ci: publish benchmarks in repository Pages

[cpunion]

Summary

• publish benchmark history to the repository pages branch without a separate data repository or dispatch token
• namespace the site output at benchmark/baseline to avoid collisions with other Pages content

This ports the two commits from cpunion#141. The cpunion source PR is merged and its source branch has been deleted; this PR is the xgo-dev counterpart.

Validation

• YAML parsing passes for .github/workflows/benchmark-publish.yml and .github/llgo-benchmark.yml
• git diff --check passes
• only the benchmark publisher workflow and benchmark site path change

[fennoai]

Review summary

This PR switches baseline benchmark publishing to the same-repo GitHub Pages model: it points site-path at benchmark/baseline and drops the external-data-repo wiring (data_repository, data_dispatch_event, secrets.data_token) from benchmark-publish.yml.

The workflow changes themselves are correct and safe. I verified against the reusable workflow xgo-dev/setup-benchmark-go-action/.github/workflows/publish.yml@v1:

• data_repository defaults to "" → falls back to the source repo (xgo-dev/llgo), matching the intent.
• data_token is required: false; same-repo commits use the built-in token, so removing the secret is fine.
• data_dispatch_event defaults to "" (no dispatch), appropriate for same-repo publishing.
• run_id (the only required input) and config_path are retained.
• The new site-path: benchmark/baseline matches the source dir ./benchmark/baseline used in .github/workflows/benchmark.yml.

The findings below are all in benchmark/baseline/README.md, which is not part of this diff and so cannot be commented inline.

Findings

1. Stale documentation — benchmark/baseline/README.md:25-28
The README still says the publisher "commits ... to the pages branch of the configured data repository. Every LLGo repository defaults to <owner>/llgo-benchmark-data." After this PR, publishing defaults to the source repo (xgo-dev/llgo) Pages, not a separate <owner>/llgo-benchmark-data repository. This description is now inaccurate.

2. Stale series paths — benchmark/baseline/README.md:30-34
The documented series paths use the old site-path prefix:

llgo/baseline/series/main/main
llgo/baseline/series/branch/<safe branch identifier>
llgo/baseline/series/pull/<number>

Since site-path is now benchmark/baseline, these should read benchmark/baseline/series/....

3. Stale token statement — benchmark/baseline/README.md:36-37
"pull request jobs never receive the benchmark repository token" describes the removed external-data-repo token mechanism, which no longer exists in this flow. Consider rewording to reflect same-repo publishing.

Non-blocking notes

• Security (low): With data_token removed, same-repo commits now use the default GITHUB_TOKEN, which this workflow grants contents: write, issues: write, pull-requests: write. Consider pinning the reusable workflow to a commit SHA instead of the mutable @v1 tag, and trimming permissions to the minimum the publish flow needs (e.g. drop issues: write if only PR comments are used). Not exploitable on its own; the workflow_run + fork-PR data flow mitigations (trusted config read from default branch, PR-number namespacing, head-SHA re-check) live in the callee and are unchanged by this diff.
• Behavior: Removing data_dispatch_event means no repository_dispatch is sent after publish. This is expected for same-repo publishing, but confirm no downstream deploy relied on that event.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

LLGo baseline benchmarks

126c9be6e54e | workflow run | long-term charts

Program measurements

Platform	Workload	File size	vs main	Build	vs main	Run	vs main
Linux	cprintf	18544 B	new	354.132 ms	new	1.364 ms	new
Linux	fmtprintf	2217600 B	new	3.207 s	new	2.728 ms	new
Linux	println	71504 B	new	352.275 ms	new	1.732 ms	new
macOS	cprintf	84672 B	new	428.033 ms	new	3.903 ms	new
macOS	fmtprintf	2361520 B	new	3.537 s	new	23.181 ms	new
macOS	println	125712 B	new	394.536 ms	new	5.405 ms	new

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs main
Linux	BenchmarkLookupPCRandom	12.330 ns/op	new
Linux	BenchmarkMergeCompilerFlags	143.900 ns/op	new
Linux	BenchmarkMergeLinkerFlags	93.880 ns/op	new
Linux	BenchmarkChannelBuffered	36.270 ns/op	new
Linux	BenchmarkChannelHandoff	23627 ns/op	new
Linux	BenchmarkDefer	42.810 ns/op	new
Linux	BenchmarkDirectCall	1.758 ns/op	new
Linux	BenchmarkGlobalRead	1.758 ns/op	new
Linux	BenchmarkGlobalWrite	2.804 ns/op	new
Linux	BenchmarkGoroutine	30801 ns/op	new
Linux	BenchmarkInterfaceCall	8.793 ns/op	new
Linux	BenchmarkRuntimeGetG	1.759 ns/op	new
macOS	BenchmarkLookupPCRandom	18.150 ns/op	new
macOS	BenchmarkMergeCompilerFlags	164.100 ns/op	new
macOS	BenchmarkMergeLinkerFlags	116.700 ns/op	new
macOS	BenchmarkChannelBuffered	47.980 ns/op	new
macOS	BenchmarkChannelHandoff	12243 ns/op	new
macOS	BenchmarkDefer	73.820 ns/op	new
macOS	BenchmarkDirectCall	1.716 ns/op	new
macOS	BenchmarkGlobalRead	1.808 ns/op	new
macOS	BenchmarkGlobalWrite	2.326 ns/op	new
macOS	BenchmarkGoroutine	28691 ns/op	new
macOS	BenchmarkInterfaceCall	10.230 ns/op	new
macOS	BenchmarkRuntimeGetG	4.682 ns/op	new

No main baseline exists yet; all metrics are marked new.

Warning

• The rendered benchmark data could not be pushed.