Skip to content

LOCAL_CI_RESULTS.md

Latest commit

 

History

History
172 lines (151 loc) · 10.7 KB

LOCAL_CI_RESULTS.md

File metadata and controls

172 lines (151 loc) · 10.7 KB
Raw

Local CI results — wire-certify.yml run off-platform

Update (2026-09-25): hosted CI is now enabled and green. GitHub Actions was re-enabled at the repository level, and wire-certify.yml executed real steps for the first time that day. On the current tree it is green: all 26 jobs passed (run 36083085187). The CI workflow is 4/4 green (run 36083085188) and Build and Deploy Docs is green too. That first hosted run caught two genuine bugs, both the same root cause — an unpinned mcp dependency resolving to 2.x, whose API renamed FastMCP→MCPServer and dropped Server.list_tools (fixed in bd771db, d6764ec). This is day one of hosted certification, not a long track record. This file, and make ci-local, remain useful and are kept up — but from here on they are a complement to hosted runs, not a substitute for them. The status note immediately below is preserved as the historical record of why local attestation was the sole certification path of record up to that point.

Update (2026-09-25, DCF-Minecraft branch feat/dcf-minecraft, PR #32): hosted CI is green on the branch — Wire Certification run 36101123986 (28 jobs, certify-minecraft included), CI 36101124002, CMake 36101123956, Docs 36101123987. Before that, the new certify-minecraft job and the jobs it touches were run locally with make ci-local (parity guard OK after adding the previously missing certify-exsecutor entry) and the full make certify (ALL CERTS PASS):

Job / check Status How
certify-minecraft ✅ PASS vectors regenerated == both committed copies; 5 Python suites (test_minecraft_{vectors,datapack,sidecar,cli}, test_bedrock_ws, 30 tests, 5/5 repeated runs green); minecraft/build.sh core (JDK 21) + Paper plugin (Nix-store JDK 25) jars; CoreSelfTest loopback
certify-java ✅ PASS + GameCertify, TextCertify, MinecraftCertify
certify-medium ✅ PASS unchanged codecs; mc: now a known scheme (Python-only; C/Rust/Go/Node exit 3, verified per binary)
certify-exsecutor ✅ PASS fasmg from nix; 246/246
Fabric mod ✅ built Gradle 9.7.1 + JDK 25 (Loom 1.18), Yarn 1.21.11+build.6, jar 65 KB
Nix ✅ PASS nix build .#dcf-python .#dcf-minecraft-core .#dcf-minecraft-paper .#dcf-minecraft-datapack; Sphinx docs build (pip venv) links the spec
Real Paper 26.2 round trip ✅ PASS minecraft/tools/devserver_test.py: Oligarchy's dev runner (Paper 26.2-121 + Geyser 2.11.2 + Floodgate 2.2.5, headless) with the plugin in datapack mode: dcf:selftest → barrels → comparators → wires → latch → plugin → UDP (proto) = golden frame; UDP → plugin → dcf:rx_commit → words + barrel 0 verified over the stdin console
Oligarchy .#test-minecraft-server ✅ PASS NixOS VM gate (KVM): jars, datapack, rendered config, sidecar unit on the console FIFO, :19134 listening, every port rule scoped to wg0

Human-run, not repeated here: prism_test.py on a Prism client (it provisions + tails; the user opens the world), Bedrock /connect on a real client.

Status note (2026-09-24): hosted GitHub Actions have never executed a real job for this repository. All 57 wire-certify.yml runs (2026-06-10 → 2026-06-21) failed at startup within seconds (an account billing lock; no step ever ran), and no run was triggered by any later commit. Separately, the workflow file itself was invalid YAML from commit 70beec5 (committed 2026-08-16; an unquoted run: one-liner in the Lua transport step) until this fix, so it could not have run even with billing cleared. Until the owner clears the billing lock, make ci-local (.github/ci-local.sh) and the dated attestations in this file are the certification path of record. workflow_dispatch is now enabled on wire-certify.yml, ci.yml and cmake-multi-platform.yml, so a manual hosted run can be triggered as soon as Actions is unblocked.

Correction (2026-10-03): the note above is superseded. The cause was not a billing lock: GitHub Actions was disabled at the repository level, and it was re-enabled on 2026-09-25 (Documentation/DCF_CODE_REVIEW.md R1). Hosted runs on main at 09866ca (2026-09-26): Wire Certification run 36275105961, CI run 36275105998 and CMake on multiple platforms run 36275106001 all passed. The attestations in this file stay as dated local records.

The Wire Certification workflow (.github/workflows/wire-certify.yml) was executed locally, job-for-job, instead of on GitHub Actions. Toolchains not installed on the host were supplied hermetically with Nix (nix shell nixpkgs#…), exactly the versions CI would provision. This file is a dated attestation, not a substitute for the hosted run.

  • Date: 2026-07-25
  • Commit: 790ebae (Phase 4 certification pass)
  • Host toolchains: Python 3, Rust (cargo), GCC, Go 1.25, GHC 9.10 (host)
  • Nix: Determinate Nix 3.16 (supplied jdk, kotlin, ghc, sbcl, swift)

Result: focused re-run of the certification-tier jobs

This refresh targets the four languages whose tier changed (Haskell/Kotlin/Swift/ Lisp) plus the new/changed jobs (spa, c-sdk-unit, lisp full-vector). The host-native adapter jobs (python/rust/c/go/…) are unchanged from the 394731f run below and were not re-executed here.

Job Status How
certify-lisp ✅ PASS nix sbcl — full 109 encode + 137 syndrome + FEC vectors read from the canonical JSON (upgraded from the embedded subset; still no Quicklisp)
certify-java ✅ PASS nix jdk — Certify + SuperPackCertify + FECCertify (all 246 + FEC)
certify-kotlin ✅ PASS nix kotlin+jdk — Frame+SuperPack+FEC+Certify → run (109 encode + 137 syndrome + FEC)
certify-haskell ✅ PASS nix ghcWithPackages — Certify.hs (109 encode + 137 syndrome + 7 superpack)
certify-swift ⏭️ CI-only the Nix Swift-on-Linux wrapper's swift test did not finish within the local budget; the swift-actions runner is authoritative (unchanged from 394731f)
spa ✅ PASS host cargo (cargo test: token codec, §13 matrix, Python↔Rust cross-language) + python unittest
c-sdk-unit ✅ PASS host cmake — plain + ASan+UBSan + TSan (unit suite, regressions C4/H2/C6, streamdb C5)

Note

The certify-lisp job now certifies the entire golden certificate (not a hand-picked subset) by reading Documentation/{golden,fec}_vectors.json through a small dependency-free JSON reader added to lisp/src/wire.lisp. TSan's first run of the new StreamDB C test found a real data race on the auto-flush running flag, now fixed.

Prior full run (retained for the host-native jobs)

  • Date: 2026-06-18 · Commit: 394731f · Result: 15 PASS · 1 SKIP · 0 FAIL
Job Status
certify-python ✅ PASS
certify-rust ✅ PASS
certify-c ✅ PASS
certify-cpp ✅ PASS
certify-go ✅ PASS
certify-perl ✅ PASS
certify-audio ✅ PASS
certify-game ✅ PASS
certify-text ✅ PASS
certify-node ✅ PASS
certify-lua ✅ PASS
certify-swift ⏭️ SKIP (nix wrapper lacks swift-test)

Local run — 2026-09-25 · b5bcda3 + uncommitted working-tree changes

  • Generated by CI_LOCAL_REPORT=1 make ci-local (.github/ci-local.sh, job-for-job with wire-certify.yml; 26 of 26 jobs run).
  • Host: Linux x86_64; gcc 13.3.0; cargo 1.94.1 (29ea6fb6a 2026-03-24); go version go1.24.7 linux/amd64; node v22.22.2; openjdk version "21.0.10" 2026-01-20; Python 3.11.15; perl v5.38.2; nix: no.
  • Result: 20 PASS · 6 SKIP · 0 FAIL
Job Status Detail
certify-python PASS 12s; partial: wirelab_mcp --selftest not run (python 'mcp' package not installed)
certify-c PASS 4s
c-sdk-unit PASS 33s
certify-hydramodem PASS 180s
certify-rust PASS 0s
spa PASS 6s
certify-wasm SKIP no wasm-pack (CI installs it with the rustwasm installer)
certify-audio PASS 0s
certify-game PASS 1s
certify-text PASS 0s
certify-qkd PASS 0s
qkd-bridge PASS 6s
certify-sstv PASS 0s
certify-snake PASS 2s
certify-lua SKIP no lua5.4 on host and no nix
certify-go PASS 13s
certify-haskell SKIP no ghc/cabal on host and no nix
certify-java PASS 2s
certify-kotlin SKIP no kotlinc on host and no nix
certify-node PASS 0s
certify-perl PASS 1s
certify-cpp PASS 4s
certify-swift SKIP no swift on host and no nix
certify-lisp SKIP no sbcl on host and no nix
certify-medium PASS 7s
io-matrix PASS 18s

Notes on this run (WP-CI-B):

  • First job-for-job run. ci-local.sh now has one function per wire-certify.yml job, in workflow order, and refuses to drift (a job-list mismatch is reported as FAIL ci-local-parity). Earlier local runs covered a subset of the steps.
  • New jobs: certify-medium (regenerate medium_vectors.json + diff both JSON copies and codec/medium_vectors.gen.h; Python/C/Rust/Go/Node/C++ medium certs; 162 cases) and io-matrix (the five punctim CLIs x file/stdio/udp-proto/udp-bare/hydra/garbage: 140 pass, 0 fail, 0 skipped). Java MediumCertify, Perl t/medium.t and hydra_symbols_certify run inside certify-java / certify-perl / certify-hydramodem.
  • Latent workflow bug found and fixed: certify-python ran python3 -m unittest python/tests/test_pipe.py from the repo root, which cannot import dcf.pipe. It would have failed on the first hosted run; it now runs from python/. certify-python also byte-diffs every python/MCP/*_vectors.json against Documentation/ (17 families, including golden/audio/pm_param, which were not checked before).
  • The SKIPs are missing toolchains on this host, which has no nix, lua, sbcl, ghc/cabal, kotlinc, swift or wasm-pack. For the last nix-backed results see the 2026-07-25 section (lisp/java/kotlin/haskell) and 2026-06-18 (lua). certify-wasm has never run locally.
  • Partial steps: numpy was pip-installed on the host for this run, as CI does. The mcp package could not be installed because of a conflict with Debian's PyJWT, so the wirelab_mcp --selftest step did not run. certify-snake skipped building snake_source/snake_mixer because DEMOD_IPC_INCLUDE is not set, which also happens on the CI runner. Durations of 0s come from warm cargo/go caches.
  • The tree had uncommitted changes because other work packages were landing at the same time (medium vectors 156→162, hydramodem comments, CMake option cleanup).