fix(desktop): skip markdown tables in speech output

apps/desktop/src/lib/speech-text.test.ts

@@ -14,4 +14,141 @@ describe('sanitizeTextForSpeech', () => {
       'Use git status after the change.'
     )
   })
+
+  it('skips markdown table data while preserving surrounding human text', () => {
+    const text = `Here is the quick takeaway: the totals remain unchanged.
+
+| Item | Value | Notes |
+| --- | ---: | --- |
+| Example A | 10 | first row |
+| Example B | 20 | second row |
+
+Full detail stays visible on screen.`
+
+    expect(sanitizeTextForSpeech(text)).toBe(
+      'Here is the quick takeaway: the totals remain unchanged. Full detail stays visible on screen.'
+    )
+  })
+
+  it('does not strip prose that merely contains a pipe character', () => {
+    const text = 'Use the summary first | keep the table on screen when it matters.'
+
+    expect(sanitizeTextForSpeech(text)).toBe('Use the summary first | keep the table on screen when it matters.')
+  })
+
+  it('does not duplicate punctuation across paragraph breaks', () => {
+    const text = `First sentence.
+
+Second sentence.`
+
+    expect(sanitizeTextForSpeech(text)).toBe('First sentence. Second sentence.')
+  })
+
+  it.each([
+    ['markdown emphasis', '**First sentence.**\n\nSecond sentence.', 'First sentence. Second sentence.'],
+    ['a closing quote', '“First sentence.”\n\nSecond sentence.', '“First sentence.” Second sentence.'],
+    ['a closing parenthesis', '(First sentence.)\n\nSecond sentence.', '(First sentence.) Second sentence.']
+  ])('does not duplicate punctuation after %s', (_label, text, expected) => {
+    expect(sanitizeTextForSpeech(text)).toBe(expected)
+  })
+
+  it('skips markdown tables without leading and trailing pipes', () => {
+    const text = `Main takeaway: total is unchanged.
+
+Item | Value
+--- | ---:
+Example A | 10
+Example B | 20
+
+Done.`
+
+    expect(sanitizeTextForSpeech(text)).toBe('Main takeaway: total is unchanged. Done.')
+  })
+
+  it('skips markdown tables nested inside blockquotes', () => {
+    const text = `Before the table.
+
+> | Item | Value |
+> | --- | ---: |
+> | Example A | 10 |
+> | Example B | 20 |
+
+After the table.`
+
+    expect(sanitizeTextForSpeech(text)).toBe('Before the table. After the table.')
+  })
+
+  it('allows marker padding plus three spaces in blockquoted tables', () => {
+    const text = `Before the table.
+
+>    | Item | Value |
+>    | --- | ---: |
+>    | Example A | 10 |
+
+After the table.`
+
+    expect(sanitizeTextForSpeech(text)).toBe('Before the table. After the table.')
+  })
+
+  it('skips explicit single-column markdown tables', () => {
+    const text = `Before the table.
+
+| Item |
+| --- |
+| Example A |
+
+After the table.`
+
+    expect(sanitizeTextForSpeech(text)).toBe('Before the table. After the table.')
+  })
+
+  it('preserves rows outside a table blockquote', () => {
+    const text = `> | Item | Value |
+> | --- | ---: |
+> | Example A | 10 |
+Outside | prose`
+
+    expect(sanitizeTextForSpeech(text)).toBe('Outside | prose')
+  })
+
+  it('preserves malformed tables with mismatched column counts', () => {
+    const text = `Heading | Detail
+--- | --- | ---
+Keep this prose.`
+
+    expect(sanitizeTextForSpeech(text)).toContain('Heading | Detail')
+  })
+
+  it('skips GFM body rows whose cell counts differ from the header', () => {
+    const text = `Before the table.
+
+| Item | Value |
+| --- | ---: |
+| Example A |
+| Example B | 20 | ignored |
+
+After the table.`
+
+    expect(sanitizeTextForSpeech(text)).toBe('Before the table. After the table.')
+  })
+
+  it('skips tables containing escaped pipe characters', () => {
+    const text = `Before the table.
+
+| Item \\| detail | Value |
+| --- | ---: |
+| Example A | 10 |
+
+After the table.`
+
+    expect(sanitizeTextForSpeech(text)).toBe('Before the table. After the table.')
+  })
+
+  it('preserves indented code that resembles a table', () => {
+    const text = `    Item | Value
+    --- | ---
+    Example A | 10`
+
+    expect(sanitizeTextForSpeech(text)).toContain('Item | Value')
+  })
 })

apps/desktop/src/lib/speech-text.ts

@@ -5,6 +5,7 @@ const CODE_BLOCK_SUMMARY = ' code block omitted '
 const INLINE_CODE_RE = /`([^`]+)`/g
 const MARKDOWN_LINK_RE = /\[([^\]]+)\]\(([^)]+)\)/g
 const PARAGRAPH_BREAK_RE = /[ \t]*\n{2,}[ \t]*/g
+const PUNCTUATED_PARAGRAPH_BREAK_RE = /([.!?])([*_~`>"'’”)}\]]*)[ \t]*\n{2,}[ \t]*/g
 const SOFT_BREAK_RE = /[ \t]*\n[ \t]*/g
 
 const THINKING_PREFIX_RE =
@@ -12,16 +13,146 @@ const THINKING_PREFIX_RE =
 
 const URL_RE = /\bhttps?:\/\/\S+/gi
 
+const MARKDOWN_TABLE_DELIMITER_CELL_RE = /^:?-{3,}:?$/
+
+interface MarkdownTableRow {
+  blockquoteDepth: number
+  cells: string[]
+}
+
+function isUnescapedPipe(row: string, index: number): boolean {
+  let backslashes = 0
+
+  for (let cursor = index - 1; cursor >= 0 && row[cursor] === '\\'; cursor -= 1) {
+    backslashes += 1
+  }
+
+  return backslashes % 2 === 0
+}
+
+function splitMarkdownTableCells(row: string): string[] {
+  const cells: string[] = []
+  let cellStart = 0
+
+  for (let index = 0; index < row.length; index += 1) {
+    if (row[index] === '|' && isUnescapedPipe(row, index)) {
+      cells.push(row.slice(cellStart, index).trim())
+      cellStart = index + 1
+    }
+  }
+
+  cells.push(row.slice(cellStart).trim())
+
+  return cells
+}
+
+function parseMarkdownTableRow(line: string): MarkdownTableRow | null {
+  let row = line
+  let blockquoteDepth = 0
+
+  while (true) {
+    const indentation = row.match(/^[ \t]*/)?.[0] ?? ''
+
+    if (indentation.includes('\t') || indentation.length > 3) {
+      return null
+    }
+
+    row = row.slice(indentation.length)
+
+    if (!row.startsWith('>')) {
+      break
+    }
+
+    blockquoteDepth += 1
+    row = row.slice(1)
+
+    if (row.startsWith(' ')) {
+      row = row.slice(1)
+    }
+  }
+
+  row = row.trimEnd()
+
+  const pipeIndexes = [...row.matchAll(/\|/g)].map(match => match.index).filter(index => isUnescapedPipe(row, index))
+
+  if (pipeIndexes.length === 0) {
+    return null
+  }
+
+  const hasLeadingPipe = pipeIndexes[0] === 0
+  const hasTrailingPipe = pipeIndexes.at(-1) === row.length - 1
+
+  if (hasLeadingPipe) {
+    row = row.slice(1)
+  }
+
+  if (hasTrailingPipe) {
+    row = row.slice(0, -1)
+  }
+
+  const cells = splitMarkdownTableCells(row)
+
+  if (cells.length < 2 && !(hasLeadingPipe && hasTrailingPipe && cells.length === 1)) {
+    return null
+  }
+
+  return { blockquoteDepth, cells }
+}
+
+function stripMarkdownTables(text: string): string {
+  const lines = text.replace(/\r\n?/g, '\n').split('\n')
+  const tableLines = new Set<number>()
+
+  let index = 1
+
+  while (index < lines.length) {
+    const delimiterRow = parseMarkdownTableRow(lines[index])
+    const headerRow = parseMarkdownTableRow(lines[index - 1])
+
+    if (
+      !delimiterRow ||
+      !headerRow ||
+      !delimiterRow.cells.every(cell => MARKDOWN_TABLE_DELIMITER_CELL_RE.test(cell)) ||
+      headerRow.cells.length !== delimiterRow.cells.length ||
+      headerRow.blockquoteDepth !== delimiterRow.blockquoteDepth
+    ) {
+      index += 1
+
+      continue
+    }
+
+    tableLines.add(index - 1)
+    tableLines.add(index)
+
+    let rowIndex = index + 1
+
+    for (; rowIndex < lines.length; rowIndex += 1) {
+      const bodyRow = parseMarkdownTableRow(lines[rowIndex])
+
+      if (!bodyRow || bodyRow.blockquoteDepth !== delimiterRow.blockquoteDepth) {
+        break
+      }
+
+      tableLines.add(rowIndex)
+    }
+
+    index = rowIndex
+  }
+
+  return lines.filter((_, index) => !tableLines.has(index)).join('\n')
+}
+
 function normalizeLineBreaks(text: string): string {
   return text
     .replace(/\r\n?/g, '\n')
     .replace(/(\p{L})-\n(\p{L})/gu, '$1$2')
+    .replace(PUNCTUATED_PARAGRAPH_BREAK_RE, '$1$2 ')
     .replace(PARAGRAPH_BREAK_RE, '. ')
     .replace(SOFT_BREAK_RE, ' ')
 }
 
 export function sanitizeTextForSpeech(text: string): string {
-  return normalizeLineBreaks(text)
+  return normalizeLineBreaks(stripMarkdownTables(text))
     .replace(FENCED_CODE_RE, CODE_BLOCK_SUMMARY)
     .replace(THINKING_PREFIX_RE, ' ')
     .replace(MARKDOWN_LINK_RE, '$1')