perf(request): cache method key

[yusukebe]

Caching the method-key to improving performance.

[yusukebe]

Hey @usualoma

I realized that the current v2 branch is not fast on Ping and Query benchmarks. The following is the result of comparing the v2 and the current npm:

============================================================
BENCHMARK RESULTS
============================================================

| Benchmark         | npm            | dev            | Difference  |
| ----------------- | -------------- | -------------- | ----------- |
| Average           | 76,276.42      | 83,202.77      | +9.08%      |
| Ping (GET /)      | 97,463.59      | 97,195.56      | -0.28%      |
| Query (GET /id)   | 94,839.04      | 94,642.80      | -0.21%      |
| Body (POST /json) | 36,526.62      | 57,769.94      | +58.16%     |

I think the changes like #310 decrease the performance. If so, it's a waste to use the changes.

But there are still some points for performance improvement.

The first one is this PR. Plus, another one is #320

npm vs PR #319 (this PR)

============================================================
BENCHMARK RESULTS
============================================================

| Benchmark         | npm            | dev            | Difference  |
| ----------------- | -------------- | -------------- | ----------- |
| Average           | 76,151.65      | 84,026.50      | +10.34%     |
| Ping (GET /)      | 97,012.20      | 98,769.61      | +1.81%      |
| Query (GET /id)   | 94,844.98      | 95,605.70      | +0.80%      |
| Body (POST /json) | 36,597.76      | 57,704.20      | +57.67%     |

npm vs PR #320

============================================================
BENCHMARK RESULTS
============================================================

| Benchmark         | npm            | dev            | Difference  |
| ----------------- | -------------- | -------------- | ----------- |
| Average           | 76,659.70      | 88,721.47      | +15.73%     |
| Ping (GET /)      | 97,443.60      | 101,423.13     | +4.08%      |
| Query (GET /id)   | 95,915.76      | 102,000.37     | +6.34%      |
| Body (POST /json) | 36,619.74      | 62,740.92      | +71.33%     |

Combining both PR will make it faster. I think it's enough as an "appeal point" for the major release.

What do you think of it? If it makes sense for you, please review the PRs.

[usualoma]

Hi @yusukebe!

Thanks for the follow-up!
You're right, my code wasn't great.

Given this approach, since method() gets called at least once in the request, I think it's better to generate it upfront rather than using ||=.

diff --git i/src/request.ts w/src/request.ts
index 67c29f7..f88b98e 100644
--- i/src/request.ts
+++ w/src/request.ts
@@ -367,7 +367,7 @@ const readBodyDirect = (request: Record<string | symbol, any>): Promise<Buffer>
 
 const requestPrototype: Record<string | symbol, any> = {
   get method() {
-    return (this[methodKey] ||= normalizeIncomingMethod(this[incomingKey].method))
+    return this[methodKey]
   },
 
   get url() {
@@ -536,6 +536,7 @@ export const newRequest = (
 ) => {
   const req = Object.create(requestPrototype)
   req[incomingKey] = incoming
+  req[methodKey] = normalizeIncomingMethod(incoming.method)
 
   const incomingUrl = incoming.url || ''
 

[yusukebe]

@usualoma Thanks! Let's go with this.