Fix benchmark dashboard artifact downloads

[abap34]

Summary

• fix benchmark dashboard artifact downloads by handling GitHub artifact redirects explicitly
• avoid sending the GitHub Authorization header to the signed object-storage URL
• allow the dashboard builder to run on local Python 3.9 via postponed annotations

Cause

The benchmark Pages builder found dashboard-entry artifacts, but Python urllib followed the artifact download redirect while keeping the GitHub Authorization header. The redirected signed storage URL rejected that request with 401, so every CI run was skipped and dev/bench/dashboard.json was generated with entries: [].

Validation

• PYTHONPYCACHEPREFIX=/tmp/almo-pycache python3 -m py_compile scripts/build_benchmark_dashboard.py
• python3 scripts/build_benchmark_dashboard.py --repo abap34/almo --out-dir /tmp/almo-dashboard-fixture-mainfix --fixture-json scripts/benchmark_pages/fixture_dashboard.json
• verified real GitHub artifacts can be fetched: latest 3 CI runs produced entries=3 with 21 benchmarks each

generated by @codex

[chatgpt-codex-connector]

To use Codex here, create an environment for this repo.

[github-actions]

CI Summary

Check	Status	Details
Tests	Success	16 passed / 0 failed
Benchmarks	Success	21 benchmark(s) captured
Coverage	Success	93.1% (1280 of 1375 lines)

Benchmark Means

Benchmark	Mean (ms)	Min (ms)	Max (ms)	Iterations
parse_small_document	9.008	8.956	9.056	5
parse_large_document	191.628	190.712	192.295	5
parse_realistic_article_document	22.843	22.778	22.942	5
parse_release_notes_document	7.402	7.358	7.519	5
parse_inline_heavy_document	347.236	346.621	347.767	5
parse_table_heavy_document	160.844	157.465	171.114	5
parse_footnote_heavy_document	56.845	56.599	57.042	5
parse_structured_document	49.583	49.163	49.996	5
parse_judge_document	0.029	0.028	0.031	5
render_small_document	18.258	18.144	18.335	5
render_large_document	228.749	216.069	243.606	5
render_realistic_article_document	32.065	31.532	32.478	5
render_release_notes_document	16.445	16.247	17.033	5
render_structured_document	67.239	66.806	67.584	5
render_table_heavy_document	216.472	215.805	218.353	5
render_footnote_heavy_document	71.605	71.493	71.714	5
render_interactive_document	12.870	12.814	12.959	5
render_judge_document	13.164	13.110	13.231	5
summary_inline_heavy_document	389.244	369.314	416.853	5
summary_realistic_article_document	32.029	31.066	33.964	5
summary_large_document	66.963	66.679	67.527	5

Benchmark Diff vs main

Benchmark	Base Mean (ms)	PR Mean (ms)	Delta (ms)	Change
parse_small_document	9.167	9.008	-0.159	✅ 1.7% speed up
parse_large_document	198.466	191.628	-6.838	✅ 3.4% speed up
parse_realistic_article_document	26.598	22.843	-3.755	✅ 14.1% speed up
parse_release_notes_document	8.646	7.402	-1.244	✅ 14.4% speed up
parse_inline_heavy_document	377.304	347.236	-30.068	✅ 8.0% speed up
parse_table_heavy_document	156.870	160.844	+3.974	⚠️ 2.5% slow down
parse_footnote_heavy_document	56.910	56.845	-0.065	➖ 0.1% change
parse_structured_document	49.852	49.583	-0.269	➖ 0.5% change
parse_judge_document	0.031	0.029	-0.002	✅ 6.5% speed up
render_small_document	18.638	18.258	-0.380	✅ 2.0% speed up
render_large_document	218.799	228.749	+9.950	⚠️ 4.5% slow down
render_realistic_article_document	32.964	32.065	-0.899	✅ 2.7% speed up
render_release_notes_document	17.182	16.445	-0.737	✅ 4.3% speed up
render_structured_document	67.879	67.239	-0.640	➖ 0.9% change
render_table_heavy_document	216.951	216.472	-0.479	➖ 0.2% change
render_footnote_heavy_document	72.933	71.605	-1.328	✅ 1.8% speed up
render_interactive_document	13.203	12.870	-0.333	✅ 2.5% speed up
render_judge_document	13.535	13.164	-0.371	✅ 2.7% speed up
summary_inline_heavy_document	389.409	389.244	-0.165	➖ 0.0% change
summary_realistic_article_document	32.658	32.029	-0.629	✅ 1.9% speed up
summary_large_document	67.901	66.963	-0.938	✅ 1.4% speed up

Benchmark dashboard: https://abap34.github.io/almo/dev/bench/?pr=178
PR preview: https://abap34.github.io/almo/dev/previews/pr-178/