floatdrop / zoxy / 28731866820

Builds Branch Commit Type Ran Committer Via Coverage
28731866820 main refactor(test): extract TestHarness for the common proxy setup The ~30-line setup block (IO loop, TestOrigin, the simple single-cluster config + router, connection pool, and a started ProxyServer on a loopback listener) was repeated in most proxy... push floatdrop github
96.09
28731649069 refactor/proxy-test-helpers Merge 5f5838891 into 5cb74b179 Pull #2 web-flow github
96.16
28731096073 refactor/proxy-test-helpers Merge 4f7bcb15b into 5cb74b179 Pull #2 web-flow github
96.27
28730770547 main docs: remove file structure and stale diagram from readme push web-flow github
96.28
28730688866 main build: add `zig build lint` for the 100-column limit (replaces CI awk) Line-length enforcement lived as an inline awk in ci.yml, invisible to local dev. Reimplement it as a Zig tool (scripts/check_line_length.zig) wired into `zig build lint`, per... push floatdrop github
96.25
28730611566 cleanup/mechanical-quick-wins Merge e4e8995f3 into b3a70575b Pull #1 web-flow github
96.28
28730354196 cleanup/mechanical-quick-wins Merge 13fbd13d7 into b3a70575b Pull #1 web-flow github
96.25
28730035603 cleanup/mechanical-quick-wins Merge 6e71345c8 into b3a70575b Pull #1 web-flow github
96.32
28717819858 main docs: refresh Phase 5 — h2load bench measured, drop stale zrk The h2 benchmark landed (bench/h2.sh), so Phase 5's "bench pending / zrk can't drive it" claims are stale. Updates, implemented slices left intact: - Trim the dead sentence off the Si... push floatdrop github
96.32
28717688477 main bench: HTTP/2 benchmark (bench/h2.sh) — h2-over-TLS vs native nginx-h2 zoxy speaks h2 only over TLS+ALPN and translates to h1 upstream, so the script drives h2load over https at a multiplexing sweep (-m), each level run direct (nginx's own TLS+ht... push floatdrop github
96.29
28717130404 main bench: pin cores + read h2load -m1 as latency, not throughput The h2load switch surfaced a scary 626k(direct) vs 188k(proxied) req/s. It was a measurement artifact, not a zoxy regression: - -m1 is closed-loop (1 req/conn), so req/s = conns/mean-... push floatdrop github
96.32
28716593661 main sim: h2c client fault menu (full parity, 2/2) Rounds out the h2c simulator to match the H1 fault menu. The multiplexed h2c client now draws a per-stream request kind — GET, small POST, a ~10 KiB POST, or a deliberately-malformed request (missing ... push floatdrop github
96.29
28716396711 main bench: swap zrk for h2load as the load generator h2load (nghttp2) speaks HTTP/2, which zrk cannot — needed ahead of the Phase 5 H2 downstream work. Add nghttp2 to the dev shell so h2load is on PATH; bench/run.sh uses it from PATH or fetches via `... push floatdrop github
96.56
28712220753 main docs: plan Phase 5 — own sans-io H2 core, H2-downstream/H1-upstream first Close the §6 open question: roll our own HTTP/2 rather than FFI nghttp2 (its allocator hook routes allocations but does not bound them; our H2 limits must be reserved const... push floatdrop github
96.6
28712001619 main docs: refresh stale H2/H3 note — pure-Zig stacks exist but don't fit The "nothing usable in pure Zig" premise is stale as of 2026 (quic-zig, httpx.zig). The conclusion is unchanged and now sharper: they bundle their own runtime + injected allocat... push floatdrop github
96.56