fix(readme-links): preserve commas in srcset URLs (#36957)

Tokenize srcset like the HTML parser: read each URL up to ASCII
whitespace, strip trailing separator commas, and skip descriptors
(including parenthesized commas) before the next candidate.

Fixes #36955

Co-authored-by: Neil <neil@stably.ai>

config/scripts/check-readme-local-links.mjs

@@ -31,14 +31,47 @@ function trackedFiles(root, candidates) {
   return new Set(stdout.split('\0').filter(Boolean))
 }
 
+function* srcsetTargets(value) {
+  let position = 0
+  while (position < value.length) {
+    while (position < value.length && /[\t\n\f\r ,]/.test(value[position])) {
+      position++
+    }
+    if (position === value.length) {
+      return
+    }
+    const start = position
+    while (position < value.length && !/[\t\n\f\r ]/.test(value[position])) {
+      position++
+    }
+    let end = position
+    while (value[end - 1] === ',') {
+      end--
+    }
+    // HTML URL tokens can contain commas; trailing commas end a URL-only candidate.
+    yield value.slice(start, end)
+    if (end < position) {
+      continue
+    }
+    let inParentheses = false
+    while (position < value.length) {
+      const character = value[position++]
+      if (character === '(') {
+        inParentheses = true
+      } else if (character === ')') {
+        inParentheses = false
+      } else if (character === ',' && !inParentheses) {
+        break
+      }
+    }
+  }
+}
+
 function* localTargets(markdown) {
   for (const [, attribute, doubleQuoted, singleQuoted] of markdown.matchAll(HTML_ATTRIBUTE)) {
     const value = doubleQuoted ?? singleQuoted
     // Only srcset lists candidates; each begins with its URL. src/href hold one whole URL.
-    const targets =
-      attribute.toLowerCase() === 'srcset'
-        ? value.split(',').map((candidate) => candidate.trim().split(/\s+/)[0])
-        : [value.trim()]
+    const targets = attribute.toLowerCase() === 'srcset' ? srcsetTargets(value) : [value.trim()]
     for (const target of targets) {
       if (target) {
         yield target

config/scripts/check-readme-local-links.test.mjs

@@ -126,6 +126,48 @@ describe('README local link check', () => {
     ])
   })
 
+  it('preserves commas inside srcset URLs', () => {
+    const root = makeFixture({
+      ...validReadmes,
+      'README.md': '<source SrcSet=" ,,\t docs/Hero,wide.png 1x,, \n docs/Hero,tall.png,, ">',
+      'docs/Hero,wide.png': 'png',
+      'docs/Hero,tall.png': 'png'
+    })
+
+    expect(findBrokenReadmeLinks(root)).toEqual([])
+  })
+
+  it('rejects a missing srcset URL even when its comma-separated fragments are tracked', () => {
+    const root = makeFixture({
+      ...validReadmes,
+      'README.md': '<source srcset="docs/Missing,wide.png 1x">',
+      'docs/Missing': 'fragment',
+      'wide.png': 'fragment'
+    })
+
+    expect(findBrokenReadmeLinks(root).map((link) => link.target)).toEqual([
+      'docs/Missing,wide.png'
+    ])
+  })
+
+  it('ignores a srcset data URL while checking the following local candidate', () => {
+    const root = makeFixture({
+      ...validReadmes,
+      'README.md': '<source srcset="data:image/png;base64,AAAA 1x, docs/missing.png 2x">'
+    })
+
+    expect(findBrokenReadmeLinks(root).map((link) => link.target)).toEqual(['docs/missing.png'])
+  })
+
+  it('does not treat commas inside srcset descriptors as new candidates', () => {
+    const root = makeFixture({
+      ...validReadmes,
+      'README.md': '<source srcset="resources/build/icon.png future(1,2), docs/missing.png 2x">'
+    })
+
+    expect(findBrokenReadmeLinks(root).map((link) => link.target)).toEqual(['docs/missing.png'])
+  })
+
   it('checks tracked media outside the detector sparse checkout', () => {
     const root = makeFixture(validReadmes)
     const workflow = parse(readFileSync(path.join(projectDir, '.github/workflows/pr.yml'), 'utf8'))