mirror of
https://github.com/backnotprop/plannotator.git
synced 2026-09-14 14:17:26 +08:00
Michael Ramos
afac44b940
Replaces the all-files renderer with Pierre's virtualized CodeView (one scroller, identical look, scales to huge diffs); deletes the legacy AllFilesDiffView/LazyFileDiff and their flag; single-file panels stay on FileDiff. @pierre/diffs pinned exact 1.2.8 (1.2.9's tree is broken on npm — see #880). Also ships: diff staleness detection ('Diff out of date · Refresh' toolbar notice backed by shared per-VCS fingerprints + GET /api/diff/fresh on both Bun and Pi servers), a stale-content guard so files edited mid-review degrade to raw-patch view instead of breaking virtualization, the Ask AI wrong-file fix (pre-existing), and toolbar/header UI polish (global settings cog, aligned sem badges, no all-files split dragger). Hardened through five review waves; tested with 1384 passing tests incl. real-git fingerprint and patch-consistency suites. Known minor gaps documented in the PR.
53 lines
2.2 KiB
TypeScript
53 lines
2.2 KiB
TypeScript
/**
|
|
* Guard against augmenting a diff with STALE file contents.
|
|
*
|
|
* The review session captures its patch once (at launch / diff switch), but
|
|
* `/api/file-content` reads the file as it is NOW. The two can drift — e.g.
|
|
* an agent edits or commits a file while the user is mid-review. Feeding
|
|
* `processFile` a patch plus contents that no longer correspond produces a
|
|
* FileDiffMetadata whose hunk/context line math is internally inconsistent;
|
|
* Pierre's virtualization then fails layout estimation for the file
|
|
* ("trailing context mismatch" console errors, disappearing content while
|
|
* scrolling). Rendering the raw patch (no expand-context for that file) is
|
|
* the correct degradation.
|
|
*
|
|
* The check: the patch's hunk headers fully determine the expected line-count
|
|
* delta between the old and new file. If the fetched contents' totals don't
|
|
* reconcile with that delta, the contents are stale relative to the patch.
|
|
* (Same-length edits slip through, but those cannot break the count math —
|
|
* worst case is a few wrong context lines until the next diff refresh.)
|
|
*/
|
|
|
|
const HUNK_HEADER_RE = /^@@ -\d+(?:,(\d+))? \+\d+(?:,(\d+))? @@/gm;
|
|
|
|
function countLines(content: string): number {
|
|
if (content.length === 0) return 0;
|
|
const segments = content.split('\n').length;
|
|
// A trailing newline does not start an extra line (git semantics).
|
|
return content.endsWith('\n') ? segments - 1 : segments;
|
|
}
|
|
|
|
export function isContentConsistentWithPatch(
|
|
patch: string,
|
|
oldContent: string | null,
|
|
newContent: string | null,
|
|
): boolean {
|
|
// Only the both-sides case is verifiable from totals alone. One-sided
|
|
// fetches (added / deleted files) pass through — processFile receives a
|
|
// single file there and cannot produce a two-sided context mismatch.
|
|
if (oldContent == null || newContent == null) return true;
|
|
|
|
let net = 0;
|
|
let sawHunk = false;
|
|
for (const match of patch.matchAll(HUNK_HEADER_RE)) {
|
|
sawHunk = true;
|
|
const oldCount = match[1] != null ? Number(match[1]) : 1;
|
|
const newCount = match[2] != null ? Number(match[2]) : 1;
|
|
net += newCount - oldCount;
|
|
}
|
|
// No hunks → nothing to reconcile against (binary / metadata-only patch).
|
|
if (!sawHunk) return true;
|
|
|
|
return countLines(newContent) - countLines(oldContent) === net;
|
|
}
|