.gitea/workflows/check.yml

@@ -9,13 +9,10 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest, macos-latest]
+        os: [linux-base]
     runs-on: ${{ matrix.os }}
     steps:
       - uses: actions/checkout@v4
-      - uses: prefix-dev/setup-pixi@v0.9.4
-        with:
-          pixi-version: v0.69.0
       - name: Check the repository
         run: pixi run check
       - name: Stress the reference implementation

.gitea/workflows/site.yml

@@ -1,33 +1,84 @@
-# Builds the website on every push to main and publishes it when the SITE_* secrets are set; see deploy/README.md.
+# Builds and checks the website on every push to main, and publishes it when SITE_PATH is set; see deploy/README.md.
 name: site
 on:
   push:
     branches: [main]
   workflow_dispatch:
+permissions:
+  contents: read
+# One publish at a time, so two pushes cannot interleave their transfers; a later run waits instead of cancelling
+# one halfway through.
+concurrency:
+  group: site-publish
+  cancel-in-progress: false
+defaults:
+  run:
+    shell: bash
 jobs:
   site:
-    runs-on: ubuntu-latest
+    runs-on: linux-base
+    timeout-minutes: 15
     steps:
       - uses: actions/checkout@v4
-      - uses: prefix-dev/setup-pixi@v0.9.4
         with:
-          pixi-version: v0.69.0
+          persist-credentials: false
+      # --locked fails when pixi.lock is out of date with pixi.toml instead of solving a new environment.
+      - name: Install the environment
+        run: pixi install --locked
+      # Everything but the byte-for-byte rebuild, including that every page, link and $id of the site resolves.
+      - name: Check the repository and the site
+        run: pixi run --frozen quick
       - name: Build the site
-        run: pixi run python tools/site.py --out _site
+        run: pixi run --frozen site --out _site
       - name: Publish the site
+        if: github.ref == 'refs/heads/main'
         env:
           SITE_HOST: ${{ secrets.SITE_HOST }}
           SITE_USER: ${{ secrets.SITE_USER }}
           SITE_PATH: ${{ secrets.SITE_PATH }}
           SITE_SSH_KEY: ${{ secrets.SITE_SSH_KEY }}
+          SITE_KNOWN_HOSTS: ${{ secrets.SITE_KNOWN_HOSTS }}
         run: |
-          if [ -z "$SITE_HOST" ]; then
+          if [ -z "$SITE_PATH" ]; then
-            echo "SITE_HOST is not set, so the site was built but not published."
+            echo "SITE_PATH is not set, so the site was built and checked but not published."
             exit 0
           fi
-          command -v rsync >/dev/null || (sudo apt-get update && sudo apt-get install -y rsync)
+          # rsync --delete empties the folder it is given, so take only an absolute path at least two levels deep.
-          mkdir -p ~/.ssh
+          dest=${SITE_PATH%/}
-          printf '%s\n' "$SITE_SSH_KEY" > ~/.ssh/site_key
+          case "$dest" in
-          chmod 600 ~/.ssh/site_key
+            /*/*) ;;
-          ssh-keyscan -H "$SITE_HOST" >> ~/.ssh/known_hosts
+            *) echo "SITE_PATH must be an absolute folder such as /srv/paperspec, not '$SITE_PATH'." >&2; exit 1 ;;
-          rsync -az --delete -e "ssh -i ~/.ssh/site_key" _site/ "$SITE_USER@$SITE_HOST:$SITE_PATH/"
+          esac
+          case "$dest/" in
+            */../*|*/./*|*//*|*[[:space:]]*) echo "SITE_PATH must be a plain path, not '$SITE_PATH'." >&2; exit 1 ;;
+          esac
+          if [ ! -f _site/index.html ] || [ ! -f _site/404.html ]; then
+            echo "_site/ has no index.html or 404.html, so it was not published." >&2
+            exit 1
+          fi
+          command -v rsync >/dev/null || { echo "rsync is not installed on the runner." >&2; exit 1; }
+          # --checksum without --times leaves unchanged files untouched, so their Last-Modified and ETag stay valid
+          # in caches; --delay-updates and --delete-delay apply the changes together at the end of the transfer.
+          opts=(--recursive --links --perms --chmod=D755,F644 --checksum --delay-updates --delete-delay --stats)
+          if [ -z "$SITE_HOST" ]; then
+            # The runner is the web server.
+            if [ ! -d "$dest" ]; then
+              echo "$dest does not exist on the runner." >&2
+              exit 1
+            fi
+            rsync "${opts[@]}" _site/ "$dest/"
+            exit 0
+          fi
+          if [ -z "$SITE_USER" ] || [ -z "$SITE_SSH_KEY" ] || [ -z "$SITE_KNOWN_HOSTS" ]; then
+            echo "Publishing over SSH needs SITE_USER, SITE_SSH_KEY and SITE_KNOWN_HOSTS as well as SITE_HOST." >&2
+            exit 1
+          fi
+          # The key and host keys live in a private folder that is removed however the step ends.
+          keys=$(mktemp -d)
+          trap 'rm -rf "$keys"' EXIT
+          printf '%s\n' "$SITE_SSH_KEY" > "$keys/id"
+          printf '%s\n' "$SITE_KNOWN_HOSTS" > "$keys/known_hosts"
+          chmod 600 "$keys/id"
+          ssh="ssh -i $keys/id -o IdentitiesOnly=yes -o BatchMode=yes -o StrictHostKeyChecking=yes"
+          ssh="$ssh -o UserKnownHostsFile=$keys/known_hosts -o ConnectTimeout=20 -o ServerAliveInterval=15"
+          rsync "${opts[@]}" --compress -e "$ssh" _site/ "$SITE_USER@$SITE_HOST:$dest/"

deploy/README.md

@@ -41,16 +41,22 @@ Add it to the server's Caddy configuration, for example with an `import` line, a
 
 ## Publish from CI
 
-The workflow `.gitea/workflows/site.yml` builds the site on every push to `main`, and publishes it with `rsync` over SSH when four repository secrets are set:
+The workflow `.gitea/workflows/site.yml` runs on every push to `main`.
+It installs the environment from `pixi.lock`, runs `pixi run quick`, builds the site, and publishes it with `rsync` when the repository secret `SITE_PATH` is set.
+A failed check stops the publish, so the site never serves a release that does not pass its own checks.
+Without `SITE_PATH` the workflow still builds and checks the site, and says that it did not publish.
 
 | Secret | Holds |
 | --- | --- |
-| `SITE_HOST` | The server's host name. |
+| `SITE_PATH` | The folder Caddy serves, such as `/srv/paperspec`: an absolute path at least two levels deep. |
+| `SITE_HOST` | The server's host name. Leave it unset when the runner is the web server, and the site is copied to `SITE_PATH` on the runner. |
 | `SITE_USER` | The account that owns the site folder. |
-| `SITE_PATH` | The folder Caddy serves, such as `/srv/paperspec`. |
 | `SITE_SSH_KEY` | A private SSH key whose public half is in that account's `authorized_keys`, and that can do nothing else. |
+| `SITE_KNOWN_HOSTS` | The server's host keys, as lines of `known_hosts`. |
 
-Without `SITE_HOST` the workflow still builds and checks the site, and says that it did not publish.
+The last three are needed only with `SITE_HOST`.
+Take `SITE_KNOWN_HOSTS` from `ssh-keyscan -H <host>` on a network you trust, and compare its fingerprints with `ssh-keygen -lf` on the server; the workflow refuses a server whose key does not match.
+Copying only changed files keeps the `Last-Modified` and `ETag` of every other file, so caches stay valid across releases.
 
 ## Distributing packages