Function getPrDiffHunks

scripts/github-actions/post-review-comments.js:2041–2127

({ github, owner, repo, prNumber, commitSha, log, cache })

// `cache` (optional) memoizes the result for the run: with several batches each
// failing 422, the inventory would otherwise be refetched per batch.
async function getPrDiffHunks({ github, owner, repo, prNumber, commitSha, log, cache }) {
  const logFn = typeof log === "function" ? log : () => {};
  if (cache && cache.diff !== undefined) return cache.diff;

  const files = new Map();
  const known = new Set();
  let complete = true;

  const PER_PAGE = 100;
  const MAX_PAGES = 30;
  let page = 1;
  while (page <= MAX_PAGES) {
    const res = await readWithPacing(
      `listFiles (page ${page})`,
      () => github.rest.pulls.listFiles({ owner, repo, pull_number: prNumber, per_page: PER_PAGE, page }),
      logFn
    );
    const batch = (res && res.data) || [];
    for (const file of batch) {
      if (!file || !file.filename) continue;
      known.add(file.filename);
      if (file.patch) {
        const parsed = parseDiffHunkInventory(file.patch);
        if (parsed.complete) {
          files.set(file.filename, parsed.ranges);
        } else {
          logFn(
            `[422-fallback] Patch data for ${file.filename} is incomplete or malformed; ` +
              `comments on that file will be treated as unknown rather than out-of-diff.`
          );
        }
      }
    }
    if (batch.length < PER_PAGE) break;
    page++;
  }
  if (page > MAX_PAGES) {
    complete = false;
    logFn(
      `[422-fallback] PR changed-file list exceeded ${MAX_PAGES * PER_PAGE} files; ` +
        `diff inventory is incomplete, so no comment will be discarded as out-of-diff.`
    );
  }

  // An empty inventory proves nothing. A PR that produced review comments
  // necessarily has changed files, so an empty listFiles response is an anomaly
  // (diff not yet materialized server-side, or a malformed/empty response body)
  // rather than evidence that every commented path sits outside the diff.
  // Trusting it would classify EVERY comment "invalid" and discard the whole
  // batch without a single posting attempt — the exact outcome the tri-state
  // classification exists to prevent. Note this is the mirror of the truncation
  // case above: too many files and zero files are both "cannot judge".
  if (known.size === 0) {
    complete = false;
    logFn(
      `[422-fallback] PR changed-file list came back empty, which cannot be right for a PR ` +
        `under review; treating the diff inventory as incomplete, so no comment will be ` +
        `discarded as out-of-diff.`