do not try to fetch remote tags if there is no internet

[nicoschmdt]

fix: #1788

[github-actions]

Automated PR Review

0. Summary

• Verdict: MINOR SUGGESTIONS ✏️

Stops the version chooser from hanging when DockerHub is unreachable by adding a 60s/10s aiohttp timeout and catching network-level exceptions in set_remote_versions. On the frontend it also refactors the internet-state handling in VersionChooser.vue (splitting is_offline from can_fetch_remote) and defers the initial remote fetch in NewVersionNotificator.vue until connectivity is available.

1. Correctness & Implementation Bugs

• 1.1 [minor] core/frontend/src/components/app/NewVersionNotificator.vue:56 — the new can_fetch_remote watcher calls this.run() unconditionally when connectivity returns, but run() has no re-entrance guard. If mounted() is still awaiting loadCurrentVersion() / loadAvailableVersions() when the internet state transitions, two overlapping run() invocations race and can leave available_versions, latest_beta, latest_stable, and should_open in an inconsistent state (last-writer-wins on the shared component fields). Consider a simple in-flight flag (or reuse a promise) before firing another run().
• 1.2 [minor] core/services/versionchooser/utils/chooser.py:379 — except (aiohttp.ClientConnectionError, OSError) is broad enough to swallow non-network OSError subclasses (e.g. an unexpected filesystem/permission error surfaced from a lower layer) and mislabel them as NO_INTERNET_ERROR ("Unable to reach DockerHub…"). Catching OSError is necessary to pick up asyncio.TimeoutError on 3.11+ (it aliases TimeoutError, which is an OSError), but consider narrowing to (aiohttp.ClientConnectionError, aiohttp.ServerTimeoutError, asyncio.TimeoutError, socket.gaierror) so a genuinely unexpected error still falls through to the logger.critical branch instead of being silently reported as "no internet".

6. Code Quality & Style

• 6.1 [minor] core/frontend/src/components/version-chooser/VersionChooser.vue:344 — the second branch in remote_alert_message (if (this.available_versions.error === this.offline_error_message) return null) is dead. resetToNoInternetAvailable no longer writes offline_error_message into available_versions.error (it's now null), and the backend's NO_INTERNET_ERROR is a different string ("Unable to reach DockerHub, can't fetch remote images"). Either drop the check, or intentionally match against the backend NO_INTERNET_ERROR constant if the goal is to avoid showing "no internet" twice when the frontend and backend disagree.
• 6.2 [nit] core/frontend/src/components/version-chooser/VersionChooser.vue:709 — with error: null, resetToNoInternetAvailable is now just clearing remote and the two latest_* fields; the alert text is entirely derived by remote_alert_message. Consider renaming to reflect the new intent (e.g. clearRemoteVersions) so the method name still describes what it does.
• 6.3 [nit] core/services/versionchooser/utils/dockerhub.py:15 — the three sessions all take the same timeout; extracting the pattern (or attaching the timeout to a shared ClientSession) would DRY things up. Not required, just noted since the file already has one session created per call (which AGENTS.md flags as a common pitfall).

7. Tests

• 7.1 [minor] core/services/versionchooser/test_versionchooser.py — no coverage is added for the new exception branch in set_remote_versions or the DOCKERHUB_TIMEOUT behavior. A small test that patches TagFetcher.fetch_remote_tags to raise aiohttp.ClientConnectionError and asserts output["error"] == NO_INTERNET_ERROR and output["remote"] == [] would lock the intended behavior in place.

Generated by PR Review Bot. This is advisory, a human reviewer must still approve.