syntaqxThe merge queue builds, tests and lands each pull request by its own merge method. Each state was built in a sandbox with git merge commits, then landed by fast-forward as it was, so a queued pull request always landed as a merge commit whatever method was asked; and since every sandbox built its own copy of the entries ahead, an entry's state did not descend from the one that landed before it, and a batch fell back to testing again. Now the work service builds every state in the repository itself, with repos' new build_queue_step: the entry landed by its method (merge commit, squash, rebase or fast-forward) on the state of the entry ahead of it, or on the default branch for the first, with the commits land would make there (merge_methods.rs make_on, shared with land_by; queue_step_on decides a step), pushed to the attempt's g1t-queue/ branch. Landing is then a fast-forward to exactly the tested commit, recorded as the entry's method with that commit as its merge commit; settle's resume check reads merge_commit. A fast-forward entry must contain what is ahead of it; a clash names the files and the nearest entry ahead that changed one (clashed_with); an entry that cannot be built leaves, and those behind are built in the next batch without it. The method is the one asked for when merging into the queue, else the merge queue rule's merge_method, each checked against the repository and the branch's rules; the commit message written then is kept (work 0044: queue_entries merge_method, commit_title, commit_message), else the repository's default, worked out when first built and kept in the tested stack (QueueStackItem method, message, author) so that narrowing a failure rebuilds the same commits; probes are built the same way. The landing gate judges the entry's method. No sandbox builds queue states any more: queue_build returns no jobs. Docs: the merge queue guide (how entries are tested and land, by method; no plan needed for building, merge_group runs billed as workflow runs), pull requests' merge methods, the work architecture page, and the API's merge and queue descriptions (OpenAPI regenerated). Deploy work migration 0044 first, then repos, then work. Checked: cargo test for g1t-repos (341, new: each entry lands by its method on the state ahead, a fast-forward entry is its own head and a clash names the files) and g1t-work (186, new: an entry keeps how it lands and one queued before methods merges; a clash names the nearest entry ahead), the API's OpenAPI test, the CLI's operation table and reference.b5c43d3History
README.md
g1t/detect
Says what a g1t.page build of the checkout would find, without building or deploying anything: the same rules g1t/deploy builds with.
steps:
- uses: actions/checkout@v5
- id: detect
uses: g1t/detect@v1
- if: steps.detect.outputs.kind == 'none'
run: echo "::error::Nothing here would be served."
- if: steps.detect.outputs.framework == 'hugo'
run: echo "A Hugo site, served from ${{ steps.detect.outputs.output-dir }}"
- if: steps.detect.outputs.install != ''
run: ${{ steps.detect.outputs.install }}yaml
Inputs
| Input | Default | |
|---|---|---|
root-dir | The checkout | Where the project is in the checkout. |
build-command | None | A build command of your own, as g1t/deploy takes it. |
output-dir | None | An output folder of your own, as g1t/deploy takes it. |
build-as | auto | What it is built as: auto, static, workers or docker (Container), as g1t/deploy takes it. |
dockerfile | None | A Dockerfile's path of your own, as g1t/deploy takes it. |
Outputs
| Output | |
|---|---|
kind | workers (a wrangler.jsonc, wrangler.json or wrangler.toml, a framework's Cloudflare adapter, or a Node.js server run as a Worker), static (a site a build writes), html (the files as they are), container (a server that is not run as a Worker, a framework set up to need a server, or a Dockerfile with nothing else to serve) or none (nothing to serve). |
framework | The framework it is built with, from the table below; node for a Node.js server run as a Worker; for a container without one, dockerfile, or the server g1t writes a Dockerfile for (node, django, flask, fastapi, python, go, rails, rack, laravel, php). Empty when there is none. |
dockerfile | A container's own Dockerfile, relative to the project. Empty when g1t writes one, and for anything that is not a container. |
port | The port a container listens on: its Dockerfile's EXPOSE, else 8080. |
entry | For a Node.js server run as a Worker, the file that starts it, relative to the project. Empty for anything else. |
install | How dependencies are installed, from the lockfile: npm ci, pnpm install --frozen-lockfile, and so on, and bundle install for Jekyll. Empty when there is nothing to install. |
build | What builds it: your command, the package's build script (npm run build), the framework's own command, or wrangler deploy --dry-run for Workers. Empty when nothing does. |
output-dir | The folder served, relative to the project (. for the project itself): yours, a Workers project's assets directory, the folder the framework writes, or the first of dist, build, out, public, _site, .output/public that exists. Empty when a build has yet to write it, as for a Cloudflare adapter's Worker. |
When kind is none, the step's log says what to change or add, such as
a Dockerfile. Docker deployments
has the order in which a project's kind is decided.
Frameworks
framework | Found by | Serves |
|---|---|---|
vite, vite-react, vite-vue, vite-svelte, vite-solid, vite-preact | vite, and the framework's Vite plugin | dist, or build.outDir |
astro | astro | dist, or outDir; a Worker with @astrojs/cloudflare |
nextjs | next | out with output: 'export'; a Worker with @opennextjs/cloudflare and a wrangler.jsonc |
sveltekit | @sveltejs/kit | build with adapter-static; a Worker with adapter-cloudflare |
nuxt | nuxt | .output/public, from nuxi generate |
react-router, remix | @react-router/dev, @remix-run/dev | build/client with ssr: false |
gatsby | gatsby | public |
docusaurus | @docusaurus/core | build |
eleventy | @11ty/eleventy | _site |
angular | @angular/core and angular.json | dist/<project>/browser |
create-react-app | react-scripts | build |
hugo | hugo.toml, or Hugo's config.toml | public |
jekyll | _config.yml and a Gemfile | _site |
mdbook | book.toml | book |
What deploys has
each one's build command. Your build-command and output-dir win over
what is detected.
It runs g1t-runner detect, found or downloaded as g1t/deploy finds it.