SubMaker v1.4.74
Bug Fixes:
-
Fixed config saves minting unintended fresh sessions after transient update failures: The config page previously fell back to creating a brand-new session on generic update errors or network failures, which could strand users on a fresh empty state and make history/session-backed data appear lost. Save now preserves the current token and asks the user to retry when session storage returns
503, when update requests fail, or when the network flakes out, instead of silently generating a replacement session. Added regression coverage for the new stable history identity and the empty-history no-scan path. -
Fixed translation history disappearing after saving settings: History identity was effectively tied to the recomputed config hash, so saving settings over the same session token could still move the user onto a different empty history namespace and make all prior cards appear to vanish. Sessions now persist a dedicated stable history user hash derived from the session token, and history reads/writes now prefer that stable session-scoped namespace instead of relying on the mutable config hash. Added a one-time legacy bridge on the history page so existing config-hash-based history is copied into the stable namespace when detected, preserving older entries after the upgrade.
-
Fixed empty translation history pages still taking 20s+ on large Redis deployments: The
/sub-historyroute could still fall back to the legacy per-user RedisSCANrecovery path when both the aggregated history store key and the sorted-set index were missing, so a user with no history could still pay the cost of walking a large shared Redis keyspace. Added a session-scoped fast path that skips the slow scan for modern empty history namespaces, so truly empty first-load history pages now return immediately instead of stalling on Redis keyspace scans. Also added stale-index cleanup during indexed reads so expired history entry IDs are pruned from the Redis zset instead of causing repeated dead lookups on later page loads. -
Fixed the history page still feeling dead while stale Redis reads were in progress:
/sub-historyno longer blocks first paint on the bounded stale-path history lookup. The route now returns the page shell immediately, shows a dedicated loading state, and fetches the history list asynchronously from a separate endpoint so the page stays responsive even when Redis is slow. Added an inline retry state for fragment-load failures instead of dumping the user onto a blank error page. -
Extended the history fast-path freshness window from 15s to 60s: The aggregated history store key is now trusted for 1 minute before the page falls back to the bounded Redis index refresh path. This keeps hot history pages on the cheap single-GET path longer and reduces how often users hit the slower indexed recovery path on busy public deployments.