Dodaios is an open-source agent OS for your desktop: it runs your coding agents, browser and apps by voice or chat, on your own machine. Formerly Personal Jarvis.
Python agent-orchestrationai-agentai-coding-assistantclaude-codecoding-agentcomputer-usedesktop-applocal-firstmcppersonal-jarvisvoice-assistantvoice-coding
perf(ci): classify lanes on pushes to main, keep the full test suite there
rubenluetke10-beep765e96029eef
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.ymlindex fd26ee3dc0..4e6a1cd7b9 100644--- a/.github/workflows/ci.yml+++ b/.github/workflows/ci.yml@@ -14,8 +14,9 @@# * Every PR ran every job (~20 runners, three 45-min macOS jobs on most),# and pushes to main cancelled each other, so main was never proven.# -> `detect` classifies the diff into lanes (scripts/ci/classify_changes.py,-# fail-open); only affected lanes run on a PR. Pushes to main, the nightly-# run and manual dispatches run everything and are never cancelled.+# fail-open); only affected lanes run. Every push to main runs the whole+# test suite and is never cancelled; the nightly run and manual+# dispatches run every lane.# * Branch protection listed five job names; renaming a job meant a settings# change. -> `CI gate` aggregates every job (scripts/ci/required_results.py)# and is the ONLY required check, for humans and for the merge train.@@ -91,6 +92,7 @@ jobs:pr_mode: ${{ steps.mode.outputs.pr_mode }}linux_shards: ${{ steps.shards.outputs.linux_shards }}windows_shards: ${{ steps.shards.outputs.windows_shards }}+ tests_full: ${{ steps.shards.outputs.tests_full }}os_matrix: ${{ steps.mode.outputs.os_matrix }}steps:- name: Checkout (full history for the diff base)@@ -108,12 +110,16 @@ jobs:IN_MACOS: ${{ inputs.include_macos }}PR_BASE: ${{ github.event.pull_request.base.sha }}run: |- full=false; macos=false; pr_mode=false+ # A push to main classifies its lanes like a pull request (the+ # concurrent-job limit is shared by every agent's PR), but always+ # runs the WHOLE test suite on Linux and Windows: post-merge proof+ # is never narrowed. Nightly and manual full runs force everything.+ full=false; macos=false; pr_mode=false; tests_full=falsecase "$EVENT" in- push) full=true ;;- schedule) full=true; macos=true ;;+ push) tests_full=true ;;+ schedule) full=true; macos=true; tests_full=true ;;workflow_dispatch)- [ "$IN_FULL" = "true" ] && full=true+ [ "$IN_FULL" = "true" ] && full=true && tests_full=true[ "$IN_MACOS" = "true" ] && macos=true ;;pull_request) pr_mode=true ;;esac@@ -130,7 +136,8 @@ jobs:fiif [ "$macos" = "true" ]; then os='["ubuntu-latest","windows-latest","macos-latest"]'; else os='["ubuntu-latest","windows-latest"]'; fi{- echo "full_mode=$full"; echo "macos=$macos"; echo "pr_mode=$pr_mode"; echo "base=$base"+ echo "full_mode=$full"; echo "tests_full=$tests_full"; echo "macos=$macos"+ echo "pr_mode=$pr_mode"; echo "base=$base"echo "os_matrix=$os"} | tee -a "$GITHUB_OUTPUT"@@ -147,18 +154,21 @@ jobs:git diff --name-only "$BASE" HEAD | tee changed.txt | python3 scripts/ci/classify_changes.pyfi- # Linux always runs the whole suite in six shards. Windows runs four- # shards only when every lane is on; otherwise one runner takes the- # tests the diff can reach.+ # Linux always runs the whole suite in six shards. Windows runs the+ # whole suite in four shards on main, nightly, full manual runs and+ # pipeline changes; on a pull request one runner takes the tests the+ # diff can reach. A push always runs the python lane's test jobs.- name: Size the test matricesid: shardsshell: bashrun: |echo 'linux_shards=[1,2,3,4,5,6]' >> "$GITHUB_OUTPUT"- if [ "${{ steps.lanes.outputs.full }}" = "true" ]; then+ if [ "${{ steps.mode.outputs.tests_full }}" = "true" ] || [ "${{ steps.lanes.outputs.full }}" = "true" ]; thenecho 'windows_shards=[1,2,3,4]' >> "$GITHUB_OUTPUT"+ echo 'tests_full=true' >> "$GITHUB_OUTPUT"elseecho 'windows_shards=[1]' >> "$GITHUB_OUTPUT"+ echo 'tests_full=false' >> "$GITHUB_OUTPUT"figates:@@ -313,7 +323,7 @@ jobs:- name: Pick the tests (full run, or what this diff can reach)id: pickenv:- FULL: ${{ needs.detect.outputs.full }}+ FULL: ${{ needs.detect.outputs.tests_full }}BASE: ${{ needs.detect.outputs.base }}SHARDS: ${{ needs.detect.outputs.windows_shards }}run: |diff --git a/docs/ci-pipeline.md b/docs/ci-pipeline.mdindex 6e2f8301b4..d32716b38b 100644--- a/docs/ci-pipeline.md+++ b/docs/ci-pipeline.md@@ -18,7 +18,7 @@ agent branch ──► pull request ──► CI (lanes) ──► CI gate ─| Stage | What it does | Script || --- | --- | --- |-| `detect` | Classifies the diff into lanes. Fails open: an empty diff, a pipeline change, a push to main, the nightly run and a manual run turn every lane on. | `scripts/ci/classify_changes.py` |+| `detect` | Classifies the diff into lanes. Fails open: an empty diff, a pipeline change, the nightly run and a manual run turn every lane on. A push to main classifies its lanes too (the concurrent-job limit is shared), but always runs the whole test suite on Linux and Windows. | `scripts/ci/classify_changes.py` || `static gates` | ~20 repository gates (keys, bundle, mirrors, privacy, docs, CLI coverage, ratchets, bash 3.2, no new German) in one job. Every gate reports. | `scripts/ci/run_gates.py` || `python contracts (fast)` | Import cleanliness on the bare install, the named contract guards, skill-routing precision and recall, plugin auth. Blocking, no baseline. | — || `tests linux 1..6` | The whole suite in six shards. Batches of files run in fresh processes with a wall-clock budget; a failed batch is re-run file by file, a failed file once more (a pass there is reported as flaky). | `scripts/ci/run_tests_parallel.py` |