fix(editor): avoid newline match arrays when saving LF Markdown (#24795)

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

src/renderer/src/components/editor/rich-markdown-source-eol-allocation.test.ts

@@ -0,0 +1,62 @@
+import { afterEach, describe, expect, it, vi } from 'vitest'
+import { RICH_MARKDOWN_MAX_SIZE_BYTES } from '../../../../shared/constants'
+import {
+  reconcileSerializedMarkdown,
+  restoreMarkdownSourceEol
+} from './rich-markdown-source-reconcile'
+
+describe('rich Markdown source line-ending allocation', () => {
+  afterEach(() => {
+    vi.restoreAllMocks()
+  })
+
+  it('serializes a supported LF edit without collecting a match for every newline', () => {
+    const source = `\`\`\`js\n${'x\n'.repeat(300_000)}const value=1;\n\`\`\`\n`
+    const canonical = source.slice(0, -1)
+    const edited = canonical.replace('value=1', 'value=2')
+    const roundTrip = vi.fn(() => null)
+    const originalMatch = RegExp.prototype[Symbol.match]
+    let largestGlobalMatchArray = 0
+    vi.spyOn(RegExp.prototype, Symbol.match).mockImplementation(function (
+      this: RegExp,
+      value: string
+    ) {
+      const result = originalMatch.call(this, value)
+      if (this.global && result) {
+        largestGlobalMatchArray = Math.max(largestGlobalMatchArray, result.length)
+      }
+      return result
+    })
+
+    expect(source.length).toBeLessThanOrEqual(RICH_MARKDOWN_MAX_SIZE_BYTES)
+    expect(
+      reconcileSerializedMarkdown({
+        originalSource: source,
+        baseCanonical: canonical,
+        edited,
+        roundTrip
+      })
+    ).toBe(`${edited}\n`)
+    expect(restoreMarkdownSourceEol(edited, source)).toBe(edited)
+    expect(roundTrip).not.toHaveBeenCalled()
+    expect(largestGlobalMatchArray).toBe(0)
+  })
+
+  it.each([
+    ['empty source', '', '\n'],
+    ['no line ending', 'text', '\n'],
+    ['LF only', 'a\nb\n', '\n'],
+    ['CRLF only', 'a\r\nb\r\n', '\r\n'],
+    ['lone CR only', 'a\rb\r', '\n'],
+    ['lone CR before CRLF', 'a\r\r\nb', '\r\n'],
+    ['CRLF first in a tie', 'a\r\nb\n', '\r\n'],
+    ['LF first in a tie', 'a\nb\r\n', '\r\n'],
+    ['LF majority', 'a\nb\r\nc\n', '\n'],
+    ['CRLF majority', 'a\r\nb\nc\r\n', '\r\n'],
+    ['Unicode separators', '\u2028\u2029', '\n'],
+    ['lone surrogates and NUL', '\ud800\n\udfff\0', '\n']
+  ])('preserves the existing decision for %s', (_name, source, eol) => {
+    const content = 'new\r\nline\n\ud800\udfff\0'
+    expect(restoreMarkdownSourceEol(content, source)).toBe(`new${eol}line${eol}\ud800\udfff\0`)
+  })
+})

src/renderer/src/components/editor/rich-markdown-source-reconcile.ts

@@ -119,6 +119,9 @@ function stripTrailingNewlines(lfText: string): string {
 }
 
 function detectDominantEol(text: string): '\n' | '\r\n' {
+  if (!text.includes('\r')) {
+    return '\n'
+  }
   const totalLf = (text.match(/\n/g) ?? []).length
   const crlf = (text.match(/\r\n/g) ?? []).length
   const lfOnly = totalLf - crlf