, last: Node): SourceSpan {
return { startIndex: first.startIndex, endIndex: last.endIndex, startRow: first.startPosition.row, endRow: last.endPosition.row }
}
/** 下一个节点紧接在上一个节点的下一行开始(中间没空行)。 */
function adjacent(previous: Node, next: Node): boolean {
return next.startPosition.row === previous.endPosition.row + 1
}
// ─── 选项、命令 ─────────────────────────────────────────────────────────────
/** 指令后面的 `--名字=值`(`--mount=type=cache,…` 的值是逗号后面那一整串)。 */
export function instructionFlags(instruction: Node): InstructionFlag[] {
return named(instruction).flatMap((node): InstructionFlag[] => {
if (node.type !== 'param' && node.type !== 'mount_param')
return []
const match = /^--([a-z][-a-z]*)=(.*)$/s.exec(node.text)
return match ? [{ node, name: match[1], value: match[2], valueStart: node.startIndex + match[1].length + 3 }] : []
})
}
/** 只写了一个 `<<EOT` 的 shell 片段:脚本是下面的 heredoc。 */
function heredocOnly(fragments: Node[]): boolean {
return fragments.length === 1 && named(fragments[0]).some(child => child.type === 'heredoc_marker') && /^<<-?["']?\w+["']?\s*$/.test(fragments[0].text)
}
/**
* shell 写法的命令交给 shell 读的几段:各个 shell 片段(去掉续行的 `\` 和下一行的缩进);
* 只写了 `<<EOT` 的是它下面 heredoc 的正文(从第一行到最后一行,换行照留)。
*/
function shellCode(command: Node): EmbeddedCode | undefined {
const fragments = named(command).filter(child => child.type === 'shell_fragment')
if (fragments.length === 0)
return undefined
if (heredocOnly(fragments)) {
const lines = named(named(command.parent).find(child => child.type === 'heredoc_block')).filter(child => child.type === 'heredoc_line')
const [first, last] = [lines[0], lines.at(-1)]
return first && last ? { node: command, languageId: SHELL, ranges: [{ startIndex: first.startIndex, endIndex: last.endIndex }] } : undefined
}
return { node: command, languageId: SHELL, ranges: fragments.map(({ startIndex, endIndex }) => ({ startIndex, endIndex })) }
}
/** 执行命令的两种写法:数组(不经过 shell)和交给 shell 的一段。 */
function commandForm(instruction: Node): CommandForm | undefined {
const array = named(instruction).find(child => child.type === 'json_string_array')
if (array)
return { form: 'exec', node: array }
const shell = named(instruction).find(child => child.type === 'shell_command')
const code = shell && shellCode(shell)
return code && { form: 'shell', code }
}
/** JSON 字符串里写的文字(去掉引号、转义)。 */
function jsonText(node: Node): string | undefined {
try {
const value: unknown = JSON.parse(node.text)
return typeof value === 'string' ? value : undefined
}
catch {
return undefined
}
}
function jsonWord(node: Node): CommandWord {
const literal = jsonText(node)
return { node, ...(literal !== undefined && { literal }) }
}
/** 数组写法的命令(RUN、CMD、ENTRYPOINT、HEALTHCHECK CMD 后面的 `["node", "server.js"]`):第一项是命令名,后面是各个词。 */
function dockerCommand(node: Node): CommandFacts | undefined {
if (node.type !== 'json_string_array' || !COMMAND_INSTRUCTIONS.has(node.parent?.type ?? ''))
return undefined
const [name, ...args] = named(node).filter(child => child.type === 'json_string')
return name ? { name: jsonWord(name), arguments: args.map(jsonWord), redirects: [] } : undefined
}
// ─── 块 ──────────────────────────────────────────────────────────────────────
/** 文件最开头连着的几行 `# syntax=…`:解析指令块;别的注释、空行、指令之后不再算。 */
function directiveBlock(items: Node[]): { block?: CodeBlock, rest: Node[] } {
const directives: ParserDirective[] = []
let count = 0
for (const item of items) {
const match = item.type === 'comment' && item.startPosition.row === count ? DIRECTIVE.exec(item.text) : null
if (!match)
break
directives.push({ name: match[1].toLowerCase(), value: match[2].trim() })
count++
}
if (count === 0)
return { rest: items }
const facts: DirectiveFacts = { directives }
return { block: { kind: 'directive', node: items[0], header: spanOf(items[0], items[count - 1]), children: [], facts }, rest: items.slice(count) }
}
/** 只写了中文名记号的注释(`# @Comment 部署 ID`):它下面紧挨着的那条指令的一部分,不单独成块。 */
function markerOnly(comment: Node): boolean {
const tag = commentTag(comment.text)
return !!tag && comment.text.slice(0, tag.start).replace(/^#/, '').trim() === ''
}
function envDeclarators(instruction: Node): DeclaratorFacts[] {
return named(instruction).filter(child => child.type === 'env_pair').flatMap((pair): DeclaratorFacts[] => {
const name = pair.childForFieldName('name')
const value = pair.childForFieldName('value')
return name ? [{ name, ...(value && { value }) }] : []
})
}
function argDeclarators(instruction: Node): DeclaratorFacts[] {
const name = instruction.childForFieldName('name')
const value = instruction.childForFieldName('default')
return name ? [{ name, ...(value && { value }) }] : []
}
/** 连着写的同种指令并成的一块:ENV / ARG 是一个变量块(各条的变量都在里面),COPY / ADD / LABEL 是一行字加一张表。 */
function mergedBlock(run: Node[]): CodeBlock {
const [first] = run
const last = run.at(-1) ?? first
const shape = { node: first, header: spanOf(first, last), children: [] }
switch (first.type) {
case 'env_instruction':
return { kind: 'variable', ...shape, facts: { declarators: run.flatMap(envDeclarators), scope: 'environment' } }
case 'arg_instruction':
return { kind: 'variable', ...shape, facts: { declarators: run.flatMap(argDeclarators), scope: 'build' } }
default: {
const block: TextBlock = { kind: first.type === 'label_instruction' ? 'label' : 'copy', ...shape }
return run.length > 1 ? { ...block, group: run } : block
}
}
}
/**
* 能并进同一块:同一种指令、紧挨着下一行;COPY / ADD 还要从同一个地方复制(都从构建目录,或 `--from` 一样)。
* 带 heredoc 的 COPY(`COPY <<EOF /f`)单独成块。
*/
function mergeable(previous: Node, next: Node): boolean {
const heredoc = (node: Node): boolean => named(node).some(child => child.type === 'heredoc_block')
const from = (node: Node): string | undefined => instructionFlags(node).find(flag => flag.name === 'from')?.value
return next.type === previous.type && adjacent(previous, next) && !heredoc(previous) && !heredoc(next) && from(previous) === from(next)
}
function instructionBlock(item: Node): CodeBlock {
if (SETTINGS.has(item.type))
return leafBlock('setting', item)
switch (item.type) {
case 'run_instruction': {
const shell = named(item).find(child => child.type === 'shell_command')
const code = shell && shellCode(shell)
return { kind: 'run', ...leafShape(item), ...(code && { embedded: code }) }
}
case 'healthcheck_instruction': {
const cmd = named(item).find(child => child.type === 'cmd_instruction')
const command = cmd && commandForm(cmd)
const none = children(item).some(child => child.type === 'NONE')
return { kind: 'health', ...leafShape(item), facts: { none, flags: instructionFlags(item), ...(command && { command }) } }
}
case 'cmd_instruction':
case 'entrypoint_instruction': {
const command = commandForm(item)
return command
? { kind: 'start', ...leafShape(item), facts: { instruction: item.type === 'cmd_instruction' ? 'cmd' : 'entrypoint', command } }
: leafBlock('statement', item)
}
case 'onbuild_instruction': {
const [inner] = named(item)
return { kind: 'setting', ...leafShape(item), children: inner ? [instructionBlock(inner)] : [] }
}
default:
return MERGED.has(item.type) || item.type === 'add_instruction' ? mergedBlock([item]) : leafBlock('statement', item)
}
}
/** 阶段里(或第一个 FROM 之前)的一串指令、注释:相邻行的注释合成一块,连着写的同种指令并成一块。 */
function instructionBlocks(items: Node[]): CodeBlock[] {
const out: CodeBlock[] = []
let comments: CodeBlock | undefined
let marker: Node | undefined
for (let i = 0; i < items.length; i++) {
const item = items[i]
if (item.type === 'comment') {
const next = items[i + 1]
if (markerOnly(item) && next && next.type !== 'comment' && adjacent(item, next)) {
marker = item
comments = undefined
continue
}
if (comments && adjacent(items[i - 1], item)) {
comments.header = { ...comments.header, endIndex: item.endIndex, endRow: item.endPosition.row }
}
else {
comments = leafBlock('comment', item)
out.push(comments)
}
continue
}
comments = undefined
const run = [item]
while (MERGED.has(item.type) && items[i + 1] && mergeable(run[run.length - 1], items[i + 1]))
run.push(items[++i])
const block = run.length > 1 ? mergedBlock(run) : instructionBlock(item)
out.push(marker ? { ...block, header: { ...block.header, startIndex: marker.startIndex, startRow: marker.startPosition.row } } : block)
marker = undefined
}
return out
}
/** 紧贴在 FROM 上面、连着的几行注释(最后一行就在 FROM 上一行):阶段的说明。 */
function stageDoc(items: Node[], at: number): Node[] {
const doc: Node[] = []
for (let i = at - 1; i >= 0 && items[i].type === 'comment' && adjacent(items[i], doc[0] ?? items[at]); i--)
doc.unshift(items[i])
return doc
}
function stageBlock(from: Node, index: number, doc: Node[], aliases: Set<string>, children: CodeBlock[]): CodeBlock {
const spec = named(from).find(child => child.type === 'image_spec')
const image = spec?.childForFieldName('name') ?? from
const name = from.childForFieldName('as')
const tag = spec?.childForFieldName('tag')
const digest = spec?.childForFieldName('digest')
const facts: StageFacts = {
...(name && { name }),
index,
...(doc.length > 0 && { doc }),
image,
stage: aliases.has(image.text.toLowerCase()),
...(tag && { tag }),
...(digest && { digest }),
flags: instructionFlags(from),
}
return { kind: 'stage', node: from, header: spanOf(doc[0] ?? from, from), children, facts }
}
/** 每个 FROM 开一个阶段,到下一个阶段(连同它上面的说明)之前的指令、注释是它的子块;第一个 FROM 之前的照常是顶层的块。 */
function dockerBlocks(root: Node): CodeBlock[] {
const { block: directives, rest: items } = directiveBlock(named(root))
const froms = items.flatMap((item, i) => item.type === 'from_instruction' ? [i] : [])
const docs = froms.map(at => stageDoc(items, at))
const starts = froms.map((at, k) => at - docs[k].length)
const out: CodeBlock[] = [...(directives ? [directives] : []), ...instructionBlocks(items.slice(0, starts[0] ?? items.length))]
const aliases = new Set<string>()
froms.forEach((at, k) => {
const from = items[at]
const stage = stageBlock(from, k, docs[k], aliases, instructionBlocks(items.slice(at + 1, starts[k + 1] ?? items.length)))
const alias = from.childForFieldName('as')
if (alias)
aliases.add(alias.text.toLowerCase())
out.push(stage)
})
return out
}
// ─── 着色、名字 ──────────────────────────────────────────────────────────────
function nameRole(text: string): TokenRole {
return CONSTANT_NAME.test(text) ? 'constant' : 'variable'
}
function dockerTokenRole(leaf: Node): TokenRole {
const parent = leaf.parent
if (leaf.type === 'comment')
return 'comment'
if (!leaf.isNamed)
return /^[A-Z]+$/.test(leaf.type) ? 'keyword' : 'plain'
switch (leaf.type) {
case 'image_name':
return 'type'
case 'image_tag':
case 'image_digest':
case 'path':
case 'json_string':
return 'string'
case 'image_alias':
return 'variable'
case 'variable':
return nameRole(leaf.text)
case 'expose_port':
return 'number'
case 'heredoc_line':
return parent?.parent?.type === 'run_instruction' ? 'plain' : 'string'
case 'unquoted_string': {
const field = parent && fieldOf(leaf, parent)
if ((parent?.type === 'env_pair' && field === 'name') || (parent?.type === 'arg_instruction' && field === 'name'))
return nameRole(leaf.text)
if (parent?.type === 'label_pair' && field === 'key')
return 'variable'
return parent?.type === 'user_instruction' ? 'variable' : 'string'
}
default:
return 'plain'
}
}
function fieldOf(node: Node, parent: Node): string | null {
for (let i = 0; i < parent.childCount; i++) {
if (parent.child(i)?.id === node.id)
return parent.fieldNameForChild(i)
}
return null
}
function rangeOf(node: Node): SourceRange {
return { startIndex: node.startIndex, endIndex: node.endIndex }
}
/** 选项值里写的阶段名(`--from=deps`、`--mount=…,from=deps`)在源码里的范围。 */
function fromRanges(instruction: Node): { name: string, range: SourceRange }[] {
return instructionFlags(instruction).flatMap((flag): { name: string, range: SourceRange }[] => {
if (flag.name === 'from')
return [{ name: flag.value, range: { startIndex: flag.valueStart, endIndex: flag.valueStart + flag.value.length } }]
const match = flag.name === 'mount' ? /(?:^|,)from=([^,]+)/.exec(flag.value) : null
if (!match)
return []
const start = flag.valueStart + match.index + match[0].indexOf('=') + 1
return [{ name: match[1], range: { startIndex: start, endIndex: start + match[1].length } }]
})
}
/**
* 同一文件里认得出的名字:阶段(定义处是 `AS` 后面的名字,用到的是 `COPY --from=`、`--mount=…,from=`、`FROM 阶段名`),
* 变量(定义处是 ARG、ENV 的名字,用到的是 `$X`、`${X}`)。
*/
function dockerSymbols(root: Node): LocalSymbol[] {
const symbols = new Map<string, LocalSymbol>()
const entry = (key: string): LocalSymbol => {
let symbol = symbols.get(key)
if (!symbol) {
symbol = { definitions: [], occurrences: [] }
symbols.set(key, symbol)
}
return symbol
}
const aliases = new Set(root.descendantsOfType('image_alias').flatMap(alias => alias ? [alias.text.toLowerCase()] : []))
const visit = (node: Node): void => {
if (node.type === 'image_alias') {
const symbol = entry(`stage:${node.text.toLowerCase()}`)
symbol.definitions.push(rangeOf(node))
symbol.occurrences.push(rangeOf(node))
}
else if (node.type === 'image_name' && aliases.has(node.text.toLowerCase())) {
entry(`stage:${node.text.toLowerCase()}`).occurrences.push(rangeOf(node))
}
else if ((node.type === 'env_pair' || node.type === 'arg_instruction') && node.childForFieldName('name')) {
const name = node.childForFieldName('name') as Node
const symbol = entry(`variable:${name.text}`)
symbol.definitions.push(rangeOf(name))
symbol.occurrences.push(rangeOf(name))
}
else if (node.type === 'variable') {
const name = NAME.exec(node.text)?.[0]
if (name)
entry(`variable:${name}`).occurrences.push({ startIndex: node.startIndex, endIndex: node.startIndex + name.length })
}
for (const { name, range } of fromRanges(node)) {
if (aliases.has(name.toLowerCase()))
entry(`stage:${name.toLowerCase()}`).occurrences.push(range)
}
for (const child of children(node))
visit(child)
}
visit(root)
for (const symbol of symbols.values())
symbol.occurrences.sort((a, b) => a.startIndex - b.startIndex)
return [...symbols.values()].filter(symbol => symbol.definitions.length > 0)
}
/**
* Dockerfile 里的中文名:紧挨在 ARG、ENV(只有一个变量的)、`FROM … AS 名字` 上面一行、只写了 `# @Comment 名字` 的注释。
* Dockerfile 的注释只能单独成行,没有行尾记号的写法。
*/
const dockerNameReader: DisplayNameReader = {
files: { dockerfile: '**/{Dockerfile,Dockerfile.*,*.dockerfile,Containerfile,Containerfile.*}' },
mentions: source => source.includes(COMMENT_MARK),
find: (root) => {
const names: DisplayName[] = []
const markers: SourceRange[] = []
for (const comment of named(root).filter(item => item.type === 'comment' && markerOnly(item))) {
const next = comment.nextNamedSibling
const tag = commentTag(comment.text)
if (!next || !tag || !adjacent(comment, next))
continue
const pairs = next.type === 'env_instruction' ? named(next).filter(child => child.type === 'env_pair') : []
const name = next.type === 'from_instruction'
? next.childForFieldName('as')
: next.type === 'arg_instruction' ? next.childForFieldName('name') : pairs.length === 1 ? pairs[0].childForFieldName('name') : null
if (!name)
continue
names.push({ name: rangeOf(name), text: tag.text })
markers.push({ startIndex: comment.startIndex + tag.start, endIndex: comment.endIndex })
}
return { names, markers }
},
}
/**
* Dockerfile(VS Code 的 languageId 是 dockerfile:Dockerfile、Dockerfile.*、*.dockerfile、Containerfile)的语言插件。
* 语法包是 tree-sitter-dockerfile(npm 包 tree-sitter-wasm 预编译的),RUN 这些 shell 代码交给 shell 的语言插件(见 EmbeddedCode)。
* @param wasmDir 放 tree-sitter-dockerfile.wasm 的目录
*/
export function createDockerfileAdapter(wasmDir: string): LanguageAdapter {
return {
id: 'dockerfile',
languageIds: ['dockerfile'],
grammarPath: () => path.join(wasmDir, 'tree-sitter-dockerfile.wasm'),
toBlocks: root => dockerBlocks(root),
tokenRole: dockerTokenRole,
command: dockerCommand,
symbols: root => dockerSymbols(root),
embedded: root => root.descendantsOfType('shell_command').flatMap((node) => {
const code = node && shellCode(node)
return code ? [code] : []
}),
displayNames: dockerNameReader,
}
}
@@ -0,0 +1,323 @@
import type { BlockView } from '../blockView'
import { describe, expect, it } from 'vitest'
import { readBlockViews } from '../blockView'
import { commandRuleLookup } from '../commandRules'
import { packageCommandRules } from '../commands/packageCommands'
import { shellCommandRules } from '../commands/shellCommands'
import { systemCommandRules } from '../commands/systemCommands'
import { createBashAdapter } from '../languages/bash'
import { createDockerfileAdapter } from '../languages/dockerfile'
import { DOCKERFILE_WASM_DIR, flatSegments, readFixture, runtime, viewLines, WASM_DIR } from '../testSupport'
import { bashZhTranslator } from './bashZh'
import { dockerfileZhTranslator } from './dockerfileZh'
const bash = createBashAdapter(WASM_DIR)
const docker = createDockerfileAdapter(DOCKERFILE_WASM_DIR)
const RULES = [...shellCommandRules, ...systemCommandRules, ...packageCommandRules]
/** RUN 里的 shell 交给 bash 的语言插件和中文直译(withShell 为假时不给,照抄)。 */
async function read(source: string, withShell = true): Promise<BlockView[]> {
const commandRule = commandRuleLookup(RULES, () => {})
const { blocks } = await readBlockViews(runtime, docker, dockerfileZhTranslator, 'dockerfile', source, {
commandRule,
...(withShell && { embedded: (id: string) => id === 'shellscript' ? { adapter: bash, translator: bashZhTranslator, commandRule } : undefined }),
})
return blocks
}
async function readLines(source: string): Promise<string[]> {
return viewLines(await read(source))
}
/** luffy 前端 frontend/client-nextjs/Dockerfile 读出来的样子(原型 NEXT_DOCKERFILE)。 */
const NEXT_READING = [
'directive 1',
' 语法: docker/dockerfile:1.7(BuildKit 按这个版本读这份文件)',
'stage 3',
' 名称: deps',
' 基础镜像: node,标签 22-alpine',
' setting 4: 工作目录 /app',
' run 5: 执行',
' command 5: 开启 corepack',
' command 5: 准备 pnpm latest(并设为当前用的版本)',
' copy 6: 复制 package.json、pnpm-lock.yaml → ./',
' run 7: pnpm 装依赖(严格按 lock 文件、不跑安装脚本)',
'stage 9',
' 名称: builder',
' 基础镜像: node,标签 22-alpine',
' setting 10: 工作目录 /app',
' run 11: 执行',
' command 11: 开启 corepack',
' command 11: 准备 pnpm latest(并设为当前用的版本)',
' copy 12: 从阶段 deps 复制 /app/node_modules → ./node_modules',
' copy 13: 复制 整个构建目录 → ./',
' variable 14',
' 环境变量: NEXT_TELEMETRY_DISABLED',
' 值: 1',
' comment 15: 部署 ID 烙进本次构建: Next 据此给静态资源加 ?dpl=、页内跳转时比对版本;',
' variable 17',
' 构建参数: DEPLOYMENT_ID',
' 默认值: 没写,构建时传',
' variable 18',
' 环境变量: NEXT_DEPLOYMENT_ID',
' 值: DEPLOYMENT_ID(构建参数)',
' run 19: pnpm 构建',
'stage 21',
' 名称: runner',
' 基础镜像: node,标签 22-alpine',
' setting 22: 工作目录 /app',
' variable 23',
' 环境变量: 4 条连着写',
' NODE_ENV | production',
' NEXT_TELEMETRY_DISABLED | 1',
' PORT | 3000',
' HOSTNAME | 0.0.0.0',
' variable 27',
' 构建参数: DEPLOYMENT_ID',
' 默认值: 没写,构建时传',
' variable 28',
' 环境变量: NEXT_DEPLOYMENT_ID',
' 值: DEPLOYMENT_ID(构建参数)',
' run 30: 执行',
' command 30: 新建系统用户组 nodejs(gid 1001)',
' command 31: 新建系统用户 nextjs(uid 1001)',
' copy 33: 复制(2 条连着写,属主都是 nextjs:nodejs)',
' 阶段 builder | /app/.next/standalone | ./',
' 阶段 builder | /app/.next/static | ./.next/static',
' comment 35: standalone 输出不含 public/, 必须显式拷贝 (字体等静态资源)',
' copy 36: 从阶段 builder 复制 /app/public → ./public(属主 nextjs:nodejs)',
' setting 38: 之后以 nextjs 身份运行',
' setting 39: 开放端口 3000',
' start 41',
' 命令: node server.js',
' 写法: 数组,不经过 shell',
]
/** luffy 后端 backend/docker/Dockerfile.backend 读出来的样子(原型 BACKEND_DOCKERFILE,原型略去的几段也在)。 */
const BACKEND_READING = [
'stage 1',
' 名称: base',
' 说明: ── 统一 Backend 镜像 ─────────────────────────────────────────────────',
' 基础镜像: python,标签 3.12-slim',
' label 6: 标签',
' maintainer | Cruldra',
' description | luffy-agent-platform - 统一 Backend 镜像 (API/Worker/Beat/Flower)',
' comment 9: git: deployment 模块创建 bare repo;postgresql-client: alembic / psql 运维;curl: healthcheck',
' run 13: 执行(10 条命令,一条失败就停)',
' command 13: 更新软件包列表',
' command 14: 安装 ca-certificates、curl、gnupg、postgresql-client、git(不装推荐的包、不用确认)',
' command 16: 建目录 /usr/share/keyrings',
' pipeline 17: 请求 https://www.postgresql.org/media/keys/ACCC4CF8.asc(不打印进度、失败就报错、跟着跳转走) → 转成二进制密钥 写到 /usr/share/keyrings/postgresql.gpg',
' command 19: 读入 /etc/os-release 里的变量和函数',
' command 20: 打印 "deb [signed-by=/usr/share/keyrings/postgresql.gpg] http://apt.postgresql.org/pub/repos/apt VERSION_CODENAME-pgdg main"(写到 /etc/apt/sources.list.d/pgdg.list)',
' command 22: 更新软件包列表',
' command 23: 安装 postgresql-16-repack(不装推荐的包、不用确认)',
' command 24: 打印 pg_repack 的版本',
' command 25: 删除 /var/lib/apt/lists/*(连同里面的东西、不提示)',
' comment 27: Node.js:agent MCP 节点 (如 context7) 用 stdio + npx 拉子进程, api 容器必须有 npx。',
' copy 32: 复制(2 条连着写)',
' 镜像 node:22-slim | /usr/local/bin/node | /usr/local/bin/node',
' 镜像 node:22-slim | /usr/local/lib/node_modules | /usr/local/lib/node_modules',
' run 34: 执行',
' command 34: 建符号链接 /usr/local/bin/npm → /usr/local/lib/node_modules/npm/bin/npm-cli.js(已有就替换)',
' command 35: 建符号链接 /usr/local/bin/npx → /usr/local/lib/node_modules/npm/bin/npx-cli.js(已有就替换)',
' copy 37: 从镜像 ghcr.io/astral-sh/uv:latest 复制 /uv、/uvx → /bin/',
' comment 39: appuser 先建、/app /lib 先归属:.venv 由 appuser 直接创建,末尾无需 chown -R',
' run 42: 执行(3 条命令,一条失败就停)',
' command 42: 新建用户 appuser(建主目录、不能登录)',
' command 43: 建目录 /app、/lib(连同上级目录)',
' command 44: 把 /app、/lib 的属主改成 appuser:appuser',
' setting 45: 之后以 appuser 身份运行',
' variable 46',
' 环境变量: UV_CACHE_DIR',
' 值: /home/appuser/.cache/uv',
' setting 48: 工作目录 /app',
' comment 50: ── 依赖层(仅随 pyproject/uv.lock 变化重建)───────────────────────────',
' copy 54: 复制(12 条连着写,属主都是 appuser:appuser)',
' backend/pyproject.toml、backend/uv.lock | ./',
' lib/agent-dsl/pyproject.toml | /lib/agent-dsl/',
' lib/agent-compiler-deepagents/pyproject.toml | /lib/agent-compiler-deepagents/',
' lib/vendor-tools/pyproject.toml | /lib/vendor-tools/',
' lib/code-sandbox-core/pyproject.toml | /lib/code-sandbox-core/',
' lib/code-sandbox-daytona/pyproject.toml | /lib/code-sandbox-daytona/',
' lib/code-sandbox-e2b/pyproject.toml | /lib/code-sandbox-e2b/',
' lib/code-sandbox-local/pyproject.toml | /lib/code-sandbox-local/',
' lib/code-publish-core/pyproject.toml | /lib/code-publish-core/',
' lib/code-publish-compose-runner/pyproject.toml | /lib/code-publish-compose-runner/',
' lib/luffy-billing-sdk/pyproject.toml | /lib/luffy-billing-sdk/',
' lib/luffy-observability/pyproject.toml | /lib/luffy-observability/',
' comment 67: 三方依赖先装;本项目与 lib editable 包此时无源码,跳过安装留给第二次 sync',
' run 68: uv 按 lock 同步依赖 · 4 个选项',
' --frozen | 严格按 lock 文件,不改它',
' --no-dev | 不装开发依赖',
' --no-install-project | 不装本项目',
' --no-install-package × 11 | 不装这些包:agent-dsl、agent-compiler-deepagents、vendor-tools、code-sandbox-core、code-sandbox-daytona、code-sandbox-e2b、code-sandbox-local、code-publish-core、code-publish-compose-runner、luffy-billing-sdk、luffy-observability',
' comment 81: ── 源码层(每次代码变更只重建这里往下)────────────────────────────────',
' copy 82: 复制 lib → /lib(属主 appuser:appuser)',
' comment 83: agent 定义 (vendor/agents/*/agent.yaml + nodes/*.py) 与代码是同一个版本, 必须随镜像原子切换;',
' copy 85: 复制(5 条连着写,属主都是 appuser:appuser)',
' vendor | /vendor',
' backend/src/ | ./src/',
' backend/alembic/ | ./alembic/',
' backend/alembic.ini | ./',
' backend/scripts/ | ./scripts/',
' comment 91: 补装本项目 + lib editable 包(三方依赖已就位,秒级)',
' run 92: uv 按 lock 同步依赖(严格按 lock 文件,不改它、不装开发依赖)',
' comment 94: 远程调试支持:装入 uv 创建的 venv,按需通过 DEBUGPY_ENABLE=1 开启。',
' run 96: uv 安装 debugpy',
' variable 98',
' 环境变量: UV_NO_SYNC',
' 值: 1',
' comment 100: 部署 ID (Jenkins IMAGE_TAG): /health 回显、日志定位后端出自哪次构建. 选择性发布时 api 与 client 镜像 tag 可不同, 客户端是否过期以 client 镜像 (浏览器包 vs BFF) 为准.',
' variable 101',
' 构建参数: DEPLOYMENT_ID',
' 默认值: 没写,构建时传',
' variable 102',
' 环境变量: DEPLOYMENT_ID',
' 值: DEPLOYMENT_ID(构建参数)',
' setting 104: 开放端口 8000',
' comment 106: 默认 CMD 走 api,celery worker/beat/flower 由 compose 覆盖 command',
' health 107',
' 间隔: 30 秒',
' 超时: 5 秒',
' 启动后宽限: 10 秒',
' 失败几次算不健康: 3 次',
' 检查命令: 请求 http://localhost:8000/health(失败就报错),失败就 退出,退出码 1',
' start 110',
' 命令: 用 uv 运行 uvicorn luffy_agent.main:app,监听 0.0.0.0:8000',
' 写法: 数组,不经过 shell',
]
describe('dockerfile 中文直译', () => {
it('luffy 前端 Dockerfile:解析指令、三个阶段、RUN 按命令拆、从阶段复制、连着写的 ENV / COPY 并表、构建参数、启动', async () => {
expect(await readLines(readFixture('clientNextjs.Dockerfile'))).toEqual(NEXT_READING)
})
it('luffy 后端 Dockerfile:阶段说明、标签表、10 条命令的 RUN(管道、重定向)、从镜像复制、12 条连着的 COPY、uv sync 的选项表、健康检查、启动', async () => {
expect(await readLines(readFixture('backend.Dockerfile'))).toEqual(BACKEND_READING)
})
it('阶段:平台、标签和摘要、接着前面的阶段建、没起名的按序号引用;基础镜像写成构建参数时注明', async () => {
expect(await readLines('FROM --platform=$BUILDPLATFORM golang:1.22@sha256:abc AS build\nFROM build\n')).toEqual([
'stage 1',
' 名称: build',
' 基础镜像: golang,标签 1.22,摘要 sha256:abc',
' 平台: BUILDPLATFORM',
'stage 2',
' 名称: 没起名,按序号 1 引用',
' 基础镜像: 阶段 build',
])
expect(await readLines(`ARG BASE=python:3.12\nFROM \${BASE}\n`)).toEqual([
'variable 1',
' 构建参数: BASE',
' 默认值: python:3.12',
'stage 2',
' 名称: 没起名,按序号 0 引用',
' 基础镜像: BASE(构建参数)',
])
})
it('变量:一条 ENV 写几个是「N 个」加一张表;没设时用默认值的展开、BuildKit 自动给的构建参数写成意思', async () => {
expect(await readLines(`FROM x\nENV A=1 B="two words" C=\${VERSION:-x} D=$TARGETARCH\n`)).toEqual([
'stage 1',
' 名称: 没起名,按序号 0 引用',
' 基础镜像: x',
' variable 2',
' 环境变量: 4 个',
' A | 1',
' B | two words',
' C | VERSION(没设就用 x)',
' D | TARGETARCH(目标架构)',
])
})
it('执行:挂载、网络的选项写成意思;数组写法交给命令规则;heredoc 按脚本分步;一条循环的子块归 RUN;&& 里带重定向的只算在最后一条上', async () => {
const source = [
'FROM x',
'RUN --mount=type=cache,target=/root/.cache/uv --mount=type=secret,id=npmrc --network=none uv sync',
'RUN ["apt-get", "update"]',
'RUN <<EOT',
'set -e',
'apt-get update',
'EOT',
'RUN for f in a b; do echo $f; done',
'RUN mkdir -p /x && echo hi > /x/a',
].join('\n')
expect((await readLines(source)).slice(3)).toEqual([
' run 2: uv 按 lock 同步依赖(缓存目录 /root/.cache/uv 跨构建保留、临时挂上密钥 npmrc、不联网)',
' run 3: 更新软件包列表',
' run 4: 执行',
' options 5',
' 出错就停: 是 -e',
' command 6: 更新软件包列表',
' run 8: 遍历 a、b 中的每个 f',
' command 8: 打印 f',
' run 9: 执行',
' command 9: 建目录 /x(连同上级目录)',
' command 9: 打印 "hi"(写到 /x/a)',
])
})
it('没有 shell 的语言插件时 RUN 照抄命令、没有子块', async () => {
expect((await read('FROM x\nRUN a && b\n', false))[0].children.map(block => viewLines([block]))).toEqual([['run 2: 执行 a && b']])
})
it('复制:heredoc 写入、ADD 下载网址 / 解压压缩包、选项的意思', async () => {
const source = 'FROM x\nCOPY <<EOF /etc/app.conf\nkey=value\nEOF\nADD https://example.com/a.tar.gz /tmp/\nADD --checksum=sha256:abc app.tar.gz /opt/\nCOPY --chmod=755 run.sh /usr/local/bin/\n'
expect((await readLines(source)).slice(3)).toEqual([
' copy 2: 写入 /etc/app.conf(内容写在下面)',
' copy 5: 下载 https://example.com/a.tar.gz → /tmp/',
' copy 6: 解压 app.tar.gz → /opt/(校验 sha256:abc)',
' copy 7: 复制 run.sh → /usr/local/bin/(权限 755)',
])
})
it('设置、健康检查、启动', async () => {
const source = [
'FROM x',
'EXPOSE 80/udp 443',
'VOLUME ["/data", "/logs"]',
'STOPSIGNAL SIGTERM',
'SHELL ["/bin/bash", "-o", "pipefail", "-c"]',
'USER 1000:1000',
'MAINTAINER someone',
'ONBUILD RUN make',
'HEALTHCHECK NONE',
'HEALTHCHECK --interval=1m30s --start-interval=500ms CMD ["curl", "-f", "http://x/"]',
'ENTRYPOINT ["/entrypoint.sh"]',
'CMD node server.js --port $PORT',
].join('\n')
expect((await readLines(source)).slice(3)).toEqual([
' setting 2: 开放端口 80(udp)、443',
' setting 3: 声明数据卷 /data、/logs',
' setting 4: 停止容器时发 SIGTERM 信号',
' setting 5: 之后的 RUN、CMD、ENTRYPOINT 用 /bin/bash -o pipefail -c 执行',
' setting 6: 之后以 1000:1000 身份运行',
' setting 7: 维护者 someone',
' setting 8: 以它为基础镜像构建时再执行',
' run 8: make',
' health 9',
' 检查: 不做健康检查(基础镜像里设的也关掉)',
' health 10',
' 间隔: 1 分 30 秒',
' 宽限期里的间隔: 500 毫秒',
' 检查命令: 请求 http://x/(失败就报错)',
' start 11',
' 入口命令: /entrypoint.sh',
' 写法: 数组,不经过 shell',
' start 12',
' 命令: node server.js --port PORT',
' 写法: 经过 shell',
])
})
it('能跳到定义的名字:从阶段复制的阶段名、表里的变量名带源码下标;RUN 的源码按 shell 着色', async () => {
const source = 'FROM node AS deps\nFROM node\nCOPY --from=deps /a /a\nENV A=1\nENV B=2\nRUN apt-get update\n'
const [, stage] = await read(source)
const [copy, env, run] = stage.children
const deps = 'translation' in copy && Array.isArray(copy.translation) ? flatSegments(copy.translation).find(segment => segment.text === 'deps') : undefined
expect(deps?.offset).toBe(source.indexOf('deps', 20))
expect('properties' in env && env.properties[0].items?.map(item => item.defines && source.slice(item.defines.startIndex, item.defines.endIndex))).toEqual(['A', 'B'])
expect(run.codeTokens.filter(token => token.role !== 'plain').map(token => [token.text, token.role])).toEqual([['RUN', 'keyword'], ['apt-get', 'function']])
})
})
@@ -0,0 +1,424 @@
import type { Inline, Node, PropertyItem, Segment, TokenRole } from '../contract'
import type { GroupReader, NodeReader, ReaderContext } from './zhTemplate'
import { createZhTemplateTranslator, embeddedText } from './zhTemplate'
/**
* Dockerfile 的中文直译:每条指令读成一行(RUN 里的 shell 由宿主交给 shell 的翻译插件读好,见 SourceDocument.embedded),
* 连着写的 COPY / LABEL 读成一行字加一张表。阶段、变量、健康检查、启动这些属性表块的行由引擎按 facts 排。
*/
const NAME = /^[A-Z_]\w*/i
const CONSTANT_NAME = /^[A-Z][A-Z0-9_]*$/
/** BuildKit 自动给的构建参数。 */
const PREDEFINED: Record<string, string> = {
TARGETPLATFORM: '目标平台',
TARGETOS: '目标系统',
TARGETARCH: '目标架构',
TARGETVARIANT: '目标架构变体',
BUILDPLATFORM: '构建机的平台',
BUILDOS: '构建机的系统',
BUILDARCH: '构建机的架构',
BUILDVARIANT: '构建机的架构变体',
}
function seg(text: string, role: TokenRole, offset?: number): Segment {
return { text, role, ...(offset !== undefined && { offset }) }
}
const hint = (text: string): Segment => seg(text, 'hint')
const word = (text: string): Segment => seg(text, 'keyword')
const verb = (text: string): Segment => seg(text, 'function')
const space = (): Segment => seg(' ', 'plain')
function named(node: Node | null | undefined): Node[] {
return node ? node.namedChildren.filter((child): child is Node => child !== null) : []
}
function joined(parts: Inline[][], separator: Inline): Inline[] {
return parts.flatMap((part, i) => i > 0 ? [separator, ...part] : part)
}
/** 括号里淡色写几条意思,用顿号隔开;一条都没有时什么都不写。 */
function notes(items: Inline[][]): Inline[] {
return items.length > 0 ? [hint('('), ...joined(items, hint('、')), hint(')')] : []
}
/** 指令后面的一个选项 `--名字=值`:值在源码里从哪开始。 */
interface Flag { name: string, value: string, valueStart: number }
function flagsOf(instruction: Node): Flag[] {
return named(instruction).flatMap((node): Flag[] => {
if (node.type !== 'param' && node.type !== 'mount_param')
return []
const match = /^--([a-z][-a-z]*)=(.*)$/s.exec(node.text)
return match ? [{ name: match[1], value: match[2], valueStart: node.startIndex + match[1].length + 3 }] : []
})
}
/** 节点所在的那条指令(父节点是根)。 */
function statementOf(node: Node): Node {
let current = node
while (current.parent?.parent)
current = current.parent
return current
}
/**
* 一段文字里夹着变量(`$HOME/x`、`"a ${B}"`):写死的部分按 role 着色(带源码下标),变量照常读;from / to 去掉引号、冒号这些。
*/
function mixed(node: Node, context: ReaderContext, role: TokenRole, from = node.startIndex, to = node.endIndex): Inline[] {
const { source } = context.document
const out: Inline[] = []
let cursor = from
for (const child of named(node)) {
if (child.type !== 'expansion')
continue
if (child.startIndex > cursor)
out.push(seg(source.slice(cursor, child.startIndex), role, cursor))
out.push(...context.say(child))
cursor = child.endIndex
}
if (cursor < to)
out.push(seg(source.slice(cursor, to), role, cursor))
return out
}
// ─── 变量 ────────────────────────────────────────────────────────────────────
/** 变量是在这条指令之前(同一阶段里;FROM 那一行看文件开头的)用 ARG 声明的构建参数,不是 ENV 设的环境变量。 */
function isBuildArgument(name: string, node: Node): boolean {
const statement = statementOf(node)
const inFrom = statement.type === 'from_instruction'
for (let previous = statement.previousNamedSibling; previous; previous = previous.previousNamedSibling) {
if (previous.type === 'from_instruction' && !inFrom)
return false
if (previous.type === 'arg_instruction' && previous.childForFieldName('name')?.text === name)
return true
if (previous.type === 'env_instruction' && named(previous).some(pair => pair.childForFieldName('name')?.text === name))
return false
}
return false
}
/**
* `$X`、`${X}`:变量名(带源码下标,全大写的按常量着色),后面淡色写它是什么:ARG 声明的写「构建参数」,BuildKit 自动给的写它的意思;
* `${X:-d}` 写「(没设就用 d)」,`${X:+d}` 写「(设了就是 d)」。
*/
const expansion: NodeReader = (node) => {
const variable = named(node).find(child => child.type === 'variable')
const name = variable && NAME.exec(variable.text)?.[0]
if (!variable || !name)
return undefined
const rest = variable.text.slice(name.length)
const modifier = /^:?([-+])(.*)$/s.exec(rest)
const origin = isBuildArgument(name, node) ? '构建参数' : PREDEFINED[name]
const remarks: Inline[][] = [
...(origin ? [[hint(origin)]] : []),
...(modifier ? [[hint(modifier[1] === '-' ? '没设就用 ' : '设了就是 '), seg(modifier[2], 'string')]] : []),
]
return [
seg(name, CONSTANT_NAME.test(name) ? 'constant' : 'variable', variable.startIndex),
...(remarks.length > 0 ? [hint('('), ...joined(remarks, hint(',')), hint(')')] : []),
]
}
/** 没引号的一段:数字按数字着色,夹着变量的拆开读;只是一个名字、一段文字的照常(名字能跳到定义)。 */
const unquoted: NodeReader = (node, context) => {
if (named(node).length > 0)
return mixed(node, context, 'string')
return /^\d+(?:\.\d+)?$/.test(node.text) ? [seg(node.text, 'number')] : undefined
}
/** 引号里的一段:去掉引号,写死的部分按字符串着色,变量照常读;空的写「空字符串」。 */
const quoted: NodeReader = (node, context) => node.endIndex - node.startIndex <= 2 ? [hint('空字符串')] : mixed(node, context, 'string', node.startIndex + 1, node.endIndex - 1)
/** JSON 字符串(数组写法里的一项):去掉引号、转义。 */
const jsonString: NodeReader = (node) => {
try {
const value: unknown = JSON.parse(node.text)
return typeof value === 'string' ? [seg(value, 'string', node.startIndex + 1)] : undefined
}
catch {
return undefined
}
}
// ─── 复制 ────────────────────────────────────────────────────────────────────
/** 名字是文件里某个阶段的名字(`AS deps`),或阶段的序号。 */
function isStage(name: string, node: Node): boolean {
return /^\d+$/.test(name) || node.tree.rootNode.descendantsOfType('image_alias').some(alias => alias?.text.toLowerCase() === name.toLowerCase())
}
/** `--from=` 的值:阶段写「阶段 X」(阶段名能跳过去),镜像写「镜像 名字:标签」。 */
function fromText(flag: Flag, node: Node): { stage: boolean, text: Inline[] } {
if (isStage(flag.value, node))
return { stage: true, text: [seg(flag.value, 'variable', flag.valueStart)] }
const at = flag.value.search(/[:@]/)
const text = at < 0
? [seg(flag.value, 'type', flag.valueStart)]
: [seg(flag.value.slice(0, at), 'type', flag.valueStart), seg(flag.value[at], 'plain'), seg(flag.value.slice(at + 1), 'string', flag.valueStart + at + 1)]
return { stage: false, text }
}
/** COPY / ADD 的选项写成意思(`--from` 另外写在前面)。 */
function copyNotes(flags: readonly Flag[], skipOwner: boolean): Inline[][] {
return flags.flatMap((flag): Inline[][] => {
switch (flag.name) {
case 'from':
return []
case 'chown':
return skipOwner ? [] : [[hint('属主 '), seg(flag.value, 'variable', flag.valueStart)]]
case 'chmod':
return [[hint('权限 '), seg(flag.value, 'string', flag.valueStart)]]
case 'parents':
return [[hint('保留目录层级')]]
case 'link':
return [[hint('单独成层')]]
case 'exclude':
return [[hint('跳过 '), seg(flag.value, 'string', flag.valueStart)]]
case 'checksum':
return [[hint('校验 '), seg(flag.value, 'string', flag.valueStart)]]
case 'keep-git-dir':
return [[hint('保留 .git 目录')]]
default:
return [[seg(`--${flag.name}=${flag.value}`, 'plain')]]
}
})
}
/** 一条 COPY / ADD 拆开的样子:从哪(阶段 / 镜像)、源、目标、选项,和动作(ADD 下载网址、解压本地的压缩包)。 */
interface CopyParts {
from?: { stage: boolean, text: Inline[] }
sources: Inline[]
target: Inline[]
flags: Flag[]
action: string
heredoc: boolean
}
const ARCHIVE = /\.(?:tar|tgz|tar\.gz|tar\.bz2|tbz2?|tar\.xz|txz)$/i
function copyParts(node: Node, context: ReaderContext): CopyParts {
const flags = flagsOf(node)
const from = flags.find(flag => flag.name === 'from')
const paths = named(node).filter(child => child.type === 'path')
const heredoc = paths.some(path => named(path).some(child => child.type === 'heredoc_marker'))
const sourcePaths = paths.slice(0, -1)
const target = paths.at(-1)
const texts = sourcePaths.map(path => path.text)
let action = '复制'
if (node.type === 'add_instruction') {
if (texts.every(text => /^https?:\/\//.test(text)))
action = '下载'
else if (texts.every(text => /^git@|\.git(?:#.*)?$/.test(text)))
action = '拉取仓库'
else if (texts.every(text => ARCHIVE.test(text)))
action = '解压'
}
return {
...(from && { from: fromText(from, node) }),
sources: joined(sourcePaths.map(path => path.text === '.' ? [hint('整个构建目录')] : context.say(path)), seg('、', 'plain')),
target: target ? target.text === '.' ? [seg('./', 'string', target.startIndex)] : context.say(target) : [],
flags,
action,
heredoc,
}
}
/** 一条 COPY / ADD:「[从阶段 X | 从镜像 Y] 复制 源 → 目标(属主 u:g)」;带 heredoc 的是「写入 目标(内容写在下面)」。 */
const copy: NodeReader = (node, context) => {
const parts = copyParts(node, context)
if (parts.heredoc)
return [verb('写入'), space(), ...parts.target, hint('(内容写在下面)'), ...notes(copyNotes(parts.flags, false))]
return [
...(parts.from ? [word(parts.from.stage ? '从阶段' : '从镜像'), space(), ...parts.from.text, space()] : []),
verb(parts.action),
space(),
...parts.sources,
space(),
word('→'),
space(),
...parts.target,
...notes(copyNotes(parts.flags, false)),
]
}
/**
* 连着写的几条 COPY(从同一个地方复制的):一行「复制(N 条连着写,属主都是 u:g)」,下面一张表「从 | 源 | 到」
* (从构建目录复制的没有「从」这一列),属主都一样才写在头一行,别的选项写在「到」那一格里。
*/
const copyGroup: GroupReader = (nodes, context) => {
const parts = nodes.map(node => copyParts(node, context))
const owners = new Set(nodes.map(node => flagsOf(node).find(flag => flag.name === 'chown')?.value))
const [owner] = owners
const shared = owners.size === 1 && owner !== undefined ? flagsOf(nodes[0]).find(flag => flag.name === 'chown') : undefined
const withFrom = parts.some(part => part.from)
const items: PropertyItem[] = parts.map(part => ({
cells: [
...(withFrom ? [part.from ? [word(part.from.stage ? '阶段' : '镜像'), space(), ...part.from.text] : [hint('构建目录')]] : []),
part.sources,
[...part.target, ...notes(copyNotes(part.flags, !!shared))],
],
}))
return {
text: [verb('复制'), hint(`(${nodes.length} 条连着写`), ...(shared ? [hint(',属主都是 '), seg(shared.value, 'variable', shared.valueStart)] : []), hint(')')],
table: { columns: [...(withFrom ? ['从'] : []), '源', '到'], items },
}
}
// ─── 标签 ────────────────────────────────────────────────────────────────────
function labelRows(nodes: Node[], context: ReaderContext): PropertyItem[] {
return nodes.flatMap(node => named(node).filter(child => child.type === 'label_pair')).map((pair) => {
const key = pair.childForFieldName('key')
const value = pair.childForFieldName('value')
return { cells: [key ? context.say(key) : [], value ? context.say(value) : [hint('空')]] }
})
}
/** LABEL:「标签」,下面一张「名称 | 值」表(连着写的几条并成一张)。 */
const labelGroup: GroupReader = (nodes, context) => ({
text: [word('标签'), ...(nodes.length > 1 ? [hint(`(${nodes.length} 条连着写)`)] : [])],
table: { columns: ['名称', '值'], items: labelRows(nodes, context) },
})
// ─── 执行、设置 ──────────────────────────────────────────────────────────────
/** `--mount=type=cache,target=…` 这类挂载的意思。 */
function mountNote(flag: Flag): Inline[] {
const fields = new Map(flag.value.split(',').map((pair) => {
const at = pair.indexOf('=')
return at < 0 ? [pair, ''] : [pair.slice(0, at), pair.slice(at + 1)]
}))
const target = fields.get('target') ?? fields.get('dst') ?? fields.get('destination')
const text = (value: string | undefined): Segment[] => value ? [seg(value, 'string')] : []
switch (fields.get('type') ?? 'bind') {
case 'cache':
return [hint('缓存目录 '), ...text(target), hint(' 跨构建保留')]
case 'secret':
return [hint('临时挂上密钥 '), ...text(fields.get('id') ?? target), ...(fields.get('env') ? [hint('(放进环境变量 '), ...text(fields.get('env')), hint(')')] : [])]
case 'ssh':
return [hint('转发 SSH 密钥')]
case 'tmpfs':
return [...text(target), hint(' 用内存临时目录')]
default: {
const from = fields.get('from')
return [hint('临时挂上 '), ...(from ? [hint('阶段 '), seg(from, 'variable'), hint(' 的 ')] : []), ...(fields.get('source') ? text(fields.get('source')) : [hint('构建目录')]), ...(target ? [hint(' 到 '), ...text(target)] : [])]
}
}
}
const NETWORK: Record<string, string> = { none: '不联网', host: '用宿主机的网络', default: '用默认网络' }
const SECURITY: Record<string, string> = { insecure: '不隔离权限', sandbox: '隔离权限' }
/** RUN 的选项:挂载、网络、权限。 */
function runNotes(node: Node): Inline[][] {
return flagsOf(node).map((flag) => {
if (flag.name === 'mount')
return mountNote(flag)
const known = flag.name === 'network' ? NETWORK[flag.value] : flag.name === 'security' ? SECURITY[flag.value] : undefined
return known ? [hint(known)] : [seg(`--${flag.name}=${flag.value}`, 'plain')]
})
}
/** 几条命令时写在「执行」后面的:3 条以上写条数,&& 连着的写「一条失败就停」。 */
const COUNTED_STEPS = 3
/**
* RUN:shell 只有一条命令(一步)时就是那条命令读成的一行(带它的表);几步时写「执行」,各步是子块。
* 数组写法的命令交给命令规则读。选项(挂载、网络)的意思写在括号里。
*/
const run: NodeReader = (node, context) => {
const flags = runNotes(node)
const shell = named(node).find(child => child.type === 'shell_command')
const reading = shell && context.document.embedded?.(shell)
if (reading?.whole) {
const { text, table } = embeddedText(reading)
return { text: [...text, ...notes(flags)], ...(table && { table }) }
}
if (reading) {
const count = reading.steps.length >= COUNTED_STEPS ? [[hint(`${reading.steps.length} 条命令${reading.chained ? ',一条失败就停' : ''}`)]] : []
return [word('执行'), ...notes([...count, ...flags])]
}
const array = named(node).find(child => child.type === 'json_string_array')
if (array)
return [...context.say(array), ...notes(flags)]
return [word('执行'), space(), seg(shell?.text ?? '', 'plain'), ...notes(flags)]
}
/** 指令关键字后面的原文(STOPSIGNAL、MAINTAINER 的值在语法树里没有单独的节点)。 */
function afterKeyword(node: Node): string {
return node.text.replace(/^\S+\s+/, '').trim()
}
/** 数组写法的各项、或空格隔开的几项,用 separator 隔开。 */
function items(node: Node, context: ReaderContext, types: string[], separator: Inline): Inline[] {
const array = named(node).find(child => child.type === 'json_string_array')
const parts = named(array ?? node).filter(child => types.includes(child.type))
return joined(parts.map(part => context.say(part)), separator)
}
const SETTINGS: Record<string, NodeReader> = {
workdir_instruction: (node, context) => [word('工作目录'), space(), ...items(node, context, ['path'], space())],
user_instruction: (node, context) => {
const user = node.childForFieldName('user')
const group = node.childForFieldName('group')
return user
? [word('之后以'), space(), context.name(user, 'variable'), ...(group ? [seg(':', 'plain'), context.name(group, 'variable')] : []), space(), word('身份运行')]
: undefined
},
expose_instruction: (node, context) => [word('开放端口'), space(), ...joined(named(node).map((port) => {
if (port.type !== 'expose_port')
return context.say(port)
const [number, protocol] = port.text.split('/')
return [seg(number, 'number'), ...(protocol ? [hint(`(${protocol})`)] : [])]
}), seg('、', 'plain'))],
volume_instruction: (node, context) => [word('声明数据卷'), space(), ...items(node, context, ['path', 'json_string'], seg('、', 'plain'))],
stopsignal_instruction: node => [word('停止容器时发'), space(), seg(afterKeyword(node), 'string'), space(), word('信号')],
shell_instruction: (node, context) => [word('之后的 RUN、CMD、ENTRYPOINT 用'), space(), ...items(node, context, ['json_string'], space()), space(), word('执行')],
maintainer_instruction: node => [word('维护者'), space(), seg(afterKeyword(node), 'string')],
onbuild_instruction: () => [word('以它为基础镜像构建时再执行')],
}
export const dockerfileZhTranslator = createZhTemplateTranslator({
id: 'dockerfile.zh-template',
label: '中文直译',
languageIds: ['dockerfile'],
keywords: {},
literals: {},
builtinTypes: {},
isTypePosition: () => false,
operators: {},
operatorParents: [],
emptyLiterals: {},
mapLiterals: {},
stringNodes: [],
commentText: (_node, text) => text.replace(/^#[ \t]?/, ''),
nodeTemplates: {},
readers: {
...SETTINGS,
expansion,
unquoted_string: unquoted,
double_quoted_string: quoted,
single_quoted_string: quoted,
json_string: jsonString,
path: (node, context) => named(node).length > 0 ? mixed(node, context, 'string') : [seg(node.text, 'string', node.startIndex)],
image_name: (node, context) => named(node).length > 0 ? mixed(node, context, 'type') : undefined,
image_tag: (node, context) => mixed(node, context, 'string', node.startIndex + 1),
image_digest: (node, context) => mixed(node, context, 'string', node.startIndex + 1),
run_instruction: run,
copy_instruction: copy,
add_instruction: copy,
label_instruction: (node, context) => labelGroup([node], context),
},
groups: {
copy_instruction: copyGroup,
label_instruction: labelGroup,
},
directives: {
syntax: { label: '语法', note: '(BuildKit 按这个版本读这份文件)' },
escape: { label: '转义字符' },
check: { label: '检查' },
},
})