Fix impossible atx headings to use setext headings

This changes several cases where ATX headings—which are formed on a
single line—could not be generated safely because of a descendant
which included a line break that could not be escaped.
Headings with a rank of 1 or 2, which include such a line break, are
now generates in setext style—even without an explicit `setext: true`.

Closes GH-26.

Author: wooorm

lib/handle/heading.js

@@ -13,17 +13,11 @@ import {containerPhrasing} from '../util/container-phrasing.js'
  */
 export function heading(node, _, context) {
   const rank = Math.max(Math.min(6, node.depth || 1), 1)
-  /** @type {Exit} */
-  let exit
-  /** @type {Exit} */
-  let subexit
-  /** @type {string} */
-  let value
 
   if (formatHeadingAsSetext(node, context)) {
-    exit = context.enter('headingSetext')
-    subexit = context.enter('phrasing')
-    value = containerPhrasing(node, context, {before: '\n', after: '\n'})
+    const exit = context.enter('headingSetext')
+    const subexit = context.enter('phrasing')
+    const value = containerPhrasing(node, context, {before: '\n', after: '\n'})
     subexit()
     exit()
 
@@ -41,10 +35,12 @@ export function heading(node, _, context) {
   }
 
   const sequence = '#'.repeat(rank)
-  exit = context.enter('headingAtx')
-  subexit = context.enter('phrasing')
-  value = containerPhrasing(node, context, {before: '# ', after: '\n'})
+  const exit = context.enter('headingAtx')
+  const subexit = context.enter('phrasing')
+  let value = containerPhrasing(node, context, {before: '# ', after: '\n'})
+
   value = value ? sequence + ' ' + value : sequence
+
   if (context.options.closeAtx) {
     value += ' ' + sequence
   }

lib/util/format-heading-as-setext.js

@@ -3,6 +3,7 @@
  * @typedef {import('../types.js').Context} Context
  */
 
+import {visit, EXIT} from 'unist-util-visit'
 import {toString} from 'mdast-util-to-string'
 
 /**
@@ -11,7 +12,24 @@ import {toString} from 'mdast-util-to-string'
  * @returns {boolean}
  */
 export function formatHeadingAsSetext(node, context) {
+  let literalWithBreak = false
+
+  // Look for literals with a line break.
+  // Note that this also
+  visit(node, (node) => {
+    console.log('n:', node.type)
+    if (
+      ('value' in node && /\r?\n|\r/.test(node.value)) ||
+      node.type === 'break'
+    ) {
+      literalWithBreak = true
+      return EXIT
+    }
+  })
+
   return Boolean(
-    context.options.setext && (!node.depth || node.depth < 3) && toString(node)
+    (!node.depth || node.depth < 3) &&
+      toString(node) &&
+      (context.options.setext || literalWithBreak)
   )
 }

package.json

@@ -43,6 +43,7 @@
     "longest-streak": "^3.0.0",
     "mdast-util-to-string": "^3.0.0",
     "parse-entities": "^3.0.0",
+    "unist-util-visit": "^4.0.0",
     "zwitch": "^2.0.0"
   },
   "devDependencies": {

test/index.js

@@ -385,12 +385,12 @@ test('blockquote', (t) => {
       children: [
         {
           type: 'heading',
-          depth: 1,
+          depth: 3,
           children: [{type: 'text', value: 'a\nb'}]
         }
       ]
     }),
-    '> # a
b\n',
+    '> ### a
b\n',
     'should support a heading (atx) in a block quote'
   )
 
@@ -597,15 +597,31 @@ test('break', (t) => {
   t.equal(to({type: 'break'}), '\\\n', 'should support a break')
 
   t.equal(
-    to(from('a  \nb\n=\n')),
-    '# a b\n',
+    to({
+      type: 'heading',
+      depth: 3,
+      children: [
+        {type: 'text', value: 'a'},
+        {type: 'break'},
+        {type: 'text', value: 'b'}
+      ]
+    }),
+    '### a b\n',
     'should serialize breaks in heading (atx) as a space'
   )
 
   t.equal(
-    to(from('a \\\nb\n=\n')),
-    '# a b\n',
-    'should serialize breaks in heading (atx) as nothing when preceded by a space'
+    to({
+      type: 'heading',
+      depth: 3,
+      children: [
+        {type: 'text', value: 'a '},
+        {type: 'break'},
+        {type: 'text', value: 'b'}
+      ]
+    }),
+    '### a b\n',
+    'should serialize breaks in heading (atx) as a space'
   )
 
   t.equal(
@@ -1080,6 +1096,50 @@ test('heading', (t) => {
     'should serialize an empty heading w/ rank 2 as atx when `setext: true`'
   )
 
+  t.equal(
+    to({
+      type: 'heading',
+      depth: 1,
+      children: [{type: 'inlineCode', value: '\n'}]
+    }),
+    '`\n`\n=\n',
+    'should serialize an heading w/ rank 1 and code w/ a line ending as setext'
+  )
+
+  t.equal(
+    to({
+      type: 'heading',
+      depth: 1,
+      children: [{type: 'html', value: '<a\n/>'}]
+    }),
+    '<a\n/>\n==\n',
+    'should serialize an heading w/ rank 1 and html w/ a line ending as setext'
+  )
+
+  t.equal(
+    to({
+      type: 'heading',
+      depth: 1,
+      children: [{type: 'text', value: 'a\nb'}]
+    }),
+    'a\nb\n=\n',
+    'should serialize an heading w/ rank 1 and text w/ a line ending as setext'
+  )
+
+  t.equal(
+    to({
+      type: 'heading',
+      depth: 1,
+      children: [
+        {type: 'text', value: 'a'},
+        {type: 'break'},
+        {type: 'text', value: 'b'}
+      ]
+    }),
+    'a\\\nb\n=\n',
+    'should serialize an heading w/ rank 1 and a break as setext'
+  )
+
   t.equal(
     // @ts-expect-error: `children` missing.
     to({type: 'heading'}, {closeAtx: true}),