refactor: hoist LoaderFunc to package-level var in phpheaders

[dunglas]

Summary

Hoists the otter.LoaderFunc closure in GetUnCommonHeader to a package-level loader var, so it is allocated once at init time instead of being re-created on every call.

This is a minor cleanup — the previous code created a new LoaderFunc closure each time GetUnCommonHeader was called. While otter's cache-hit path is fast enough that this doesn't show a measurable difference in end-to-end benchmarks, avoiding the repeated allocation is strictly better.

What changed

Before (closure created per call):

func GetUnCommonHeader(ctx context.Context, key string) string {
    phpHeaderKey, err := headerKeyCache.Get(
        ctx,
        key,
        otter.LoaderFunc[string, string](func(_ context.Context, key string) (string, error) {
            return "HTTP_" + headerNameReplacer.Replace(strings.ToUpper(key)) + "\x00", nil
        }),
    )
    ...
}

After (closure allocated once):

var loader = otter.LoaderFunc[string, string](func(_ context.Context, key string) (string, error) {
    return "HTTP_" + headerNameReplacer.Replace(strings.ToUpper(key)) + "\x00", nil
})

func GetUnCommonHeader(ctx context.Context, key string) string {
    phpHeaderKey, err := headerKeyCache.Get(ctx, key, loader)
    ...
}

Benchmarks

Apple M1 Pro, 8 runs, benchstat comparison — no regressions, no extra allocations:

Benchmark	main	PR	vs base
HelloWorld	41.81µ ± 2%	42.75µ ± 5%	~ (p=0.065)
ServerSuperGlobal	73.36µ ± 2%	74.20µ ± 3%	~ (p=0.105)
UncommonHeaders	69.03µ ± 3%	68.71µ ± 1%	~ (p=0.382)

All results within noise. Zero change in allocations.

[dunglas]

I implemented @maxm86545 suggestion, this is now ready to be merged:

Benchmarks

Apple M1 Pro, 6 runs, benchstat comparison (main vs this PR):

Latency (sec/op)

Benchmark	main	PR	vs base
HelloWorld	55.07µ ± 107%	44.08µ ± 5%	~ (p=0.065)
ServerSuperGlobal	85.78µ ± 64%	76.37µ ± 6%	-10.98% (p=0.004)
UncommonHeaders	75.59µ ± 13%	70.53µ ± 1%	-6.70% (p=0.002)

Memory (B/op)

Benchmark	main	PR	vs base
HelloWorld	1.404Ki ± 1%	1.384Ki ± 1%	-1.43% (p=0.026)
ServerSuperGlobal	18.15Ki ± 34%	16.15Ki ± 8%	-11.02% (p=0.004)
UncommonHeaders	15.96Ki ± 14%	15.37Ki ± 1%	~ (p=0.065)

Allocations (allocs/op)

Benchmark	main	PR	vs base
HelloWorld	17	17	~
ServerSuperGlobal	17	17	~
UncommonHeaders	18	21	+16.67% (p=0.002)

Summary

ServerSuperGlobal is the most realistic benchmark (simulates a typical browser request with common headers + env vars). It shows an 11% improvement in both latency and memory, with strong
statistical significance (p=0.004).

The optimization works by:

1. Early return when all request headers are common (the typical case) — skipping the uncommon header path entirely
2. Batching uncommon header lookups into a single GetUnCommonHeaders call using Get+LoaderFunc (per review feedback — avoids cache-miss storms vs BulkGet)

The +3 allocs in UncommonHeaders come from the uncommonKeys slice and uncommonHeaders map needed to batch the lookups. This only applies when uncommon headers are present.

[henderkes]

Based on just the benchmarks with zero further insight on the topic, I think it's worth it. Runtime performance is almost always worth it over a slightly lower memory footprint and as a php server, it's not like FrankenPHP is expected to consume little memory.

[Copilot]

Pull request overview

This PR aims to improve request handling performance by batching uncommon HTTP header name conversions/caching, and updates tests accordingly.

Changes:

• Replace per-header uncommon header conversion with a batched GetUnCommonHeaders helper backed by the existing Otter cache.
• Update CGI header-to-$_SERVER registration to precompute uncommon header names in bulk.
• Adjust header correctness tests to use the new batched helper (and add a duplicate copy in phpmainthread_test.go).

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
internal/phpheaders/phpheaders.go	Replaces single-key conversion with a batched API that returns a map of converted header keys.
cgi.go	Refactors header registration to count/register common headers first, then batch-convert/register uncommon headers.
internal/phpheaders/phpheaders_test.go	Updates the common-header correctness test to use the new batched API.
phpmainthread_test.go	Adds a second copy of the common-header correctness test in the main package tests.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated no new comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[dunglas]

Re-running the benchmarks properly, it's not worth it. I just made a tiny cleanup.