omp-edit/lib/store.js

// Session snapshot store + stale-tag recovery.
//
// Every tagged read/grep/edit records the full LF text under (file, tag). An edit that names
// an older tag is rebased onto the current file when the anchored lines (and a neighbour on
// each side) survived unchanged — the same rule oh-my-pi uses — otherwise it is rejected
// with fresh context so the model can retry.
import { fileTag, splitLines } from './text.js'
import { opAnchors, remapOp } from './hashline.js'

const MAX_SNAPSHOTS_PER_FILE = 12
const MAX_FILES = 400
const MAX_SNAPSHOT_BYTES = 4 * 1024 * 1024

export class SnapshotStore {
  constructor() {
    /** @type {Map<string, Array<{tag:string,text:string,at:number}>>} key -> newest last */
    this.files = new Map()
    /** @type {Map<string,string[]>} named registers persist for the session */
    this.registers = new Map()
  }

  /** Record text for key and return its tag. */
  record(key, text) {
    const tag = fileTag(text)
    if (text.length > MAX_SNAPSHOT_BYTES) return tag
    let list = this.files.get(key)
    if (!list) {
      if (this.files.size >= MAX_FILES) this.files.delete(this.files.keys().next().value)
      list = []
      this.files.set(key, list)
    } else {
      // refresh LRU position
      this.files.delete(key); this.files.set(key, list)
    }
    const i = list.findIndex((s) => s.tag === tag)
    if (i >= 0) list.splice(i, 1)
    list.push({ tag, text, at: Date.now() })
    while (list.length > MAX_SNAPSHOTS_PER_FILE) list.shift()
    return tag
  }

  byTag(key, tag) {
    return this.files.get(key)?.find((s) => s.tag === tag)
  }

  /** Paths (keys) that have a snapshot with this tag — helps explain a tag pasted onto the wrong file. */
  keysWithTag(tag) {
    const out = []
    for (const [k, list] of this.files) if (list.some((s) => s.tag === tag)) out.push(k)
    return out
  }

  forget(key) { this.files.delete(key) }
}

// ---------------------------------------------------------------- line diff (Myers, O(ND))

/** Map old 1-based line -> new 1-based line for lines common to both texts. */
export function lineMap(prevLines, curLines) {
  const map = new Map()
  // Trim common prefix/suffix first (cheap and usually most of the file).
  let a0 = 0
  while (a0 < prevLines.length && a0 < curLines.length && prevLines[a0] === curLines[a0]) { map.set(a0 + 1, a0 + 1); a0++ }
  let ae = prevLines.length; let be = curLines.length
  const tail = []
  while (ae > a0 && be > a0 && prevLines[ae - 1] === curLines[be - 1]) { tail.push([ae, be]); ae--; be-- }
  const A = prevLines.slice(a0, ae); const B = curLines.slice(a0, be)
  for (const [i, j] of lcsPairs(A, B)) map.set(a0 + i + 1, a0 + j + 1)
  for (const [i, j] of tail) map.set(i, j)
  return map
}

function lcsPairs(A, B) {
  const N = A.length; const M = B.length
  if (!N || !M) return []
  if (N * M > 4_000_000) return greedyPairs(A, B)
  const max = N + M
  const V = new Int32Array(2 * max + 2)
  const trace = []
  let found = false
  for (let d = 0; d <= max && !found; d++) {
    trace.push(V.slice())
    for (let k = -d; k <= d; k += 2) {
      let x = k === -d || (k !== d && V[max + k - 1] < V[max + k + 1]) ? V[max + k + 1] : V[max + k - 1] + 1
      let y = x - k
      while (x < N && y < M && A[x] === B[y]) { x++; y++ }
      V[max + k] = x
      if (x >= N && y >= M) { found = true; break }
    }
  }
  // Backtrack.
  const pairs = []
  let x = N; let y = M
  for (let d = trace.length - 1; d >= 0 && (x > 0 || y > 0); d--) {
    const v = trace[d]
    const k = x - y
    const prevK = k === -d || (k !== d && v[max + k - 1] < v[max + k + 1]) ? k + 1 : k - 1
    const prevX = v[max + prevK]
    const prevY = prevX - prevK
    while (x > prevX && y > prevY) { x--; y--; pairs.push([x, y]) }
    if (d > 0) { x = prevX; y = prevY }
  }
  return pairs.reverse()
}

function greedyPairs(A, B) {
  // Fallback for huge diffs: match unique lines in order.
  const pos = new Map()
  B.forEach((l, j) => pos.set(l, pos.has(l) ? -1 : j))
  const pairs = []
  let last = -1
  A.forEach((l, i) => { const j = pos.get(l); if (j !== undefined && j > last) { pairs.push([i, j]); last = j } })
  return pairs
}

// ---------------------------------------------------------------- recovery

/**
 * Try to rebase a section parsed against `prevText` onto `curText`.
 * @returns {{ops:object[], offsets:Set<number>} | null}
 */
export function rebaseOps(prevText, curText, ops) {
  const prev = splitLines(prevText).lines
  const cur = splitLines(curText).lines
  const map = lineMap(prev, cur)
  const anchors = new Set(ops.flatMap(opAnchors))
  const dupPrev = dupes(prev); const dupCur = dupes(cur)
  // Validate: each anchor maps, and for unique lines at least one neighbour moved by the same
  // offset; for duplicated lines every existing neighbour must agree.
  const sorted = [...anchors].sort((a, b) => a - b)
  const groups = []
  for (const a of sorted) {
    const g = groups.at(-1)
    if (g && a === g.end + 1) g.end = a
    else groups.push({ start: a, end: a })
  }
  for (const g of groups) {
    const before = g.start > 1 ? g.start - 1 : null
    const after = g.end < prev.length ? g.end + 1 : null
    for (let line = g.start; line <= g.end; line++) {
      const mapped = map.get(line)
      if (mapped === undefined) return null
      const off = mapped - line
      const okB = before != null && map.get(before) === before + off
      const okA = after != null && map.get(after) === after + off
      const dup = dupPrev.has(prev[line - 1]) || dupCur.has(cur[mapped - 1])
      if (dup) {
        if (before == null && after == null) return null
        if (before != null && !okB) return null
        if (after != null && !okA) return null
      } else if (!okA && !okB && (before != null || after != null)) {
        return null
      }
    }
  }
  const offsets = new Set()
  const remapped = ops.map((op) => remapOp(op, (n) => { const m = map.get(n); offsets.add(m - n); return m }))
  // Ranges must stay contiguous after the remap (nothing inserted inside them).
  for (const op of remapped) {
    if ((op.kind === 'replace' || op.kind === 'cut')) {
      const orig = ops[remapped.indexOf(op)]
      if (op.end - op.start !== orig.end - orig.start) return null
    }
  }
  return { ops: remapped, offsets }
}

function dupes(lines) {
  const seen = new Set(); const d = new Set()
  for (const l of lines) { if (seen.has(l)) d.add(l); else seen.add(l) }
  return d
}