feat(pkg-manifest): preserve comments in json5 manifests (#5677)

Use npm package `strip-comments-strings` to find all comments in any
  manifest file as it is read. Save them as part of the "file formatting"
  detected by the manifest reader, noting for each the text of the lines
  they are on, before, and after, and the line number as a fallback.

  When the manifest is written, attempt to place each comment back in
  the resulting JSON5 text, so that the text of the line it is on,
  before, or after (in that priority order) matches the text at time of
  reading. Otherwise, so that no comments are lost, replace the comment
  on the same line number, adding a notation that it may have been
  relocated (due to sorting dependencies, for example, it may no longer
  be in the same "logical position" in the file, even though it is on
  the same physical line number).

  When comments are in fairly ordinary positions and the manifest does
  not change too drastically (i.e. once the dependencies are sorted as
  pnpm prefers, and there are not many parameters added all at once),
  this strategy results in exact preservation of the comments, as a new
  test shows.

  The motivation for this commit is to take advantage of the feature of
  JSON5 that it allows human-readable comments. For this feature to be
  useful in the case of package.json5 manifests, those comments must be
  preserved across manifest changes.

  Partially resolves #2008. [That issue requests comment preservation
  also for YAML manifests, but I have no experience working with YAML,
  and it also requests that key order be preserved, but I did not
  address key order because current code in the the pnpm manifest
  reader/writer _explicitly_ reorders keys -- clearly deliberately --
  so I did not want to simply remove code that appeared to have been
  purposefully written and included.]

Co-authored-by: Zoltan Kochan <z@kochan.io>

.changeset/famous-ants-lie.md

@@ -0,0 +1,5 @@
+---
+"@pnpm/text.comments-parser": major
+---
+
+Initial release.

.changeset/strong-houses-visit.md

@@ -0,0 +1,7 @@
+---
+"@pnpm/read-project-manifest": minor
+"@pnpm/write-project-manifest": minor
+"pnpm": patch
+---
+
+Comments in `package.json5` are preserver [#2008](https://github.com/pnpm/pnpm/issues/2008).

__typings__/typed.d.ts

@@ -45,6 +45,29 @@ declare module 'split-cmd' {
   export function splitToObject (cmd: string): { command: string, args: string[] }
 }
 
+declare module 'strip-comments-strings' {
+  export interface CodeItem {
+    // What feature of the code has been found:
+    type: string,
+    // The indices of the feature in the original code string:
+    index: number,
+    indexEnd: number,
+    // The test of the feature
+    content: string
+  }
+  export interface CodeAttributes {
+    // The remaining code text after all features have been stripped:
+    text: string,
+    // The items found:
+    comments: CodeItem[],
+    regexes: CodeItem[],
+    strings: CodeItem[]
+  }
+  export function parseString (str: string): CodeAttributes;
+  export type CodeItemReplacer = (item: CodeItem) => string;
+  export function stripComments (
+    str: string, replacer?: CodeItemReplacer): string;
+}
 
 declare module 'bin-links/lib/fix-bin' {
   function fixBin (path: string, execMode: number): Promise<void>;

pkg-manifest/read-project-manifest/fixtures/commented-package-json5/modified.json5

@@ -0,0 +1,9 @@
+/* This is an example of a package.json5 file with comments. */
+{
+    /* pnpm should keep comments at the same indentation level */
+    name: 'foo',
+    version: '1.0.0', // it should keep in-line comments on the same line
+    // It should allow in-line comments with no other content
+    type: 'commonjs',
+}
+/* And it should preserve comments at the end of the file. Note no newline. */

pkg-manifest/read-project-manifest/fixtures/commented-package-json5/package.json5

@@ -0,0 +1,9 @@
+/* This is an example of a package.json5 file with comments. */
+{
+    /* pnpm should keep comments at the same indentation level */
+    name: 'foo',
+    version: '1.0.0', // it should keep in-line comments on the same line
+    // It should allow in-line comments with no other content
+    type: 'module',
+}
+/* And it should preserve comments at the end of the file. Note no newline. */

pkg-manifest/read-project-manifest/package.json

@@ -32,6 +32,7 @@
     "@gwhitney/detect-indent": "7.0.1",
     "@pnpm/error": "workspace:*",
     "@pnpm/graceful-fs": "workspace:*",
+    "@pnpm/text.comments-parser": "workspace:*",
     "@pnpm/types": "workspace:*",
     "@pnpm/write-project-manifest": "workspace:*",
     "fast-deep-equal": "^3.1.3",

pkg-manifest/read-project-manifest/src/index.ts

@@ -2,9 +2,9 @@ import { promises as fs, Stats } from 'fs'
 import path from 'path'
 import { PnpmError } from '@pnpm/error'
 import { ProjectManifest } from '@pnpm/types'
+import { extractComments, CommentSpecifier } from '@pnpm/text.comments-parser'
 import { writeProjectManifest } from '@pnpm/write-project-manifest'
 import readYamlFile from 'read-yaml-file'
-
 import detectIndent from '@gwhitney/detect-indent'
 import equal from 'fast-deep-equal'
 import isWindows from 'is-windows'
@@ -76,7 +76,7 @@ export async function tryReadProjectManifest (projectDir: string): Promise<{
       fileName: 'package.json5',
       manifest: data,
       writeProjectManifest: createManifestWriter({
-        ...detectFileFormatting(text),
+        ...detectFileFormattingAndComments(text),
         initialManifest: data,
         manifestPath,
       }),
@@ -117,6 +117,15 @@ export async function tryReadProjectManifest (projectDir: string): Promise<{
   }
 }
 
+function detectFileFormattingAndComments (text: string) {
+  const { comments, text: newText, hasFinalNewline } = extractComments(text)
+  return {
+    comments,
+    indent: detectIndent(newText).indent,
+    insertFinalNewline: hasFinalNewline,
+  }
+}
+
 function detectFileFormatting (text: string) {
   return {
     indent: detectIndent(text).indent,
@@ -143,7 +152,7 @@ export async function readExactProjectManifest (manifestPath: string) {
     return {
       manifest: data,
       writeProjectManifest: createManifestWriter({
-        ...detectFileFormatting(text),
+        ...detectFileFormattingAndComments(text),
         initialManifest: data,
         manifestPath,
       }),
@@ -174,6 +183,7 @@ async function readPackageYaml (filePath: string) {
 function createManifestWriter (
   opts: {
     initialManifest: ProjectManifest
+    comments?: CommentSpecifier[]
     indent?: string | number | undefined
     insertFinalNewline?: boolean
     manifestPath: string
@@ -184,6 +194,7 @@ function createManifestWriter (
     updatedManifest = normalize(updatedManifest)
     if (force === true || !equal(initialManifest, updatedManifest)) {
       await writeProjectManifest(opts.manifestPath, updatedManifest, {
+        comments: opts.comments,
         indent: opts.indent,
         insertFinalNewline: opts.insertFinalNewline,
       })

pkg-manifest/read-project-manifest/test/index.ts

@@ -82,6 +82,26 @@ test('preserve space indentation in json5 file', async () => {
   expect(rawManifest).toBe("{\n  name: 'foo',\n  dependencies: {\n    bar: '1.0.0',\n  },\n}\n")
 })
 
+test('preserve comments in json5 file', async () => {
+  const originalManifest = await fs.readFile(
+    path.join(fixtures, 'commented-package-json5/package.json5'), 'utf8')
+  const modifiedManifest = await fs.readFile(
+    path.join(fixtures, 'commented-package-json5/modified.json5'), 'utf8')
+
+  process.chdir(tempy.directory())
+  await fs.writeFile('package.json5', originalManifest, 'utf8')
+
+  const { manifest, writeProjectManifest } = await readProjectManifest(process.cwd())
+
+  // Have to make a change to get it to write anything:
+  const newManifest = Object.assign({}, manifest, { type: 'commonjs' })
+
+  await writeProjectManifest(newManifest)
+
+  const resultingManifest = await fs.readFile('package.json5', 'utf8')
+  expect(resultingManifest).toBe(modifiedManifest)
+})
+
 test('do not save manifest if it had no changes', async () => {
   process.chdir(tempy.directory())
 

pkg-manifest/read-project-manifest/tsconfig.json

@@ -18,6 +18,9 @@
     {
       "path": "../../packages/types"
     },
+    {
+      "path": "../../text/comments-parser"
+    },
     {
       "path": "../write-project-manifest"
     }

pkg-manifest/write-project-manifest/package.json

@@ -29,6 +29,7 @@
   },
   "homepage": "https://github.com/pnpm/pnpm/blob/main/pkg-manifest/write-project-manifest#readme",
   "dependencies": {
+    "@pnpm/text.comments-parser": "workspace:*",
     "@pnpm/types": "workspace:*",
     "json5": "^2.2.1",
     "write-file-atomic": "^4.0.2",

pkg-manifest/write-project-manifest/src/index.ts

@@ -1,5 +1,6 @@
 import { promises as fs } from 'fs'
 import path from 'path'
+import { insertComments, CommentSpecifier } from '@pnpm/text.comments-parser'
 import { ProjectManifest } from '@pnpm/types'
 import JSON5 from 'json5'
 import writeFileAtomic from 'write-file-atomic'
@@ -14,6 +15,7 @@ export async function writeProjectManifest (
   filePath: string,
   manifest: ProjectManifest,
   opts?: {
+    comments?: CommentSpecifier[]
     indent?: string | number | undefined
     insertFinalNewline?: boolean
   }
@@ -25,9 +27,21 @@ export async function writeProjectManifest (
 
   await fs.mkdir(path.dirname(filePath), { recursive: true })
   const trailingNewline = opts?.insertFinalNewline === false ? '' : '\n'
+  const indent = opts?.indent ?? '\t'
 
-  const json = (fileType === 'json5' ? JSON5 : JSON)
-    .stringify(manifest, undefined, opts?.indent ?? '\t')
+  const json = (
+    fileType === 'json5'
+      ? stringifyJson5(manifest, indent, opts?.comments)
+      : JSON.stringify(manifest, undefined, indent)
+  )
 
   return writeFileAtomic(filePath, `${json}${trailingNewline}`)
 }
+
+function stringifyJson5 (obj: object, indent: string | number, comments?: CommentSpecifier[]) {
+  const json5 = JSON5.stringify(obj, undefined, indent)
+  if (comments) {
+    return insertComments(json5, comments)
+  }
+  return json5
+}

pkg-manifest/write-project-manifest/tsconfig.json

@@ -11,6 +11,9 @@
   "references": [
     {
       "path": "../../packages/types"
+    },
+    {
+      "path": "../../text/comments-parser"
     }
   ]
 }

pnpm-lock.yaml

@@ -3787,6 +3787,9 @@ importers:
       '@pnpm/graceful-fs':
         specifier: workspace:*
         version: link:../../fs/graceful-fs
+      '@pnpm/text.comments-parser':
+        specifier: workspace:*
+        version: link:../../text/comments-parser
       '@pnpm/types':
         specifier: workspace:*
         version: link:../../packages/types
@@ -3830,6 +3833,9 @@ importers:
 
   pkg-manifest/write-project-manifest:
     dependencies:
+      '@pnpm/text.comments-parser':
+        specifier: workspace:*
+        version: link:../../text/comments-parser
       '@pnpm/types':
         specifier: workspace:*
         version: link:../../packages/types
@@ -5577,6 +5583,16 @@ importers:
         specifier: ^3.0.2
         version: 3.0.2
 
+  text/comments-parser:
+    dependencies:
+      strip-comments-strings:
+        specifier: 1.2.0
+        version: 1.2.0
+    devDependencies:
+      '@pnpm/text.comments-parser':
+        specifier: workspace:*
+        version: 'link:'
+
   workspace/filter-workspace-packages:
     dependencies:
       '@pnpm/error':
@@ -16100,6 +16116,10 @@ packages:
     resolution: {integrity: sha512-3xurFv5tEgii33Zi8Jtp55wEIILR9eh34FAW00PZf+JnSsTmV/ioewSgQl97JHvgjoRGwPShsWm+IdrxB35d0w==}
     engines: {node: '>=8'}
 
+  /strip-comments-strings/1.2.0:
+    resolution: {integrity: sha512-zwF4bmnyEjZwRhaak9jUWNxc0DoeKBJ7lwSN/LEc8dQXZcUFG6auaaTQJokQWXopLdM3iTx01nQT8E4aL29DAQ==}
+    dev: false
+
   /strip-final-newline/2.0.0:
     resolution: {integrity: sha512-BrpvfNAE3dcvq7ll3xVumzjKjZQ5tI1sEUIKr3Uoks0XUl45St3FlatVqef9prk4jRDzhW6WZg+3bk93y6pLjA==}
     engines: {node: '>=6'}
@@ -17589,6 +17609,7 @@ time:
   /string.prototype.replaceall/1.0.6: '2021-10-05T01:00:00.092Z'
   /strip-ansi/6.0.1: '2021-09-23T16:34:41.798Z'
   /strip-bom/4.0.0: '2019-04-28T04:40:47.887Z'
+  /strip-comments-strings/1.2.0: '2022-06-12T23:34:53.852Z'
   /symlink-dir/5.1.0: '2022-11-20T16:48:15.918Z'
   /syncpack/8.3.9: '2022-10-28T16:58:42.312Z'
   /tar-stream/2.2.0: '2020-12-29T10:22:57.508Z'

pnpm-workspace.yaml

@@ -20,6 +20,7 @@ packages:
   - resolving/*
   - reviewing/*
   - store/*
+  - text/*
   - workspace/*
   - "!**/example/**"
   - "!**/test/**"

text/comments-parser/README.md

@@ -0,0 +1,14 @@
+# @pnpm/text.comments-parser
+
+> Extracts and inserts comments from/to text
+
+## Installation
+
+```
+pnpm i @pnpm/text.comments-parser
+
+```
+
+## License
+
+[MIT](LICENSE)

text/comments-parser/jest.config.js

@@ -0,0 +1 @@
+module.exports = require('../../jest.config.js')

text/comments-parser/package.json

@@ -0,0 +1,41 @@
+{
+  "name": "@pnpm/text.comments-parser",
+  "description": "Extracts and inserts comments from/to text",
+  "version": "0.0.0",
+  "bugs": {
+    "url": "https://github.com/pnpm/pnpm/issues"
+  },
+  "main": "lib/index.js",
+  "types": "lib/index.d.ts",
+  "files": [
+    "lib",
+    "!*.map"
+  ],
+  "keywords": [
+    "pnpm7"
+  ],
+  "license": "MIT",
+  "engines": {
+    "node": ">=14.6"
+  },
+  "repository": "https://github.com/pnpm/pnpm/blob/main/text/comments-parser",
+  "scripts": {
+    "start": "tsc --watch",
+    "_test": "jest",
+    "test": "pnpm run compile && pnpm run _test",
+    "lint": "eslint src/**/*.ts test/**/*.ts",
+    "prepublishOnly": "pnpm run compile",
+    "compile": "tsc --build && pnpm run lint --fix"
+  },
+  "dependencies": {
+    "strip-comments-strings": "1.2.0"
+  },
+  "homepage": "https://github.com/pnpm/pnpm/blob/main/text/comments-parser#readme",
+  "funding": "https://opencollective.com/pnpm",
+  "devDependencies": {
+    "@pnpm/text.comments-parser": "workspace:*"
+  },
+  "exports": {
+    ".": "./lib/index.js"
+  }
+}

text/comments-parser/src/CommentSpecifier.ts

@@ -0,0 +1,9 @@
+export interface CommentSpecifier {
+  type: string
+  content: string
+  lineNumber: number
+  after?: string
+  on: string
+  whitespace: string
+  before?: string
+}

text/comments-parser/src/extractComments.ts

@@ -0,0 +1,72 @@
+import { parseString, stripComments } from 'strip-comments-strings'
+import { CommentSpecifier } from './CommentSpecifier'
+
+export function extractComments (text: string) {
+  const hasFinalNewline = text.endsWith('\n')
+  if (!hasFinalNewline) {
+    /* For the sake of the comment parser, which otherwise loses the
+     * final character of a final comment
+     */
+    text += '\n'
+  }
+  const { comments: rawComments } = parseString(text)
+  const comments: CommentSpecifier[] = []
+  let stripped = stripComments(text)
+  if (!hasFinalNewline) {
+    stripped = stripped.slice(0, -1)
+  }
+  let offset = 0 // accumulates difference of indices from text to stripped
+  for (const comment of rawComments) {
+    /* Extract much more context for the comment needed to restore it later */
+    // Unfortunately, JavaScript lastIndexOf does not have an end parameter:
+    const preamble: string = stripped.slice(0, comment.index - offset)
+    const lineStart = Math.max(preamble.lastIndexOf('\n'), 0)
+    const priorLines = preamble.split('\n')
+    let lineNumber = priorLines.length
+    let after = ''
+    let hasAfter = false
+    if (lineNumber === 1) {
+      if (preamble.trim().length === 0) {
+        lineNumber = 0
+      }
+    } else {
+      after = priorLines[lineNumber - 2]
+      hasAfter = true
+      if (priorLines[0].trim().length === 0) {
+        /* JSON5.stringify will not have a whitespace-only line at the start */
+        lineNumber -= 1
+      }
+    }
+    let lineEnd = stripped.indexOf(
+      '\n', (lineStart === 0) ? 0 : lineStart + 1)
+    if (lineEnd < 0) {
+      lineEnd = stripped.length
+    }
+    const whitespaceMatch = stripped
+      .slice(lineStart, comment.index - offset)
+      .match(/^\s*/)
+
+    const newComment: CommentSpecifier = {
+      type: comment.type,
+      content: comment.content,
+      lineNumber,
+      on: stripped.slice(lineStart, lineEnd),
+      whitespace: whitespaceMatch ? whitespaceMatch[0] : '',
+    }
+
+    if (hasAfter) {
+      newComment.after = after
+    }
+    const nextLineEnd = stripped.indexOf('\n', lineEnd + 1)
+    if (nextLineEnd >= 0) {
+      newComment.before = stripped.slice(lineEnd, nextLineEnd)
+    }
+    comments.push(newComment)
+    offset += comment.indexEnd - comment.index
+  }
+  return {
+    text: stripped,
+    comments: comments.length ? comments : undefined,
+    hasFinalNewline,
+  }
+}

text/comments-parser/src/index.ts

@@ -0,0 +1,3 @@
+export * from './extractComments'
+export * from './insertComments'
+export * from './CommentSpecifier'

text/comments-parser/src/insertComments.ts

@@ -0,0 +1,83 @@
+import { CommentSpecifier } from './CommentSpecifier'
+
+export function insertComments (json: string, comments: CommentSpecifier[]) {
+  // We need to reintroduce the comments. So create an index of
+  // the lines of the manifest so we can try to match them up.
+  // We eliminate whitespace and quotes in the index entries,
+  // because pnpm may have changed them.
+  const jsonLines = json.split('\n')
+  const index = {}
+  const canonicalizer = /[\s'"]/g
+  for (let i = 0; i < jsonLines.length; ++i) {
+    const key = jsonLines[i].replace(canonicalizer, '')
+    if (key in index) {
+      index[key] = -1 // Mark this line as occurring twice
+    } else {
+      index[key] = i
+    }
+  }
+
+  // A place to put comments that come _before_ the lines they are
+  // anchored to:
+  const jsonPrefix: Record<string, string> = {}
+  for (const comment of comments) {
+    // First if we can find the line the comment was on, that is
+    // the most reliable locator:
+    let key = comment.on.replace(canonicalizer, '')
+    if (key && index[key] !== undefined && index[key] >= 0) {
+      jsonLines[index[key]] += ' ' + comment.content
+      continue
+    }
+    // Next, if it's not before anything, it must have been at the very end:
+    if (comment.before === undefined) {
+      jsonLines[jsonLines.length - 1] += comment.whitespace + comment.content
+      continue
+    }
+    // Next, try to put it before something; note the comment extractor
+    // used the convention that position 0 is before the first line:
+    let location = (comment.lineNumber === 0) ? 0 : -1
+    if (location < 0) {
+      key = comment.before.replace(canonicalizer, '')
+      if (key && index[key] !== undefined) {
+        location = index[key]
+      }
+    }
+    if (location >= 0) {
+      if (jsonPrefix[location]) {
+        jsonPrefix[location] += ' ' + comment.content
+      } else {
+        const inlineWhitespace = comment.whitespace.startsWith('\n')
+          ? comment.whitespace.slice(1)
+          : comment.whitespace
+        jsonPrefix[location] = inlineWhitespace + comment.content
+      }
+      continue
+    }
+    // The last definite indicator we can use is that it is after something:
+    if (comment.after) {
+      key = comment.after.replace(canonicalizer, '')
+      if (key && index[key] !== undefined && index[key] >= 0) {
+        jsonLines[index[key]] += comment.whitespace + comment.content
+        continue
+      }
+    }
+    // Finally, try to get it in the right general location by using the
+    // line number, but warn the user the comment may have been relocated:
+    location = comment.lineNumber - 1 // 0 was handled above
+    let separator = ' '
+    if (location >= jsonLines.length) {
+      location = jsonLines.length - 1
+      separator = '\n'
+    }
+    jsonLines[location] += separator + comment.content +
+      ' /* [comment possibly relocated by pnpm] */'
+  }
+  // Insert the accumulated prefixes:
+  for (let i = 0; i < jsonLines.length; ++i) {
+    if (jsonPrefix[i]) {
+      jsonLines[i] = jsonPrefix[i] + '\n' + jsonLines[i]
+    }
+  }
+  // And reassemble the manifest:
+  return jsonLines.join('\n')
+}

text/comments-parser/test/index.ts

@@ -0,0 +1,19 @@
+import { extractComments, insertComments } from '@pnpm/text.comments-parser'
+
+test('extract and insert JSON5 comments', () => {
+  const json5WithComments = `/* This is an example of a package.json5 file with comments. */
+{
+    /* pnpm should keep comments at the same indentation level */
+    name: 'foo',
+    version: '1.0.0', // it should keep in-line comments on the same line
+    // It should allow in-line comments with no other content
+    type: 'commonjs',
+}
+/* And it should preserve comments at the end of the file. Note no newline. */`
+  const { comments } = extractComments(json5WithComments)
+  expect(insertComments(`{
+    name: 'foo',
+    version: '1.0.0',
+    type: 'commonjs',
+}`, comments!))
+})

text/comments-parser/tsconfig.json

@@ -0,0 +1,12 @@
+{
+  "extends": "@pnpm/tsconfig",
+  "compilerOptions": {
+    "outDir": "lib",
+    "rootDir": "src"
+  },
+  "include": [
+    "src/**/*.ts",
+    "../../__typings__/**/*.d.ts"
+  ],
+  "references": []
+}

text/comments-parser/tsconfig.lint.json

@@ -0,0 +1,8 @@
+{
+  "extends": "./tsconfig.json",
+  "include": [
+    "src/**/*.ts",
+    "test/**/*.ts",
+    "../../__typings__/**/*.d.ts"
+  ]
+}