Nest 3/4: compile hello-nest ahead of time

[puskuruk]

apps/hello-nest (a decorated controller on NodeAdapter, @nestjs/core 12.1.3) compiles with geatsc. tools/nest/aot resolves @Module, @Controller, @Injectable and the route decorators with the TypeScript checker into a generated module function; scripts/build.mjs runs it as a compiler source-transform plugin, and reflect-metadata and the three @nestjs packages map to typed shims the generated application drives.

Needs the four compiler fixes (compiler release 1.0.38); the floor bump in package.json follows once that release exists.

• 044a5ac Add tools/nest/aot: resolve the decorators with the checker into a generated module function
• 23259ca Map reflect-metadata and @nestjs/common, core and platform-node to typed shims the AOT-generated application drives
• be3c7bd Rewrite a Nest entry through the AOT resolver as a compiler source-transform plugin in build.mjs
• cd5462c Add hello-nest on NodeAdapter with platform-node 12.1.3
• 16bf804 Add scripts/test-hello-nest.mjs comparing the compiled hello-nest against its Node oracle over eleven requests

Verification: node scripts/test-hello-nest.mjs → HELLO_NEST_PARITY: 10 of 11 identical. The one deviation is the JSON.parse error message wording on POST /json with a malformed body (compiler item D7).

Benchmark

Same app under Node and compiled, bench/http-matrix.py (Nest 4/4), 2026-10-19, arm64 Ubuntu 24.04 container on an 8-vCPU Colima VM (Apple M1), VM otherwise idle: 2 s warmup, 8 s samples, wrk -t4 -c64, three interleaved rounds, server on vCPUs 0-3, wrk on 4-7. Not the bench host, so sizes carry the VM's variance (the same binaries moved 25-46% between two runs an hour apart); orders held in every round.

Server	1w /	1w /json	1w PSS MiB	1w startup ms	4w /	4w /json	4w PSS MiB	4w startup ms	4w p50	4w p99
nest-gea	103,000	88,705	9.4	13	210,652	194,149	10.9	15	225 µs	1.15 ms
nest-node	42,828	38,581	119.2	698	77,146	81,775	334.8	1,700	718 µs	7.72 ms
hono-gea	132,018	97,355	6.3	7	254,633	211,371	8.5	11	179 µs	880 µs
hono-node	29,372	25,538	115.7	89	56,799	48,976	260.2	148	840 µs	2.89 ms
gea-raw	156,805	153,935	4.6	14	291,858	278,909	5.9	15	154 µs	990 µs
node-raw	43,572	43,894	86.8	101	103,868	96,322	185.7	219	633 µs	2.69 ms

Four workers, GET /: compiled Nest 2.73× Nest on Node (2.40× at one worker), 2.03× bare node:http on Node, 0.83× compiled Hono; PSS 1/31; startup 13-15 ms against 698 ms and 1,700 ms. Binary 7,619,264 bytes (-Os -flto, stripped). Raw records and the full ten-server table are in Nest 4/4.

Related

Stack: #6 (1/4) → #7 (2/4) → this PR → #9 (4/4, benchmark harness and raw records). Needs the four compiler fixes geastack/compiler#7, geastack/compiler#8, geastack/compiler#9 and geastack/compiler#10 (release 1.0.38); the compiler floor bump follows that release.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}