You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Both join this repository's full test matrix only, flagged
continue-on-error; the shipped defaults in `repomatic/matrix_axes.py` stay on
GA images, so no downstream repo inherits a preview.
https://github.com/actions/runner-images#available-images
With no `[tool.repomatic.test-matrix.*]` overrides, this is the built-in default: the six runners from the inventory below, the default Python versions, and the rows the transform chain contributes: the `3.15` prerelease flagged `unstable`, the free-threaded `3.14t` build pinned to a single runner as a stable smoke test, and `windows-11-arm` dropped on `3.10`.
32
+
The grid above is the built-in default plus this repository's own probe. The default supplies the six runners from the inventory below, the default Python versions, and the rows the transform chain contributes: the `3.15` prerelease flagged `unstable`, the free-threaded `3.14t` build pinned to a single runner as a stable smoke test, and `windows-11-arm` dropped on `3.10`. The two extra `ubuntu-26.04*` columns come from a temporary `[tool.repomatic.test-matrix.*]` override described in [§ Preview images under evaluation](#preview-images-under-evaluation); a project that sets no overrides gets the six-runner default.
33
33
34
34
The reduced pull-request matrix keeps one runner per OS and two Python versions, for faster feedback:
35
35
@@ -142,6 +142,25 @@ The same spirit covers the matrix's other invariants: its lowest Python should e
142
142
|[`windows-11-arm`](https://github.com/actions/runner-images/blob/main/images/windows/Windows11-Arm64-Readme.md)| Windows | ARM64 | no | Compute ties `windows-2025`; full-matrix only, for native ARM64 execution coverage. |
143
143
|[`windows-2025`](https://github.com/actions/runner-images/blob/main/images/windows/Windows2025-Readme.md)| Windows | x86-64 | yes | Compute tied with `windows-11-arm`; the PR-set Windows pick. |
144
144
145
+
### Preview images under evaluation
146
+
147
+
GitHub ships Ubuntu 26.04 as `ubuntu-26.04` and `ubuntu-26.04-arm`, both still in preview. Neither is a `repomatic` default: the shipped runner sets stay on GA images, so no downstream repo inherits a preview image.
148
+
149
+
This repository probes them on itself through its own `[tool.repomatic]` config, which is the pattern any project can copy:
`variations` keeps both images out of the PR matrix, and `unstable` flags every resulting cell `continue-on-error`, so an image still being provisioned never reddens the default branch. Read the two cells differently. `ubuntu-26.04-arm` against `ubuntu-24.04-arm` is a clean base-version comparison: same architecture, same full image. `ubuntu-26.04` has no matching control, because the matrix's x86 Linux runner is the lean `ubuntu-slim`, and comparing those two would credit the base version for the other image's leanness, the confound [§ Architecture speed is workload-dependent](#architecture-speed-is-workload-dependent) warns about. Treat the x86 cell as a compatibility signal rather than a speed measurement.
161
+
162
+
Accumulate timings with the [§ Measuring your own](#measuring-your-own) recipe. The question to settle before these images reach GA is whether `ubuntu-24.04-arm` should give way to `ubuntu-26.04-arm` as the default single-runner pick, which would also move the PR Linux slot, the `3.14t` smoke test, and the `linux-arm64` Nuitka build host.
163
+
145
164
### Speed tendencies
146
165
147
166
Relative speed is workload-dependent, so the only authoritative numbers are your own. The tendencies below come from `repomatic`'s own full test suite, taken as the median across the five most recent successful runs on all six runners. Two numbers matter and can disagree: **job wall-clock** (the `startedAt`/`completedAt` delta, what you pay in CI minutes) and **compute** (just the test-execution steps, with checkout and environment setup stripped out). When they diverge, a non-compute step is the cause.
0 commit comments