[MaxtuneLee]

Summary

revalidatePath() strips the trailing slash before constructing its _N_T_ implicit cache tag. Pages served at trailing-slash URLs (e.g. /en/) were building their own tags with the slash, so the tags never matched and revalidation silently did nothing.

Fix: normalize the pathname (strip trailing /) in both tag-building sites before generating the _N_T_ tag, matching what revalidatePath() produces.

Files touched

File	Change
packages/vinext/src/server/app-page-cache.ts	Strip trailing slash from stem before building _N_T_${stem} tag in buildAppPageCacheTags()
packages/vinext/src/server/implicit-tags.ts	Same normalization in buildPageCacheTags() — normalizedPathname used for the implicit tag
tests/fixtures/app-trailing-slash-isr/app/[lang]/legacy/page.tsx	Extend fixture page to render ISO timestamp + <span id="lang"> + RevalidateButton
tests/fixtures/app-trailing-slash-isr/app/[lang]/revalidate-button.tsx	New client component — triggers /api/revalidate and surfaces result
tests/fixtures/app-trailing-slash-isr/app/api/revalidate/route.ts	New API route — calls revalidatePath() with the path supplied by the button
tests/e2e/app-router/trailingslash-revalidate.browser.spec.ts	New Playwright spec — ports the revalidation test from Next.js (see Reference)
tests/app-trailing-slash-isr.test.ts	Add timestamp content check

Testing plan

• Pass the new Playwright spec: PLAYWRIGHT_PROJECT=app-router-chrome-browser-specific pnpm playwright test trailingslash-revalidate
  • Both withSlash=true and withSlash=false variants pass
  • Verified the timestamp changes after clicking the revalidate button
• Pass the existing vitest integration test: pnpm vitest run tests/app-trailing-slash-isr.test.ts
  • Checks /en/legacy/ returns 200 with timestamp content
  • Checks /en redirects to /en/ and renders correct lang param

Reference

https://github.com/vercel/next.js/blob/canary/test/e2e/app-dir/trailingslash/trailingslash.test.ts
Fix #1832

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/@vinext/cloudflare@2390

npm i https://pkg.pr.new/vinext@2390

commit: 399f246

[github-actions]

Performance benchmarks

Compared 399f246 against base da086c0 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	122.0 KB	122.0 KB	⚫ -0.0%
Client entry size (gzip)	vinext	117.0 KB	117.0 KB	⚫ -0.0%
Dev server cold start	vinext	2.53 s	2.55 s	⚫ +0.7%
Production build time	vinext	3.02 s	2.99 s	⚫ -0.9%
RSC entry closure size (gzip)	vinext	94.3 KB	94.3 KB	⚫ 0.0%
Server bundle size (gzip)	vinext	160.0 KB	160.1 KB	⚫ +0.0%

View detailed results and traces

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

[james-elicx]

/bigbonk review for issues

[ask-bonk]

Review: normalize trailing slash in implicit cache tags

The fix is correct and well-scoped. I traced the producer/consumer sides:

• Producer (shims/cache.ts:135-136): revalidatePath() strips the trailing slash (stem = path.endsWith("/") ? path.slice(0,-1) : path) and builds _N_T_${stem || "/"}.
• Consumers (the two builders this PR touches): now strip the trailing slash with pathname.length > 1 && pathname.endsWith("/") before building _N_T_${stem}.

These line up for all cases I checked:

• /en/legacy/ -> _N_T_/en/legacy (matches revalidatePath("/en/legacy") and revalidatePath("/en/legacy/")).
• Root / -> the guard length > 1 keeps it /, producing _N_T_/, matching revalidatePath where stem || "/" yields _N_T_/. Good — the guard is the key detail that prevents the root case from regressing to _N_T_.

The type: "page"/type: "layout" tags were already correct since they derive from segment splitting (empty trailing segment is skipped), so limiting the change to the no-type tag is right.

Dev/prod parity check (per AGENTS.md): I verified the Pages Router path (server/pages-page-response.ts:643) already normalizes via stem || "/", so no parallel gap is left behind. Both App Router builders (app-page-cache.ts and implicit-tags.ts) are used in distinct code paths, so updating both is correct.

Test coverage: The e2e spec correctly exercises the regression — before this fix the render-time tag was _N_T_/en/legacy/ which matched neither withSlash variant. The self-build/serve pattern matches the existing hash-rsc-requests.browser.spec.ts.

No blocking issues. A couple of minor nits below on the test fixtures.

[ask-bonk]

Review posted successfully. Summary of my findings on PR #2390:

Verdict: Correct, well-scoped fix — no blocking issues

The fix is sound. I traced both sides of the cache-tag matching:

• Producer (shims/cache.ts:135-136): revalidatePath() strips trailing slash → _N_T_${stem || "/"}
• Consumers (the two App Router builders this PR touches): now strip trailing slash before building _N_T_${stem}

These line up for all cases including the tricky root / case — the pathname.length > 1 guard correctly prevents / from collapsing to _N_T_ (it stays _N_T_/, matching the producer's stem || "/").

Key verifications:

• The type: "page"/type: "layout" tags were already correct (they derive from segment splitting which skips empty trailing segments), so limiting the change to the no-type tag is right.
• Dev/prod parity (per AGENTS.md): The Pages Router path (pages-page-response.ts:643) already normalizes via stem || "/", so no parallel gap is left behind. Both modified App Router builders are used in distinct, active code paths — updating both is necessary.
• The e2e test genuinely exercises the regression: before the fix, the render-time tag was _N_T_/en/legacy/ which matched neither withSlash variant.
• The self-build/serve test pattern matches the existing hash-rsc-requests.browser.spec.ts.

Two minor non-blocking nits posted inline: redundant nested startTransition in the fixture button, and a note that the e2e test compares against initialGeneratedAt rather than refreshedGeneratedAt (equivalent given the prior toPass block).

github run