[Next.js/dylib] Preserve NextRequest and nextUrl.searchParams across App Route imports

What happened?

The Perch Next.js lower-bound fixture on Perry 0.5.1503 could construct a NextRequest in one module and call an imported route function, but could not rely on the callee receiving a usable request argument. NextRequest.nextUrl.searchParams was also not reliable through that path, so the benchmark had to ignore the request and hard-code its workload.

This remains a distinct end-to-end gap even though important lower-level fixes have landed: #7252 roots cross-module call arguments and #6738 backs URLSearchParams subclasses. The missing regression is the actual production Next request shape across the generated App Route/import boundary.

Reproduction app (provided)

Use the exact pinned app and verifier attached to #8034. Do not create a new app. Its route is deliberately re-exported across a module boundary:

AppRouteRouteModule.handle
  -> app/api/benchmark/route.ts
  -> lib/route-impl.ts::GET/POST

Build with:

bash make-next-app-route-fixture.sh
cd perry-next-app-route-fixture
npm install --package-lock-only
npm ci
npm run build

The focused request is:

curl -i \
  -H 'x-request-id: request-7' \
  'http://127.0.0.1:3100/api/benchmark?id=request-7&iterations=7'

The returned JSON must contain:

{
  "method": "GET",
  "pathname": "/api/benchmark",
  "id": "request-7",
  "iterations": 7,
  "beforeAwait": "request-7",
  "afterAwait": "request-7",
  "requestBody": ""
}

The attached verifier also sends a POST and requires requestBody === "perry-request-body".

Required implementation behavior

Preserve the NextRequest object passed by the generated App Route module into the re-exported user handler.
Preserve method, URL, nextUrl, ReadonlyURLSearchParams, headers, and body state across the imported call.
Keep the request live across allocations, dynamic import, promise continuation, timer continuation, and GC movement.
Do not special-case the fixture or lower NextRequest to a partial plain object.
Keep direct calls, re-exports, sync handlers, and async handlers on the same correct path.

Regression coverage

Add a small compiler fixture for a request-shaped class passed through a re-exported imported function, plus the production integration test from #8034. Run the small fixture with forced evacuation / GC verification so a stale-but-lucky pointer cannot pass.

Acceptance criteria

All request-field assertions in [Next.js/dylib] Add a pinned production App Route parity fixture and CI gate #8034's verify.mjs pass through the production routeModule.handle path.
GET query data and POST body data are correct on the first cold call and subsequent warm calls.
The 20 concurrent requests never exchange IDs or headers.
The test passes with forced/verified GC and with normal GC.
No direct call to the userland GET, hard-coded query value, response fabrication, or other compatibility fallback exists in the Perry test path.

Baseline / references

Observed baseline: Perry 0.5.1503 at 564c56308d221a51b50308d9165578fbb176e877
Fixture and full commands: [Next.js/dylib] Add a pinned production App Route parity fixture and CI gate #8034
Related groundwork: fix(url): route URLSearchParams subclass instances to their native backing #6738, fix(gc): root every argument of a cross-module direct call (#7240, by @jdalton) #7252

fix(codegen): optnone post-RS4GC relocation-bloated functions

feat(gc): sharpen the from-space scan and stack-map walk instruments

fix(codegen): optnone post-RS4GC relocation-bloated functions

feat(gc): sharpen the from-space scan and stack-map walk instruments

fix(codegen): unbreak in-process RS4GC on inline asm and relocation-g…

feat(gc): name the HOLDERS of a stale from-space address, not just th…

feat(gc): sharpen the from-space scan and stack-map walk instruments

fix(gc): root heap values held across JS-running calls in runtime and…

feat(gc): name the HOLDERS of a stale from-space address, not just th…

fix(next): pass production App Route dylib gate (#8082)

Closing: everything this issue asked for has landed on main except the forced-evacuation arm, which was split out as #8163 and stays open there.

PR #8082's body says "Closes #8034 and #8036", but GitHub only links the first reference in that phrasing (each issue needs its own keyword), so #8034 auto-closed on merge and this one did not. Closing manually per that intent.

What landed, and where it is gated

Small compiler fixture with forced/verified evacuation — fix(gc): preserve request graphs across route imports #8044 (fe0d49792): test-files/test_gap_gc_next_request_import.ts + test-files/fixtures/gc_next_request_import/ (request-shaped class through a re-exported imported handler; allocation, dynamic import, promise + timer continuation, body read; 20 concurrent handlers started before the first await; explicit gc() while all 20 are suspended). Registered in test-parity/gc_repsel_corpus.txt, so scripts/gc_repsel_matrix.sh runs it under the force_verify / verify_evac / cons_scan_off arms in gc-moving-witnesses — green on main (latest scheduled run 31932625278, 2026-08-16 06:56 UTC, 3be2016c1).
Production integration test from [Next.js/dylib] Add a pinned production App Route parity fixture and CI gate #8034 — test(next): land the pinned production App Route fixture and its dylib gate #8161 (e75191b93): tests/fixtures/next-app-route/ with the routeModule.handle bypass guard armed, tests/test_next_app_route_dylib.sh grepping the host log for the guard (its signal cannot reach verify.mjs's exit code), wired as next-app-route.yml (workflow_dispatch + nightly 17 3 * * * UTC; merged after today's slot, so its first scheduled run is 2026-08-17).
Runtime/codegen fixes and the release-tier fixture — fix(next): pass production App Route dylib gate #8082 (53e8a21e3): tests/release/packages/next-app-route/fixture.sh (Node production oracle, separate runtime/stdlib provider images, untouched webpack output compiled as an app-only dylib, 10 cold processes × 2 verifier passes) — normal arm PASS on that head; plus the stacked defects it took to get there: Scope split unknown-function fallback wrappers by module #8078, fix(codegen): unbreak in-process RS4GC on inline asm and relocation-grown functions #8128, fix(gc): root heap values held across JS-running calls in runtime and ext-http #8131, fix(next): make computed relative chunk requires resolve in a compiled App Route #8146, fix(hir): a nested class's own name beats an enclosing binding in new #8153, fix(gc): prune FUNCTION_CLASS_IDS keys whose closure died (#8040) #8168.

What is NOT closed here — acceptance bullet 4, "passes with forced/verified GC": the production route under PERRY_GC_FORCE_EVACUATE=1 … PERRY_GC_SCHEDULE_SEED=8036 still dies with TypeError: value is not a function from a holder outside the GC heap. That is #8163, opt-in in the fixture behind PERRY_NEXT_ROUTE_FORCED_GC=1 (off, fails loudly when set — not a SKIP, not continue-on-error).

One reconciliation note for whoever touches the fixtures next: the release-tier copy's perry-host.js does not carry the bypass guard that tests/fixtures/next-app-route/perry-host.js has; it relies on the forbidden-diagnostic grep instead. Both copies should converge on the armed host.

turnloop P3: JS timers on a per-agent heap, Node's event-loop phase o…

Navigation Menu

Repository navigation

[Next.js/dylib] Preserve NextRequest and nextUrl.searchParams across App Route imports #8036

What happened?

Reproduction app (provided)

Required implementation behavior

Regression coverage

Acceptance criteria

Baseline / references

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[Next.js/dylib] Preserve NextRequest and nextUrl.searchParams across App Route imports #8036

Description

What happened?

Reproduction app (provided)

Required implementation behavior

Regression coverage

Acceptance criteria

Baseline / references

Activity

proggeramlug commented on Aug 16, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions