fix(update): 聚合 Spark 和 APM 升级通知

docs/superpowers/specs/2026-04-15-update-notifier-apm-aggregation-design.md

@@ -0,0 +1,129 @@
+# Update Notifier APM Aggregation Design
+
+## Background
+
+`tool/update-upgrade/ss-update-notifier.sh` currently counts Spark (`aptss`) updates, filters them through `hold` state and `~/.config/spark-store/ignored_apps.conf`, then sends one desktop notification. A separate APM-side notifier pattern exists, but it is not merged into the current Spark notifier script.
+
+The goal is to let the current notifier aggregate both Spark and APM upgradable items into one notification, while keeping the existing user-level ignored-update behavior and avoiding hard failures on systems that do not provide `aptss`.
+
+## Goals
+
+1. Keep a single notifier script: `tool/update-upgrade/ss-update-notifier.sh`.
+2. Count both Spark and APM upgradable applications in that script.
+3. Continue to use one shared ignored-update file: `~/.config/spark-store/ignored_apps.conf`.
+4. Apply ignored filtering to both Spark and APM using exact `pkgname|newVersion` keys.
+5. Apply `hold` filtering independently for Spark and APM.
+6. Aggregate the remaining Spark and APM counts into one notification.
+7. If `aptss` is unavailable, skip the Spark branch without failing the script.
+
+## Non-Goals
+
+1. Do not create a second notifier service or script.
+2. Do not change the ignored-update file format.
+3. Do not change Electron or update-center UI behavior in this task.
+4. Do not add a compatibility layer for `/etc/spark-store/ignored_apps.conf`.
+
+## Recommended Approach
+
+Extend the existing notifier in place and keep Spark and APM as two counting branches inside the same script.
+
+Spark keeps its current `aptss`-based flow. APM adds a second branch that parses `apm list --upgradable`, applies APM `hold` detection via `amber-pm-debug dpkg-query`, and reuses the same ignored-entry set already loaded from user config files. The final notification count becomes `spark_count + apm_count`.
+
+This keeps the script small, preserves the current Spark path, and avoids introducing a second source of notification truth.
+
+## Data Sources
+
+### Spark branch
+
+- Command availability gate: `command -v aptss`
+- Refresh commands: `aptss update`, `LANGUAGE=en_US aptss ssupdate`
+- Upgradable list source: `/opt/durapps/spark-store/bin/update-upgrade/ss-do-upgrade-worker.sh upgradable-list`
+- Hold check: `dpkg-query -W -f='${db:Status-Want}' <pkg>`
+
+### APM branch
+
+- Command availability gate: `command -v apm`
+- Refresh commands: `LANGUAGE=en_US apm update`, followed by `apm clean`
+- Upgradable list source: `env LANGUAGE=en_US apm list --upgradable`
+- Output compatibility: support both `[upgradable from: <version>]` and legacy `[from: <version>]` variants when extracting the current version
+- Hold check: `amber-pm-debug dpkg-query -W -f='${db:Status-Want}' <pkg>`
+
+## Filtering Rules
+
+### Ignored entries
+
+The script continues to load ignored entries from `~/.config/spark-store/ignored_apps.conf`, using the existing user-detection plus `/home/*` scan behavior.
+
+Each valid line is still interpreted as:
+
+```text
+pkgname|version
+```
+
+Matching rule:
+
+- Spark item is ignored when `pkgname|sparkNewVersion` exists in the ignored set.
+- APM item is ignored when `pkgname|apmNewVersion` exists in the ignored set.
+
+Ignored matching is intentionally source-agnostic. If Spark and APM expose the same package name and target version, one ignore entry suppresses both.
+
+### Hold entries
+
+- Spark item is excluded if `dpkg-query` reports `hold`.
+- APM item is excluded if `amber-pm-debug dpkg-query` reports `hold`.
+
+### Invalid or stale version entries
+
+Each branch keeps its own version sanity check before counting:
+
+- Spark continues to skip items where `newVersion <= currentVersion`.
+- APM does the same after parsing `apm list --upgradable` output from either supported bracket variant.
+
+## Availability Rules
+
+### Missing `aptss`
+
+If `aptss` is not installed or not in `PATH`:
+
+1. Skip Spark refresh commands entirely.
+2. Skip Spark upgradable counting entirely.
+3. Continue with APM counting if `apm` is available.
+
+### Missing `apm`
+
+If `apm` is not installed or not in `PATH`:
+
+1. Skip APM refresh commands entirely.
+2. Skip APM upgradable counting entirely.
+3. Continue with Spark counting if `aptss` is available.
+
+### Both unavailable
+
+If both `aptss` and `apm` are unavailable, the script exits without sending a notification.
+
+## Notification Behavior
+
+The script sends one notification only when:
+
+```text
+spark_effective_count + apm_effective_count > 0
+```
+
+The notification remains a single desktop message. The implementation may update the wording to mention both Spark and APM updates, but the key requirement is one aggregated notification rather than separate per-source notifications.
+
+## Implementation Boundaries
+
+1. Keep the current `detect-notify-user` and ignored-config discovery logic.
+2. Add APM parsing as a second source-specific helper path instead of rewriting the whole script.
+3. Keep the shell implementation POSIX-compatible with the current Bash usage already present in the file.
+4. Avoid changing unrelated installer or update-center code in this task.
+
+## Verification
+
+1. `bash -n tool/update-upgrade/ss-update-notifier.sh`
+2. Manual dry-run reasoning for all four cases:
+   - Spark only
+   - APM only
+   - Spark + APM
+   - neither available
+3. Confirm ignored entries suppress both branches via exact `pkg|newVersion` matching.