Hmbown/CodeWhale · error

Complete PR review plan found

Error message

Complete PR review plan found {} file patches; expected {}

What it means

After splitting the diff into per-file patches, plan_pr_review verifies the count matches the PR's changed_files count from the GitHub view and that it is non-empty. A mismatch means the diff and PR metadata disagree or the diff parser missed files.

Solutions

1. Re-fetch both the PR view and diff so they come from the same revision
2. Check the diff is complete/untruncated (compare byte counts, no `gh` output truncation)
3. Add a parser test for the diff shape (renames, binary files, new files) that miscounts patches
4. Retry — a transient gh failure can produce a partial diff

Example fix

// before
let diff = fetch_cached_diff(pr)?; // may be stale
// after
let (view, diff) = fetch_view_and_diff_together(pr)?; // same revision

// sanity check before planning
if diff.is_empty() || view.changed_files == 0 { bail!("empty diff or metadata"); }

match plan_pr_review(...) { Err(e) if e.to_string().contains("expected") => refetch_and_retry(pr).await, ... }

When it happens

Trigger: The `gh pr diff` output is truncated or partial (network error, huge PR); diff format changed so pr_file_patches misparses file boundaries; diff was cached from an older revision than view.changed_files; changed_files metadata reports 0 or a stale count.

Common situations: Very large PRs where gh truncates output; reviewing a PR whose head moved between metadata fetch and diff fetch; parser regression after a git diff format change (e.g. rename with mode change).

Related errors

• PR review plan did not partition the complete diff…
• cannot parse exactly one Rust TRUSTED_KEYS table
• Cargo metadata is not valid JSON
• Codewhale stream-json line
• Codewhale stream-json line

AI-assisted analysis of Hmbown/CodeWhale@73e0f67d83 (2026-09-22). Data as JSON: /api/errors/b33b64a4ae05f707. Report an issue: GitHub.

Appendix: source

Thrown at crates/tui/src/tools/review.rs:437

/// skipped in diff order and named in `manifest.skipped_files` (AC4). Only a
/// plan that covers nothing still errors.
///
/// Known limitations, beside the behaviour: skips are whole files — a file
/// with one oversized hunk is skipped entirely, never truncated — and files
/// stay in diff order rather than re-sorted by estimated risk.
pub(crate) fn plan_pr_review(
    diff: &str,
    view: &super::review_pr::GhPullRequest,
    max_chars: usize,
    max_passes: usize,
) -> anyhow::Result<PrReviewPlan> {
    anyhow::ensure!(max_chars > 0, "Review max_chars must be positive");
    anyhow::ensure!(
        (1..=MAX_REVIEW_PASSES).contains(&max_passes),
        "Review max_passes must be from 1 to {MAX_REVIEW_PASSES}"
    );
    let patches = pr_file_patches(diff);
    anyhow::ensure!(
        patches.len() == view.changed_files && !patches.is_empty(),
        "Complete PR review plan found {} file patches; expected {}",
        patches.len(),
        view.changed_files
    );

    // A whole file stays together whenever it fits. An oversized text file
    // splits only at complete hunk boundaries, with the full file header
    // replayed into every part so each part stays a self-describing patch;
    // no line is elided, shortened or reordered. Sizes use the model
    // representation, so a binary payload already omitted there can never
    // drive a split.
    let mut atoms: Vec<PrReviewAtom<'_>> = Vec::new();
    for patch in patches {
        let patch_chars = super::review_pr::model_diff(patch).chars().count();
        if patch_chars <= max_chars {
            atoms.push(PrReviewAtom::Piece(PrReviewPiece {
                diff: Cow::Borrowed(patch),

View on GitHub (pinned to 73e0f67d83)