Answer the stability review: the parse layer strips a space or a tab, never more

src/markdown/parse/blocks.test.ts

@@ -24,6 +24,14 @@ test('keeps the link reference definitions a paragraph gives up, the first of a
   assert.deepEqual(definitions('[a]: /url\\(x\n'), [['a', { destination: '/url\\(x' }]])
   assert.deepEqual(definitions('[a]: <>\n'), [['a', { destination: '' }]])
   assert.deepEqual(definitions('[a]: /url "He said \\"hi\\""\n'), [['a', { destination: '/url', title: 'He said \\"hi\\"' }]])
+  assert.deepEqual(definitions('[a]: /url\\\n[b]: /b\n'), [
+    ['a', { destination: '/url\\' }],
+    ['b', { destination: '/b' }],
+  ])
+  assert.deepEqual(definitions('[\u00a0a]: /one\n[a]: /two\n'), [
+    ['\u00a0a', { destination: '/one' }],
+    ['a', { destination: '/two' }],
+  ])
 })
 
 test('leaves the paragraph a line no definition spells', () => {

src/markdown/parse/blocks.ts

@@ -70,7 +70,6 @@ function openBlock(walk: Walk, lines: readonly string[], index: number, line: st
   return readHtmlBlock(walk, lines, index, html)
 }
 
-// The document's last line ending closes its line rather than opening an empty one.
 function normalizeInput(markdown: string): string {
   return markdown
     .replace(/\r\n?/g, '\n')

src/markdown/parse/link-reference-definitions.ts

@@ -1,11 +1,11 @@
-import { holdsControlCharacter } from '../commonmark-grammar.ts'
+import { holdsControlCharacter, isAsciiPunctuation } from '../commonmark-grammar.ts'
 
 export type LinkDefinition = { destination: string; title?: string }
 
 type ReadDefinition = { definition: LinkDefinition; label: string; length: number }
 type ReadValue = { length: number; value: string }
 
-const bracketedDestination = /^<((?:[^\n<>\\]|\\[\s\S])*)>/
+const bracketedDestination = /^<((?:[^\n<>\\]|\\[^\n])*)>/
 const label = /^\[((?:[^[\]\\]|\\[\s\S]){1,999})\]:/
 const restOfLine = /^[ \t]*(?:\n|$)/
 const titleClosers: Readonly<Record<string, string>> = { '"': '"', "'": "'", '(': ')' }
@@ -24,13 +24,12 @@ export function readLinkDefinitions(definitions: Map<string, LinkDefinition>, te
 function readDefinition(text: string): ReadDefinition | undefined {
   const matched = label.exec(text)
   if (matched === null) return undefined
-  const raw = matched[1] ?? ''
-  if (raw.trim() === '') return undefined
+  const name = normalizeLabel(matched[1] ?? '')
+  if (name === '') return undefined
   const afterLabel = skipSpace(text, matched[0].length)
   const destination = readDestination(text, afterLabel)
   if (destination === undefined) return undefined
   const afterDestination = afterLabel + destination.length
-  const name = raw.replace(/[ \t\n]+/g, ' ').trim().toLowerCase()
   const titled = readTitledEnd(text, afterDestination)
   if (titled !== undefined) return { definition: { destination: destination.value, title: titled.value }, label: name, length: titled.length }
   const plain = endOfLine(text, afterDestination)
@@ -38,6 +37,14 @@ function readDefinition(text: string): ReadDefinition | undefined {
   return { definition: { destination: destination.value }, label: name, length: plain }
 }
 
+// CommonMark's label matching: the whitespace a label holds collapses, and its case folds.
+function normalizeLabel(raw: string): string {
+  return raw
+    .replace(/^[ \t\n]+|[ \t\n]+$/g, '')
+    .replace(/[ \t\n]+/g, ' ')
+    .toLowerCase()
+}
+
 function readTitledEnd(text: string, offset: number): ReadValue | undefined {
   const afterSpace = skipSpace(text, offset)
   if (afterSpace === offset) return undefined
@@ -56,7 +63,7 @@ function readDestination(text: string, offset: number): ReadValue | undefined {
   while (index < text.length) {
     const character = text.charAt(index)
     if (character === ' ' || holdsControlCharacter(character)) break
-    if (character === '\\') {
+    if (escapesNext(text, index)) {
       index += 2
       continue
     }
@@ -77,7 +84,7 @@ function readTitle(text: string, offset: number): ReadValue | undefined {
   let index = offset + 1
   while (index < text.length) {
     const character = text.charAt(index)
-    if (character === '\\') {
+    if (escapesNext(text, index)) {
       index += 2
       continue
     }
@@ -88,6 +95,11 @@ function readTitle(text: string, offset: number): ReadValue | undefined {
   return undefined
 }
 
+// A backslash escapes ASCII punctuation only, so a line ending always ends the destination it follows.
+function escapesNext(text: string, index: number): boolean {
+  return text.charAt(index) === '\\' && isAsciiPunctuation(text.charAt(index + 1))
+}
+
 // The label, the destination and the title each take at most one line ending with them.
 function skipSpace(text: string, offset: number): number {
   const rest = text.slice(offset)

src/markdown/parse/markdown-to-adf.test.ts

@@ -129,8 +129,14 @@ test('gives up the link reference definitions a paragraph opens with', () => {
   assert.deepEqual(content(markdownToAdf('[a]: /url\n===\n')), [paragraph('===')])
 })
 
+test('keeps the whitespace CommonMark strips no more of than a space or a tab', () => {
+  assert.deepEqual(content(markdownToAdf('\u00a0Part.\u00a0\n')), [paragraph('\u00a0Part.\u00a0')])
+  assert.deepEqual(content(markdownToAdf('  \u3000Part.\t\n')), [paragraph('\u3000Part.')])
+})
+
 test('normalizes the line endings and the null character CommonMark replaces', () => {
   assert.deepEqual(content(markdownToAdf('One\r\ntwo.\r\n')), [paragraph('One two.')])
+  assert.deepEqual(content(markdownToAdf('One\rtwo.\r')), [paragraph('One two.')])
   assert.deepEqual(content(markdownToAdf('```\r\nx\r\n```\r\n')), [{ content: [text('x')], type: 'codeBlock' }])
   assert.deepEqual(content(markdownToAdf('a\u0000b\n')), [paragraph('a\ufffdb')])
 })

src/markdown/parse/markdown-to-adf.ts

@@ -2,6 +2,7 @@ import type { AdfDocument, AdfNode } from '../../adf/document.ts'
 import type { ClaimedConstruct, LeafBlock } from './blocks.ts'
 import { failure, success, type ConvertErrorPath, type Result } from '../../result.ts'
 import { parseBlocks } from './blocks.ts'
+import { trimSpace } from '../commonmark-grammar.ts'
 
 export function markdownToAdf(markdown: string): Result<AdfDocument> {
   const content: AdfNode[] = []
@@ -40,7 +41,7 @@ function withContent(node: AdfNode, text: string): AdfNode {
 function inlineContent(text: string): AdfNode[] {
   const line = text
     .split('\n')
-    .map((part) => part.trim())
+    .map((part) => trimSpace(part))
     .join(' ')
   return line === '' ? [] : [{ text: line, type: 'text' }]
 }