F2 SLICE-2: Anchorer (fingerprint, resolution ladder, live shift) (#4)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-10 06:50:29 -07:00
parent 9d93ccca19
commit e5817eef0a
2 changed files with 182 additions and 0 deletions
@@ -0,0 +1,113 @@
+/**
+ * Anchorer — hybrid anchoring (spec §6.5). The durable FINGERPRINT is the source
+ * of truth for an anchor's location (INV-3); live OFFSET ranges are a within-
+ * session optimization. The resolution ladder is:
+ *   exact-unique → context-disambiguated → lineHint tiebreak → orphaned.
+ * It NEVER guesses on an unbreakable tie — it orphans instead (INV-1).
+ *
+ * vscode-free: operates on plain strings and character-offset ranges, so it is
+ * unit-testable in Node. The vscode layer converts OffsetRange <-> vscode.Range.
+ */
+import type { Fingerprint } from "./model";
+
+export interface OffsetRange {
+  start: number;
+  end: number;
+}
+
+/** A document edit: the half-open range [start,end) was replaced with text of `newLength`. */
+export interface TextEdit {
+  start: number;
+  end: number;
+  newLength: number;
+}
+
+const MAX_CTX_CHARS = 120;
+const MAX_CTX_LINES = 3;
+
+function lineNumberAt(doc: string, offset: number): number {
+  let line = 0;
+  const bound = Math.min(offset, doc.length);
+  for (let i = 0; i < bound; i++) if (doc.charCodeAt(i) === 10) line++;
+  return line;
+}
+
+function leadingContext(doc: string, start: number): string {
+  let s = doc.slice(Math.max(0, start - MAX_CTX_CHARS), start);
+  const lines = s.split("\n");
+  if (lines.length > MAX_CTX_LINES) s = lines.slice(lines.length - MAX_CTX_LINES).join("\n");
+  return s;
+}
+
+function trailingContext(doc: string, end: number): string {
+  let s = doc.slice(end, Math.min(doc.length, end + MAX_CTX_CHARS));
+  const lines = s.split("\n");
+  if (lines.length > MAX_CTX_LINES) s = lines.slice(0, MAX_CTX_LINES).join("\n");
+  return s;
+}
+
+export function buildFingerprint(docText: string, range: OffsetRange): Fingerprint {
+  return {
+    text: docText.slice(range.start, range.end),
+    before: leadingContext(docText, range.start),
+    after: trailingContext(docText, range.end),
+    lineHint: lineNumberAt(docText, range.start),
+  };
+}
+
+function allIndexesOf(hay: string, needle: string): number[] {
+  if (needle.length === 0) return [];
+  const out: number[] = [];
+  let from = 0;
+  let idx = hay.indexOf(needle, from);
+  while (idx !== -1) {
+    out.push(idx);
+    from = idx + 1;
+    idx = hay.indexOf(needle, from);
+  }
+  return out;
+}
+
+function contextMatches(doc: string, i: number, fp: Fingerprint): boolean {
+  const beforeOk = fp.before.length === 0 || doc.slice(Math.max(0, i - fp.before.length), i) === fp.before;
+  const afterStart = i + fp.text.length;
+  const afterOk = fp.after.length === 0 || doc.slice(afterStart, afterStart + fp.after.length) === fp.after;
+  return beforeOk && afterOk;
+}
+
+const rangeAt = (i: number, text: string): OffsetRange => ({ start: i, end: i + text.length });
+
+/**
+ * Re-resolve a fingerprint against current text. Returns the resolved range or
+ * "orphaned" when no confident, unique match exists (INV-1).
+ */
+export function resolve(docText: string, fp: Fingerprint): OffsetRange | "orphaned" {
+  const occ = allIndexesOf(docText, fp.text);
+  if (occ.length === 0) return "orphaned";
+  if (occ.length === 1) return rangeAt(occ[0], fp.text);
+
+  // Multiple exact matches: try context disambiguation.
+  const ctx = occ.filter((i) => contextMatches(docText, i, fp));
+  if (ctx.length === 1) return rangeAt(ctx[0], fp.text);
+
+  // Still ambiguous: break ties by proximity to lineHint within the best pool.
+  const pool = ctx.length > 0 ? ctx : occ;
+  const dists = pool.map((i) => ({ i, d: Math.abs(lineNumberAt(docText, i) - fp.lineHint) }));
+  const min = Math.min(...dists.map((x) => x.d));
+  const closest = dists.filter((x) => x.d === min);
+  if (closest.length === 1) return rangeAt(closest[0].i, fp.text);
+
+  // Unbreakable tie — refuse to guess.
+  return "orphaned";
+}
+
+/** Maintain a live range across an in-session edit (spec §6.4 `shift`). */
+export function shift(range: OffsetRange, edit: TextEdit): OffsetRange {
+  const delta = edit.newLength - (edit.end - edit.start);
+  const point = (p: number): number => {
+    if (p <= edit.start) return p;
+    if (p >= edit.end) return p + delta;
+    return edit.start; // inside the replaced span — clamp to its start
+  };
+  return { start: point(range.start), end: point(range.end) };
+}
@@ -0,0 +1,69 @@
+import { describe, it, expect } from "vitest";
+import { buildFingerprint, resolve, shift, type OffsetRange } from "../src/anchorer";
+
+const DOC = "line0\nline1 needle here\nline2\nneedle again on line3\n";
+
+describe("buildFingerprint", () => {
+  it("captures the exact text, bounded context, and 0-based lineHint", () => {
+    const start = DOC.indexOf("needle here");
+    const range: OffsetRange = { start, end: start + "needle".length };
+    const fp = buildFingerprint(DOC, range);
+    expect(fp.text).toBe("needle");
+    expect(fp.lineHint).toBe(1);
+    expect(fp.before.endsWith("line1 ")).toBe(true);
+    expect(fp.after.startsWith(" here")).toBe(true);
+    expect(fp.before.length).toBeLessThanOrEqual(120);
+    expect(fp.after.length).toBeLessThanOrEqual(120);
+  });
+});
+
+describe("resolve", () => {
+  it("exact-unique: returns the single occurrence's range", () => {
+    const fp = { text: "line2", before: "", after: "", lineHint: 0 };
+    const start = DOC.indexOf("line2");
+    expect(resolve(DOC, fp)).toEqual({ start, end: start + 5 });
+  });
+
+  it("context-disambiguated: picks the occurrence whose before/after match", () => {
+    const fp = { text: "needle", before: "line1 ", after: " here", lineHint: 99 };
+    const start = DOC.indexOf("needle here");
+    expect(resolve(DOC, fp)).toEqual({ start, end: start + 6 });
+  });
+
+  it("lineHint tiebreak: when context does not disambiguate, the closest line wins", () => {
+    // both occurrences of 'needle', no usable context, hint points at line 3
+    const fp = { text: "needle", before: "", after: "", lineHint: 3 };
+    const start = DOC.indexOf("needle again");
+    expect(resolve(DOC, fp)).toEqual({ start, end: start + 6 });
+  });
+
+  it("orphaned: returns 'orphaned' when the text is gone", () => {
+    const fp = { text: "absent-text", before: "", after: "", lineHint: 0 };
+    expect(resolve(DOC, fp)).toBe("orphaned");
+  });
+
+  it("orphaned: returns 'orphaned' rather than guessing on an unbreakable tie (INV-1)", () => {
+    const doc = "needle\n....\nneedle\n"; // two identical occurrences, equidistant from hint
+    const fp = { text: "needle", before: "", after: "", lineHint: 1 };
+    expect(resolve(doc, fp)).toBe("orphaned");
+  });
+});
+
+describe("shift", () => {
+  it("edit entirely before the range shifts both endpoints by the delta", () => {
+    const r: OffsetRange = { start: 10, end: 15 };
+    // insert 2 chars at offset 0 (replace [0,0) with len 2)
+    expect(shift(r, { start: 0, end: 0, newLength: 2 })).toEqual({ start: 12, end: 17 });
+  });
+
+  it("edit entirely after the range leaves it unchanged", () => {
+    const r: OffsetRange = { start: 2, end: 5 };
+    expect(shift(r, { start: 10, end: 12, newLength: 0 })).toEqual({ start: 2, end: 5 });
+  });
+
+  it("edit overlapping the range clamps the touched endpoints to the edit start", () => {
+    const r: OffsetRange = { start: 5, end: 10 };
+    // replace [3,7) with 1 char (delta = -3)
+    expect(shift(r, { start: 3, end: 7, newLength: 1 })).toEqual({ start: 3, end: 7 });
+  });
+});