omp-edit/lib/block.js

// Syntactic block resolution for `N*` anchors.
//
// OMP resolves blocks with tree-sitter. This is a dependency-free approximation that covers
// the same shapes for the common languages:
//   - Markdown: a heading runs through deeper headings until the next same/higher heading.
//   - Indentation languages (Python, YAML, Nim, CoffeeScript, …) and any opener ending in ':'
//     with no brackets: the opener plus every following line indented deeper (blank lines
//     inside are kept, trailing blank lines are not).
//   - Bracket languages (JS/TS/C/Go/Rust/Java/JSON/CSS/…): from the opener to the line that
//     closes the first bracket left open, string- and comment-aware.
// Leading decorators/attributes (`@x`, `#[x]`, `[Attr]`) are grouped with the declaration
// that follows them.
import path from 'node:path'
import { EditError } from './text.js'

const INDENT_EXT = new Set(['.py', '.pyi', '.yaml', '.yml', '.nim', '.coffee', '.sass', '.styl', '.pug', '.haml', '.slim', '.gd'])
const MARKDOWN_EXT = new Set(['.md', '.markdown', '.mdx', '.rst'])
const HASH_COMMENT_EXT = new Set(['.py', '.pyi', '.rb', '.sh', '.bash', '.zsh', '.yaml', '.yml', '.toml', '.pl', '.r', '.nim', '.coffee', '.cmake', '.mk', '.dockerfile', '.ex', '.exs', '.jl', '.tf', '.hcl'])

const indentOf = (line) => (line.match(/^[ \t]*/)[0].replace(/\t/g, '    ').length)
const isBlank = (line) => line.trim() === ''
const isDecorator = (line) => /^\s*(@[\w.]|#\[|\[[A-Z]\w*(\(|\]))/.test(line)

/**
 * Resolve the block that starts at 1-based `line`.
 * @returns {{start:number,end:number}} 1-based inclusive span
 */
export function resolveBlock(filePath, lines, line) {
  if (line < 1 || line > lines.length) throw new EditError(`Block anchor ${line} is outside the file (1-${lines.length}).`)
  const ext = path.extname(filePath).toLowerCase()
  const base = path.basename(filePath).toLowerCase()
  const opener = lines[line - 1]
  if (isBlank(opener)) throw new EditError(`Block anchor ${line} is a blank line; anchor the construct opener (the first line of the function/class/heading).`)

  if (MARKDOWN_EXT.has(ext)) return markdownBlock(lines, line)

  // Decorators/attributes group with the declaration they precede.
  let declLine = line
  while (declLine <= lines.length && isDecorator(lines[declLine - 1]) && !/[{(:]\s*$/.test(lines[declLine - 1].replace(/\)\s*$/, ''))) {
    if (declLine === lines.length) break
    declLine++
  }

  const decl = lines[declLine - 1]
  const useIndent = INDENT_EXT.has(ext) || (/:\s*(#.*)?$/.test(decl) && !/[{[(]/.test(stripStrings(decl)))
  const span = useIndent
    ? indentBlock(lines, declLine)
    : bracketBlock(lines, declLine, HASH_COMMENT_EXT.has(ext) || base === 'dockerfile' || base === 'makefile')
  if (!span || span.end <= line) {
    throw new EditError(
      `Line ${line} does not open a multi-line block (${JSON.stringify(opener.trim().slice(0, 60))}). ` +
      'Anchor the construct opener, never a closer, blank line or inner statement; for a single line use an explicit range (PUT N.=N:).',
    )
  }
  return { start: line, end: span.end }
}

function markdownBlock(lines, line) {
  const m = /^(#{1,6})\s/.exec(lines[line - 1])
  if (!m) {
    // Paragraph / list block: until the next blank line.
    let end = line
    while (end < lines.length && !isBlank(lines[end])) end++
    if (end === line) throw new EditError(`Line ${line} is a single-line Markdown block; use an explicit range.`)
    return { start: line, end }
  }
  const level = m[1].length
  let end = lines.length
  let fence = false
  for (let i = line; i < lines.length; i++) {
    if (/^\s*(```|~~~)/.test(lines[i])) fence = !fence
    if (fence) continue
    const h = /^(#{1,6})\s/.exec(lines[i])
    if (h && h[1].length <= level) { end = i; break }
  }
  while (end > line && isBlank(lines[end - 1])) end--
  return { start: line, end }
}

function indentBlock(lines, line) {
  const base = indentOf(lines[line - 1])
  let end = line
  // A multi-line opener (e.g. a def whose parameters wrap) continues until the ':' line.
  let i = line
  if (!/:\s*(#.*)?$/.test(lines[line - 1])) {
    let depth = bracketDelta(lines[line - 1])
    while (depth > 0 && i < lines.length) { depth += bracketDelta(lines[i]); i++ }
    end = i
  }
  for (let k = i; k < lines.length; k++) {
    const l = lines[k]
    if (isBlank(l)) continue
    if (indentOf(l) > base) end = k + 1
    else break
  }
  return { start: line, end }
}

function stripStrings(s) {
  return s.replace(/(["'`])(?:\\.|(?!\1).)*\1/g, '""')
}

function bracketDelta(s) {
  let d = 0
  for (const c of stripStrings(s)) {
    if (c === '(' || c === '[' || c === '{') d++
    else if (c === ')' || c === ']' || c === '}') d--
  }
  return d
}

/**
 * Scan from the opener with a small lexer (strings, template literals, line and block
 * comments) and end at the line where depth returns to zero after first opening.
 */
function bracketBlock(lines, line, hashComments) {
  let depth = 0
  let opened = false
  let inBlock = false
  let quote = null
  for (let i = line - 1; i < lines.length; i++) {
    const s = lines[i]
    for (let j = 0; j < s.length; j++) {
      const c = s[j]
      const n = s[j + 1]
      if (inBlock) { if (c === '*' && n === '/') { inBlock = false; j++ } continue }
      if (quote) {
        if (c === '\\') { j++; continue }
        if (c === quote) quote = null
        continue
      }
      if (c === '/' && n === '/') break
      if (hashComments && c === '#') break
      if (c === '/' && n === '*') { inBlock = true; j++; continue }
      if (c === '"' || c === "'" || c === '`') { quote = c; continue }
      if (c === '{' || c === '(' || c === '[') { depth++; opened = true }
      else if (c === '}' || c === ')' || c === ']') {
        depth--
        if (opened && depth <= 0) {
          // Keep scanning the rest of this line for a chained opener: `} else {`, `}).then(() => {`.
          const rest = stripStrings(s.slice(j + 1))
          if (bracketDelta(rest) > 0) { depth += 0; continue }
          return { start: line, end: i + 1 }
        }
      }
    }
    if (quote && quote !== '`') quote = null // unterminated single-line string
    if (!opened && i > line - 1 + 3) return null // opener must open within a few lines
  }
  return null
}