Files
thoughtsync/desktop/packaging/publish-release.sh
T
bvandeusenandClaude Opus 5 72968897ab
CI & Build / Build now, or wait for Android? (push) Successful in 3s
Android / Build, or is the channel already serving this? (push) Successful in 3s
CI & Build / Python lint (push) Successful in 2s
CI & Build / TypeScript typecheck (push) Successful in 6s
Desktop (Tauri) / Build, or is the channel already serving this? (push) Successful in 3s
CI & Build / Python tests (push) Successful in 10s
CI & Build / integration (push) Successful in 44s
CI & Build / Build & push image (push) Skipped
Desktop (Tauri) / Tauri desktop (Linux) (push) Failing after 3m20s
Desktop (Tauri) / Windows installer (cross-compiled) (push) Successful in 3m20s
Desktop (Tauri) / Update manifest (push) Skipped
Android / Kotlin + Rust (APK) (push) Successful in 9m13s
channels: the dev channel publishes on dev-rolling, so its tag stops shadowing the branch
The rolling dev release lived on a tag named `dev`, beside the branch
named `dev`. Once a clone had fetched tags, `git push origin dev` failed
with "src refspec dev matches more than one" (Scribe #2184, note #3042),
and every session had to know to spell out refs/heads/dev.

The channel is still `dev` everywhere a person sees it: the app's
setting, `install.sh --channel dev`, the stored pref. Only the release
tag moves, to `dev-rolling`, matching roundtable-android. `stable` has no
branch to collide with and keeps its name.

- packaging/channel-tag.sh is the one channel -> tag mapping CI reads:
  the publish steps in android.yml and desktop.yml, the manifest job,
  fetch-clients.sh and guard-forward.sh. guard-forward exits 2 on an
  unmapped channel instead of fetching an empty URL and passing.
- update.rs and install.sh carry their own copy because neither can run
  it; update.rs gains a test that no channel feed is named like a branch.
- tests/test_channel_tag.py runs the script: no tag is a branch name,
  dev is exactly dev-rolling, an unknown channel fails with no output.
- publish-release.sh titles the release "ThoughtSync dev (rolling)", so
  the tag name does not leak into what people read.

TEMPORARY bridge: desktop apps installed before this have
.../download/dev/latest.json compiled in. The dev manifest job sets
BRIDGE_TAG=dev, and write-manifest.sh writes the same latest.json to
the old `dev` release. Its URLs name dev-rolling assets, so those apps
update once into a build that reads the new tag. The bridge, and the old
release and tag, are removed once installed apps have crossed over.
Until then the push still needs the explicit refspec, as
ci-requirements.md now says.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DwoKYuw3qJmUUYsJeNherB
2026-09-10 18:34:11 -04:00

207 lines
11 KiB
Bash
Executable File

#!/usr/bin/env bash
#
# Publish the built desktop bundles as assets on a Fabled-Git (Forgejo) Release.
#
# WHY: CI Actions artifacts are ephemeral, per-run, and auth-gated — useless as a
# distribution/fetch target. The install script (install.sh) and the in-app
# updater both need a STABLE, versioned URL. A Release attached to the pushed
# `v*` tag is that target: `/releases/latest` always points at the newest one,
# and each asset has a permanent browser_download_url.
#
# WHEN: CI-only, and ONLY on a `v*` tag build (the workflow gates this step with
# `if: startsWith(github.ref, 'refs/tags/v')`). Cutting the tag is the operator's
# action (rule 2) — this script never creates a tag, it only publishes a Release
# for a tag that already exists.
#
# The build + de-bundle steps run first; this consumes their output:
# target/release/bundle/appimage/*.AppImage (de-bundled)
# target/release/bundle/deb/*.deb
# target/release/bundle/arch/*.pkg.tar.* (prebuilt pacman)
# target/x86_64-pc-windows-msvc/release/bundle/nsis/*.exe
#
# Instance-agnostic: server + repo come from the runner's github.* context
# (Forgejo populates them for compatibility), so nothing is hardcoded to one host.
#
# Idempotent: re-running for the same tag reuses the existing Release and
# replaces same-named assets, so a re-run (or workflow_dispatch retry) is safe.
set -euo pipefail
: "${GITHUB_TOKEN:?GITHUB_TOKEN is required (runner-injected; needs contents:write)}"
: "${GITHUB_SERVER_URL:?GITHUB_SERVER_URL is required}"
: "${GITHUB_REPOSITORY:?GITHUB_REPOSITORY is required (owner/repo)}"
: "${GITHUB_REF_NAME:?GITHUB_REF_NAME is required (the tag, e.g. v0.1.0)}"
# The release to publish to. Defaults to the pushed tag (the versioned, stable
# case). The channel publish steps call this with the channel's fixed tag —
# `stable`, or `dev-rolling` for the dev channel (packaging/channel-tag.sh) — to
# maintain a rolling release whose tag never moves, because Forgejo has no
# `/releases/latest/download/<asset>` route for an updater to point at.
RELEASE_TAG="${RELEASE_TAG:-$GITHUB_REF_NAME}"
RELEASE_PRERELEASE="${RELEASE_PRERELEASE:-false}"
API="$GITHUB_SERVER_URL/api/v1/repos/$GITHUB_REPOSITORY"
TAG="$RELEASE_TAG"
AUTH=(-H "Authorization: token $GITHUB_TOKEN")
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
REPO_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
BUNDLE_ROOT="$REPO_ROOT/target/release/bundle"
# Cross-compiled Windows output lands under the target triple, not the host root.
WIN_BUNDLE_ROOT="$REPO_ROOT/target/x86_64-pc-windows-msvc/release/bundle"
# --- collect the assets to upload -------------------------------------------
shopt -s nullglob
# nullglob (set above) drops the patterns that didn't match, which is what lets the
# Linux job and the Windows job each run this script against the SAME release and
# upload only what they actually built — they run in separate workspaces, so neither
# can see the other's bundles. The release is created once and reused (409 path).
# The `.sig` files are the updater's whole trust story — a bundle published without
# its signature is one the app will refuse, so they ship together or not at all.
# They only exist when the build ran with a signing key (M10.9); nullglob drops
# them silently otherwise, which is the correct behaviour for an unsigned build.
# The Android client publishes here too, from its own job and its own workspace
# — the same nullglob arrangement that already lets the Linux and Windows jobs
# share one release. Its two files are staged under android/dist by the workflow:
# a STABLY NAMED apk (a fixed name is the whole point of the fixed `dev` tag —
# Forgejo has no /releases/latest/download route) and the sidecar the server reads
# its version out of, because an APK keeps that in a binary manifest.
ASSETS=(
"$BUNDLE_ROOT"/appimage/*.AppImage
"$BUNDLE_ROOT"/appimage/*.AppImage.sig
"$BUNDLE_ROOT"/deb/*.deb
"$BUNDLE_ROOT"/arch/*.pkg.tar.*
"$WIN_BUNDLE_ROOT"/nsis/*.exe
"$WIN_BUNDLE_ROOT"/nsis/*.exe.sig
"$REPO_ROOT"/android/dist/thoughtsync.apk
"$REPO_ROOT"/android/dist/thoughtsync-android.json
)
# nullglob drops PATTERNS that match nothing — it does nothing for a path with no
# wildcard in it, which stays in the array as a literal and reaches curl as a file
# that isn't there (exit 26). The Android entries above are exactly that shape, and
# adding them broke the desktop publish that had been working. Filter on existence
# instead, which is what the array actually means and covers every entry rather
# than only the ones that happen to contain a `*`.
present=()
for a in "${ASSETS[@]}"; do
[ -f "$a" ] && present+=("$a")
done
ASSETS=("${present[@]}")
if [ ${#ASSETS[@]} -eq 0 ]; then
echo "ERROR: nothing to publish — no desktop bundles under $BUNDLE_ROOT and no APK under $REPO_ROOT/android/dist." >&2
exit 1
fi
echo "==> Publishing release $TAG with ${#ASSETS[@]} asset(s):"
for a in "${ASSETS[@]}"; do echo " $(basename "$a") ($(du -h "$a" | cut -f1))"; done
# curl wrapper that returns the response body on stdout and fails the script on
# an HTTP >=400 that we didn't explicitly allow (via ALLOW_CODES).
api() {
local method="$1" url="$2"; shift 2
local out code
out="$(curl -sS -X "$method" "${AUTH[@]}" -w $'\n%{http_code}' "$url" "$@")"
code="${out##*$'\n'}"
out="${out%$'\n'*}"
if [ "$code" -ge 400 ] && [[ " ${ALLOW_CODES:-} " != *" $code "* ]]; then
echo "ERROR: $method $url -> HTTP $code" >&2
echo "$out" >&2
return 1
fi
printf '%s' "$out"
}
# First integer value of a "id": <n> pair — the release id is the first "id" in
# the release object. Avoids a jq dependency (not guaranteed in the CI image).
first_id() { grep -oE '"id"[[:space:]]*:[[:space:]]*[0-9]+' | head -1 | grep -oE '[0-9]+'; }
# --- create (or reuse) the release for the tag ------------------------------
# The install command printed on a release has to install THAT release's channel.
# install.sh defaults to stable, so the rolling dev release must opt in explicitly
# — otherwise someone following the instructions here lands on a tagged build and
# wonders why the version they were sent isn't what they got.
#
# Both CHANNELS are rolling pointer releases (M314 step 3): `dev` republishes on
# every green push to dev, `stable` on every merge to main. Each says so, because a
# release that prunes its own assets behaves differently from a versioned one and a
# reader deserves to know which they are looking at.
# Keyed on the TAG, which for the dev channel is `dev-rolling`, not `dev` — a tag
# named `dev` shadowed the branch (Scribe #2184). CHANNEL_LABEL is the name a person
# reads in the release title, so the rename does not leak into it.
if [ "$TAG" = "dev-rolling" ]; then
CHANNEL_LABEL='dev (rolling)'
INSTALL_TAIL='sh -s -- --channel dev'
# Backticks BARE, not `\``. The heredoc below is unquoted, so there the backslash
# is the shell's — it suppresses command substitution and never reaches the JSON.
# Here single quotes already do that job, so a backslash would survive into the
# body as `\``, which is not a legal JSON escape: Forgejo answers 422.
CHANNEL_NOTE='\n\nThis is the rolling **dev** channel: republished on every green push to `dev`, and pruned to the current build.'
elif [ "$TAG" = "stable" ]; then
CHANNEL_LABEL='stable'
# install.sh defaults to stable, so no flag.
INSTALL_TAIL='sh'
CHANNEL_NOTE='\n\nThis is the rolling **stable** channel: republished on every merge to `main`, and pruned to the current build. No tag is required for a build to arrive here.'
else
CHANNEL_LABEL="$TAG"
INSTALL_TAIL='sh'
CHANNEL_NOTE=''
fi
echo "==> Creating release for $TAG"
BODY=$(cat <<JSON
{"tag_name":"$TAG","name":"ThoughtSync $CHANNEL_LABEL","draft":false,"prerelease":$RELEASE_PRERELEASE,
"body":"ThoughtSync $CHANNEL_LABEL.\n\n**Desktop**\n\n- **Debian / Ubuntu** — native \`.deb\`\n- **Arch / CachyOS** — native \`.pkg.tar.*\`\n- **everything else** — \`.AppImage\` (de-bundled graphics: renders on any GPU/Wayland setup)\n\nInstall / update — picks the right one for your system:\n\`\`\`\ncurl -fsSL $GITHUB_SERVER_URL/$GITHUB_REPOSITORY/raw/branch/dev/desktop/packaging/install.sh | $INSTALL_TAIL\n\`\`\`\n\n**Android** — \`thoughtsync.apk\`. Copy it and \`thoughtsync-android.json\` into your server's \`/var/thoughtsync/client/\` and the server will offer it to your devices; see docs/android-distribution.md.$CHANNEL_NOTE"}
JSON
)
# An explicit body, replacing the install instructions above.
#
# Used by the RELEASE lane, whose job is a changelog rather than artifacts (M314
# step 7). It goes through this script rather than making its own API calls so that
# the create-or-PATCH-on-409 path is shared: a fixed-tag release that only ever
# POSTs keeps whatever text its FIRST build wrote, which is #2182 exactly, and
# re-implementing that correctly in a second place is how it comes back.
#
# CONTRACT: already JSON-escaped, without surrounding quotes. The caller knows
# whether it has a JSON encoder; this script cannot assume python3 is on PATH in
# every image that sources it.
if [ -n "${RELEASE_BODY_JSON:-}" ]; then
BODY="{\"tag_name\":\"$TAG\",\"name\":\"ThoughtSync $CHANNEL_LABEL\",\"draft\":false,\"prerelease\":$RELEASE_PRERELEASE,\"body\":\"$RELEASE_BODY_JSON\"}"
fi
# 409 = a release for this tag already exists (re-run) — fall through to lookup.
release="$(ALLOW_CODES=409 api POST "$API/releases" -H "Content-Type: application/json" -d "$BODY")"
RELEASE_ID="$(printf '%s' "$release" | first_id || true)"
if [ -z "${RELEASE_ID:-}" ]; then
echo " release exists; fetching it by tag"
release="$(api GET "$API/releases/tags/$TAG")"
RELEASE_ID="$(printf '%s' "$release" | first_id)"
# And refresh its description. A fixed-tag release (dev, and any re-run) keeps
# whatever text the FIRST build wrote — including the install command. A stale
# one sends people to the wrong channel, silently.
echo " refreshing its description"
api PATCH "$API/releases/$RELEASE_ID" -H "Content-Type: application/json" -d "$BODY" >/dev/null
fi
[ -n "${RELEASE_ID:-}" ] || { echo "ERROR: could not resolve release id" >&2; exit 1; }
echo " release id = $RELEASE_ID"
# Existing assets (name -> id), so a re-run replaces rather than duplicates.
existing="$(api GET "$API/releases/$RELEASE_ID/assets")"
# --- upload each asset ------------------------------------------------------
for asset in "${ASSETS[@]}"; do
name="$(basename "$asset")"
# If an asset with this name already exists, delete it first (Forgejo rejects
# a duplicate name). Match the "id" that precedes this asset's "name".
old_id="$(printf '%s' "$existing" \
| grep -oE "\"id\"[[:space:]]*:[[:space:]]*[0-9]+[^}]*\"name\"[[:space:]]*:[[:space:]]*\"$name\"" \
| first_id || true)"
if [ -n "${old_id:-}" ]; then
echo "==> Replacing existing asset $name (id $old_id)"
api DELETE "$API/releases/$RELEASE_ID/assets/$old_id" >/dev/null
fi
echo "==> Uploading $name"
api POST "$API/releases/$RELEASE_ID/assets?name=$name" -F "attachment=@$asset" >/dev/null
done
echo "==> Done. Release $TAG published with $(printf '%s\n' "${ASSETS[@]##*/}" | tr '\n' ' ')"