FlClash: delay tests fail after rename

Record whether FlClash still opens web pages before and after the subscription renames nodes, and whether the same manual-selection group and the groups that reference it keep showing Timeout. This test probes latency to a test address; it is not a bandwidth test.

Project issue #2635 reports this on Windows with FlClash 0.8.99+2026100301. At this review it remained a user report without a maintainer-confirmed general fix. The diagnosis and recovery advice come from v0.8.99 source code; not every Timeout has the same cause.

When v0.8.99 tests a select group, it first reads selectedMap from the Profile—the previously saved node name—then finds the node to test. If the old name is no longer in the current configuration, a missing test target can also return value: -1. That result alone does not establish a real network timeout.

Look for changed node names after an update to the same Profile, while real requests still succeed through the expected proxy and failures are limited to the original selection group or groups that reference it. The known report is from Windows. Other platforms should use this guidance only if their own name and request comparisons match; it does not establish that every platform is affected.

Make a local backup of the current Profile, overrides and settings. Record the full client version, group name, old and new node names, and test URL. Subscription links may contain credentials, so keep backups and original configurations on your device.

Keep the same network, Profile and test URL. Expand the affected manual selection group on the proxies page and confirm that the new name appears among its current members. Run one delay test for the new node and one for the group, waiting for each to finish. Repeated batch refreshes can hide the difference.

Open a normally accessible HTTPS page with that group's current selection, then inspect FlClash connection records. A useful successful-connection comparison must enter FlClash, use the expected proxy and finish loading. A browser request that bypasses the proxy does not qualify.

If visible logs say that the old node name is missing from the applied config, compare it with the current member list. Missing logs do not rule out a stale selection, and a failure number alone does not identify the cause. Do not edit the database or expose the control interface to the LAN just to collect evidence.

On the nodes page, reselect a valid node from the current list and wait for saving and switching to finish. If a database-write or switch-failure notification appears, stop clicking repeatedly and keep the error.

v0.8.99 saves the selection in selectedMap and rolls it back if saving or switching the core fails. Restarting alone may not clear an old record. Reselect using the steps below and check again, but do not treat v0.8.99 as a confirmed fix for this report.

If the old selection returns after a restart, keep the backup and original Profile. Use the normal new-file or URL import entry to add the same trusted subscription as another Profile with a distinct label. Do not delete the original or restore a full app backup to create this comparison.

In v0.8.99, a new import creates a separate Profile with an empty saved selection for comparison with the original Profile. Rules supplied by the subscription are still imported with its content. Do not assume that extra rules, scripts or customizations tied to the old Profile have migrated; app-level overrides may also affect the copy. Check the running configuration first.

Check the copy's node list and required rules first, then select a valid member and switch briefly only when ready. If the copy works while the original reproduces the problem, keep it as a temporary recovery option and check necessary customizations individually. This still does not replace a maintainer's root-cause confirmation.

If the new copy also fails, restore the previously working Profile and settings changed during this attempt. Stop FlClash traffic handling and confirm ordinary networking works. Pause tasks that must use a proxy; direct access is not equivalent recovery for them. Do not clear all data, edit the database, disable security software or promise that installing an older version will cure the problem.

Recovery requires a valid selected name, sensible delay-test and real-request comparisons, and a selection that survives a full exit. A green delay number means one probe succeeded, not that every website or protocol works. Even if the test URL is unreachable, record successful real requests separately.

If the problem remains, report the full FlClash and operating-system versions, how the node name changed, current group members, individual and group tests using the same URL, the actual outbound for real requests, and results after reselection and a full restart. In public evidence, use consistent placeholder names for groups and nodes, and redact subscription credentials, passwords, personal addresses and the control-interface secret.

If the results match issue #2635, add a redacted comparison. Describe differences separately for invalid YAML, a disconnected node or failures confined to batch testing.