test(app-router): cover consumed prefetch cache handoff by NathanDrake2406 · Pull Request #2758 · cloudflare/vinext

NathanDrake2406 · 2026-07-30T08:24:45Z

Summary

add a deterministic App Router browser regression for the consumed-prefetch to committed-cache handoff
delay the redundant cache tee after a full prefetch is consumed and assert that remounting the destination Link does not issue a duplicate RSC request
retain the existing upstream Next.js five-minute parallel-route reuse coverage

Why

The original production fix on this branch was superseded by the broader cache-publication work merged in #2760. After merging current main, the conflict was resolved in favor of that newer implementation, leaving this PR as focused regression coverage for the same race.

The test remains useful because it deterministically delays the cache branch that previously created the false-miss window. Current main passes by publishing the already-buffered consumed-prefetch snapshot without waiting for that redundant branch to drain.

Upstream parity reference:

test/e2e/app-dir/app-client-cache/client-cache.parallel-routes.test.ts
https://github.com/vercel/next.js/blob/canary/test/e2e/app-dir/app-client-cache/client-cache.parallel-routes.test.ts

Closes #2756.

Validation

vp check tests/e2e/app-router/nextjs-compat/client-cache.spec.ts
targeted browser regression: 1/1 passed
targeted browser regression repeated 10 times in parallel: 10/10 passed

Keep the buffered snapshot discoverable while navigation publishes the committed visited response. This prevents a remounted prefetch-enabled Link from observing a false cache miss and starting a duplicate RSC request.

NathanDrake2406 · 2026-07-30T08:25:10Z

@codex review

pkg-pr-new · 2026-07-30T08:25:52Z

Open in StackBlitz

npm i https://pkg.pr.new/@vinext/cloudflare@2758

npm i https://pkg.pr.new/create-vinext-app@2758

npm i https://pkg.pr.new/@vinext/types@2758

npm i https://pkg.pr.new/vinext@2758

commit: f7c1526

chatgpt-codex-connector · 2026-07-30T08:29:00Z

Codex Review: Didn't find any major issues. Can't wait for the next one!

Reviewed commit: e5e80479aa

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

github-actions · 2026-07-30T08:32:14Z

Performance benchmarks

Compared f7c1526 against base dc05459 using alternating same-runner rounds. Next.js was unchanged and skipped.

0 improved · 0 regressed · 6 within ±1.5%

Scenario	Framework	Baseline	Current	Change
Client bundle size (gzip)	vinext	134.5 KB	134.5 KB	⚫ -0.0%
Client entry size (gzip)	vinext	122.1 KB	122.1 KB	⚫ 0.0%
Dev server cold start	vinext	3.06 s	3.03 s	⚫ -1.2%
Production build time	vinext	3.28 s	3.31 s	⚫ +0.9%
RSC entry closure size (gzip)	vinext	112.3 KB	112.3 KB	⚫ +0.0%
Server bundle size (gzip)	vinext	189.7 KB	189.7 KB	⚫ +0.0%

View detailed results and traces

_{🟢 improvement · 🔴 regression · ⚫ change below 1.5% · paired base/head}

# Conflicts: # packages/vinext/src/server/app-browser-entry.ts

james-elicx · 2026-07-31T12:16:54Z

/bigbonk review for issues

ask-bonk

Reviewed the test-only change (+59 lines). This is well-constructed, deterministic regression coverage for the consumed-prefetch → committed-cache handoff race.

Verification done:

Confirmed the final diff is test-only; the source fix (e5e8047) was superseded by #2760, as the PR body states.
Confirmed the mechanism under test still exists in current main: consumedPrefetchSnapshot at app-browser-entry.ts:1974/2007/2285 is exactly what lets navigation publish the buffered snapshot without awaiting the delayed cacheBufferPromise (line 2207/2286). The test deliberately stalls that redundant cache tee via releaseOldNavigationTail, so a regression that reintroduced the await cacheBufferPromise dependency would hang the poll at line 335 or surface a duplicate request at line 358 — i.e. the test would fail loudly. Good.
The self-restoring global ReadableStream.prototype.tee override (line 308-329) is safe here: no browser-side code runs between install (page.evaluate at 298) and targetLink.click() (333), so the first tee() after install is guaranteed to be the navigation's navBody.tee() at app-browser-entry.ts:2200. This matches the existing prototype-patch pattern already used at lines 502-514.
vp check passes (format, lint, types).

Only minor, non-blocking observations below. These mirror patterns already present in this file, so they are consistency notes rather than defects. No changes required.

ask-bonk · 2026-07-31T12:19:46Z

+        const cacheReader = cacheBranch.getReader();
+        const delayedCacheBranch = new ReadableStream({
+          async start(controller) {
+            while (true) {


Non-blocking: the while (true) drain loop has no error propagation — if cacheReader.read() rejects, start() rejects, releaseOldNavigationTail is never set, and the poll at line 335 times out rather than failing with a clear message. This is acceptable for a test (it fails loudly either way) and is identical to the existing pattern at lines 462-478 / 598-614, so I'd keep it consistent. Just flagging that a rejected read surfaces as a timeout, not a targeted assertion failure.

ask-bonk · 2026-07-31T12:19:46Z

+        "tee",
+      ) as typeof ReadableStream.prototype.tee;
+      ReadableStream.prototype.tee = function (this: ReadableStream<unknown>) {
+        ReadableStream.prototype.tee = originalTee;


Non-blocking: the override self-restores on the first tee() call, so exactly one tee() gets delayed. That's the intended contract here (only the navigation's cache tee at app-browser-entry.ts:2200 should be stalled), and the timing guarantees it fires first. Worth a one-line comment noting the "delay exactly the first post-install tee" intent, since the self-restore is load-bearing and easy to accidentally remove in a future refactor.

ask-bonk · 2026-07-31T12:19:55Z

Review posted successfully to PR #2758.

Summary

I reviewed this test-only PR (+59 lines to client-cache.spec.ts) and posted a COMMENT review with 2 non-blocking observations.

Verdict: LGTM. This is well-constructed, deterministic regression coverage. Key findings:

Confirmed test-only diff — the source fix (e5e8047) was superseded by fix(app-router): preserve page result render ordering #2760, as the PR body states. Final diff touches only the spec.
The test is meaningful against current main — consumedPrefetchSnapshot still exists in app-browser-entry.ts (lines 1974/2007/2285). The test stalls the redundant cache tee via releaseOldNavigationTail, so a regression that reintroduced an await cacheBufferPromise dependency would hang the poll (line 335) or surface a duplicate RSC request (line 358) — it fails loudly.
The global self-restoring ReadableStream.prototype.tee override is safe — no browser-side code runs between install and the navigation click, so the first tee() is guaranteed to be the navigation's. This matches the existing prototype-patch pattern already in the file (lines 502-514).
vp check passes (format, lint, types).

Non-blocking notes posted:

The while (true) drain loop surfaces a rejected read as a poll timeout rather than a targeted assertion — acceptable and consistent with existing patterns in the file.
Suggested a one-line comment documenting that the self-restore is load-bearing ("delay exactly the first post-install tee"), since it's easy to accidentally remove in a refactor.

No changes required.

github run

fix(app-router): preserve consumed prefetch during cache handoff

e5e8047

Keep the buffered snapshot discoverable while navigation publishes the committed visited response. This prevents a remounted prefetch-enabled Link from observing a false cache miss and starting a duplicate RSC request.

NathanDrake2406 marked this pull request as ready for review July 30, 2026 08:32

Merge remote-tracking branch 'origin/main' into codex/pr2758-owner

0a5deb5

# Conflicts: # packages/vinext/src/server/app-browser-entry.ts

james-elicx changed the title ~~fix(app-router): preserve consumed prefetch during cache handoff~~ test(app-router): cover consumed prefetch cache handoff Jul 31, 2026

james-elicx added 3 commits July 31, 2026 12:18

test(app-router): verify prefetch handoff delay

3716e6a

test(app-router): latch prefetch handoff delay

f3f205b

test(app-router): hold cache latch through remount

f7c1526

ask-bonk Bot reviewed Jul 31, 2026

View reviewed changes

james-elicx merged commit 87220d0 into cloudflare:main Aug 3, 2026
57 checks passed

NathanDrake2406 mentioned this pull request Aug 15, 2026

test(app-router): cover cross-path hash push URL replacement #2926

Closed

Navigation Menu

Uh oh!

test(app-router): cover consumed prefetch cache handoff - #2758

test(app-router): cover consumed prefetch cache handoff#2758
james-elicx merged 5 commits into
cloudflare:mainfrom
NathanDrake2406:nathan/fix-client-cache-prefetch-flake

NathanDrake2406 commented Jul 30, 2026 •

edited by james-elicx

Loading

Uh oh!

NathanDrake2406 commented Jul 30, 2026

Uh oh!

pkg-pr-new Bot commented Jul 30, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot commented Jul 30, 2026

Uh oh!

github-actions Bot commented Jul 30, 2026 •

edited

Loading

Uh oh!

james-elicx commented Jul 31, 2026

Uh oh!

ask-bonk Bot left a comment

Uh oh!

ask-bonk Bot Jul 31, 2026

Uh oh!

ask-bonk Bot Jul 31, 2026

Uh oh!

ask-bonk Bot commented Jul 31, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

NathanDrake2406 commented Jul 30, 2026 • edited by james-elicx Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Why

Validation

Uh oh!

NathanDrake2406 commented Jul 30, 2026

Uh oh!

pkg-pr-new Bot commented Jul 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot commented Jul 30, 2026

Uh oh!

github-actions Bot commented Jul 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Performance benchmarks

Uh oh!

james-elicx commented Jul 31, 2026

Uh oh!

ask-bonk Bot left a comment

Choose a reason for hiding this comment

Uh oh!

ask-bonk Bot Jul 31, 2026

Choose a reason for hiding this comment

Uh oh!

ask-bonk Bot Jul 31, 2026

Choose a reason for hiding this comment

Uh oh!

ask-bonk Bot commented Jul 31, 2026

Summary

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

NathanDrake2406 commented Jul 30, 2026 •

edited by james-elicx

Loading

pkg-pr-new Bot commented Jul 30, 2026 •

edited

Loading

github-actions Bot commented Jul 30, 2026 •

edited

Loading