Pushes to main cancel the Workspace run they supersede, as pull requests
already do; the newest push's run tests everything before it. On 16 Nov main
had 144 pushes, each a three-OS run that could not be cancelled, which is much
of what fills the org's macOS runners. Release tags, manual runs and a new
scheduled run every 6 hours keep a group of their own and always finish, and the
scheduled run saves the caches (the gate already saves on scheduled runs of the
default branch). scripts/test-workspace-concurrency.py checks the new policy.
The per-crate workflows, which no longer save caches, cancel superseded runs on
every branch, main included.
The Workspace gate already runs fmt, both clippy runs and the tests of every
crate on Linux, Windows and macOS. The per-crate workflows drop those steps and
keep their own checks: wasm32 builds and tests, no-default and single-feature
clippy and tests, cross-target clippy, docs, release and sample tests,
--ignored tests, ffmpeg oracle tests, the Language-data proofing builds, the PDF
Unicode fixtures, read-only contract and vendored backends. All of them run on
Linux. docpath.yml, print.yml, i18n.yml and ui.yml had no check of their own and
are removed.
They also stop caching dependencies: about 25 rust-cache entries no longer
compete with the gate's for the repository's 10 GB, which the gate's main run
missed on every OS. The data caches (Opus vectors, CRW samples) stay.
Mutable action tags are a supply-chain risk: a moved tag runs different code
without a diff. Every workflow action is now pinned through
.github/workflows/actions.lock, and CI verifies the lockfile so an out-of-sync
or tampered one fails instead of running.
Each third-party action that StepSecurity publishes a secure drop-in
replacement for now uses that replacement, pinned to a commit with its
version as a comment. Versions follow what we used: the newest
StepSecurity release in the same major, or the closest major they
publish. dtolnay/rust-toolchain refs that named a toolchain (stable,
nightly, 1.76) become the replacement's `toolchain` input.
Adds step-security/harden-runner (audit, pinned to v2.22.4 like
craft-common-actions) as the first step of each job that did not have it,
so every workflow in the org's public repos reports its egress to
StepSecurity. Audit, not block: block-mode allow lists come from each
job's observed traffic.
* ci: save the dependency cache from main only, and cancel superseded PR runs
* ci: give runs outside pull requests their own concurrency group
---------
Co-authored-by: FredApps <FredApps@users.noreply.github.com>
* ci: window and ui run macOS only when started by hand
macOS runners are kept for releases. Pushes and pull requests run Linux and Windows; workflow_dispatch runs all three.
* ci: the other workflows run macOS only when started by hand
atomic, docpath, ai-core, i18n, print, pdf, update and keymap follow window and ui: Linux and Windows on pushes and pull requests (keymap: Windows), all their platforms on workflow_dispatch.
Pin third-party actions to full-length commit SHAs (with version comments)
instead of mutable tags/branches to prevent supply-chain tampering via
retagged releases. Resolves SonarCloud rule githubactions:S7637.
* feat(window): shared crate deciding who draws the window frame
Add craft-window, which decides whether the desktop or the app draws the
window frame (titlebar, borders, window buttons) and applies it to an
egui ViewportBuilder. This fixes the drift where some Craft apps drew
their own titlebar on KDE while others kept KWin's.
The rule: macOS, Windows and X11 keep the desktop frame; a Wayland
compositor with zxdg_decoration_manager_v1 (KDE, Sway, Hyprland, COSMIC)
keeps its frame; one without it (GNOME) leaves the frame to the app.
<PREFIX>_TITLEBAR=app|system|desktop overrides. When the app draws the
frame, the decision carries the window-button Layout from GNOME's
button-layout so the app's header bar matches the desktop's order.
Decision logic only; drawing the buttons stays in each app. Extracted and
generalized from pdfcraft's decorations.rs. forbid(unsafe_code); default
build has no egui dependency (egui feature adds decorate()).
* style(window): rustfmt the Wayland probe and Dispatch impl
cargo fmt --check failed CI: collapse the thread-spawn builder chain and
the Dispatch::event signature onto the lines rustfmt wants (verified with
rustfmt 1.98 against the workspace rustfmt.toml).
* fix(window): address review — WAYLAND_SOCKET, bounded gsettings, checkout creds
- Skip the Wayland probe when WAYLAND_SOCKET is set: connect_to_env would
take ownership of the inherited fd and unset the variable, closing the
descriptor winit needs. Keep the desktop frame instead.
- Run the gsettings button-layout lookup behind the same 500 ms bounded
worker as the Wayland probe, so a stalled gsettings can't block startup.
Factored the worker into a shared bounded() helper.
- window.yml: set persist-credentials: false on checkout so the token
isn't left in .git/config while cargo runs dependency build scripts.
- README: document the socket guard and the bounded gsettings lookup.
Verified with rustfmt 1.98 and clippy -D warnings (both feature sets).
* fix(window): reap the gsettings child on timeout; document probe tradeoffs
Follow-up to review: the button-layout lookup now spawns gsettings with
piped output, reads it on a worker, and kills+reaps the child whether it
finishes or times out — so a stalled gsettings leaves neither a blocked
process nor a parked worker (it is a one-shot startup call).
Also document, on the Wayland probe, that an inherited WAYLAND_SOCKET
(rare; normal launches get WAYLAND_DISPLAY) deliberately falls back to the
desktop frame to protect winit's connection — the same frame an app that
doesn't probe would get, a safe default rather than a regression.
Verified: rustfmt 1.98 clean, clippy -D warnings clean (default + egui),
tests pass.