feature: implement well formed JSON.stringify behavior

Unicode code points in the inclusive range 0xD800 to 0xDFFF are from
now on escaped as defined by the spec. This was not considered a
breaking change in the proposal and as such it's neither considered
a breaking change here.

Author: BridgeAR

CHANGELOG.md

@@ -4,6 +4,7 @@
 
 - Added `maximumBreadth` option to limit stringification at a specific object or array "width" (number of properties / values)
 - Added `maximumDepth` option to limit stringification at a specific nesting depth
+- Implemented the [well formed stringify proposal](https://github.com/tc39/proposal-well-formed-stringify) that is now part of the spec
 - Fixed maximum spacer length (10)
 - Fixed TypeScript definition
 - Fixed duplicated array replacer values serialized more than once

benchmark.js

@@ -2,7 +2,7 @@
 
 const Benchmark = require('benchmark')
 const suite = new Benchmark.Suite()
-const stringify = require('.')
+const stringify = require('.').configure({ deterministic: true })
 
 // eslint-disable-next-line
 const array = Array({ length: 10 }, (_, i) => i)

index.js

@@ -18,9 +18,9 @@ exports.configure = configure
 module.exports = stringify
 
 // eslint-disable-next-line
-const strEscapeSequencesRegExp = /[\x00-\x1f\x22\x5c]/
+const strEscapeSequencesRegExp = /[\u0000-\u001f\u0022\u005c\ud800-\udfff]/
 // eslint-disable-next-line
-const strEscapeSequencesReplacer = /[\x00-\x1f\x22\x5c]/g
+const strEscapeSequencesReplacer = /[\u0000-\u001f\u0022\u005c\ud800-\udfff]/g
 
 // Escaped special characters. Use empty strings to fill up unused entries.
 const meta = [
@@ -40,10 +40,14 @@ const meta = [
 ]
 
 function escapeFn (str) {
-  return meta[str.charCodeAt(0)]
+  const charCode = str.charCodeAt(0)
+  return meta.length > charCode
+    ? meta[charCode]
+    : `\\u${charCode.toString(16).padStart(4, '0')}`
 }
 
-// Escape control characters, double quotes and the backslash.
+// Escape C0 control characters, double quotes, the backslash and every code
+// unit with a numeric value in the inclusive range 0xD800 to 0xDFFF.
 function strEscape (str) {
   // Some magic numbers that worked out fine while benchmarking with v8 8.0
   if (str.length < 5000 && !strEscapeSequencesRegExp.test(str)) {
@@ -54,23 +58,17 @@ function strEscape (str) {
   }
   let result = ''
   let last = 0
-  let i = 0
-  for (; i < str.length; i++) {
+  for (let i = 0; i < str.length; i++) {
     const point = str.charCodeAt(i)
     if (point === 34 || point === 92 || point < 32) {
-      if (last === i) {
-        result += meta[point]
-      } else {
-        result += `${str.slice(last, i)}${meta[point]}`
-      }
+      result += `${str.slice(last, i)}${meta[point]}`
+      last = i + 1
+    } else if (point >= 55296 && point <= 57343) {
+      result += `${str.slice(last, i)}${`\\u${point.toString(16).padStart(4, '0')}`}`
       last = i + 1
     }
   }
-  if (last === 0) {
-    result = str
-  } else if (last !== i) {
-    result += str.slice(last)
-  }
+  result += str.slice(last)
   return result
 }
 

package.json

@@ -38,6 +38,7 @@
   "typings": "index.d.ts",
   "devDependencies": {
     "@types/json-stable-stringify": "^1.0.32",
+    "@types/node": "^16.11.1",
     "benchmark": "^2.1.4",
     "clone": "^2.1.2",
     "fast-json-stable-stringify": "^2.1.0",

test.js

@@ -995,3 +995,21 @@ test('should throw when maximumBreadth receives malformed input', (assert) => {
   })
   assert.end()
 })
+
+test('check for well formed stringify implementation', (assert) => {
+  for (let i = 0; i < 2 ** 16; i++) {
+    const string = String.fromCharCode(i)
+    const actual = stringify(string)
+    const expected = JSON.stringify(string)
+    // Older Node.js versions do not use the well formed JSON implementation.
+    if (Number(process.version.split('.')[0].slice(1)) >= 12 || i < 0xd800 || i > 0xdfff) {
+      assert.equal(actual, expected)
+    } else {
+      assert.not(actual, expected)
+    }
+  }
+  // Trigger special case
+  const longStringEscape = stringify(`${'a'.repeat(100)}\uD800`)
+  assert.equal(longStringEscape, `"${'a'.repeat(100)}\\ud800"`)
+  assert.end()
+})