mirror of
https://github.com/vitorpamplona/amethyst.git
synced 2026-10-06 11:48:24 +00:00
Claude
75a0f738c8
LegacyPreferenceCleanup runs after each successful account load and deletes that account's `secret_keeper_<npub>` file only when it can show nothing in it would be lost. It refuses today, deliberately, and says why. The check is not one rule, because the two halves of the migration are in different states. The plain per-account groups stopped being legacy-written when they moved, so that file is a frozen snapshot of the day they migrated — comparing values would flag every setting the user has changed since. For those the question is "did the copy run", which the migration marker answers exactly: CopyOnceMigration commits the values and the marker as one Preferences, so the marker cannot be set without them. The secrets and the private key *are* still written to both stores, so for those the stronger question is available and is asked: read both back, require them to agree. Coverage runs from the file's own keys rather than a checklist. A checklist fails silently in the one direction that matters — a key added later that no migration carries — so instead every key present must be claimed by a table, be one of the secrets, or be on the accepted-loss list, and anything else stops the deletion by name. LegacyKeyCoverageTest reflects over PrefKeys and fails the build if a key falls outside all of those, so that surfaces at the commit that adds it rather than as a file that quietly never gets deleted. A store that cannot be read is a reason, never a pass. Nothing is deleted yet, and not because the check fails: LEGACY_WRITES_RETIRED is false. The identity, key, secret and roster stores still mirror into the legacy file so a rolled-back build finds a complete account, and while that is true, deleting the file achieves nothing — the next save recreates it — and would look like it had worked. Retiring those writes ends the rollback window and waits on the device pass, which nothing here has had: no AndroidKeyStore path has executed in this environment, so what is tested is the decision logic against fakes. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AXvKXakvup4inNFfAhhr4L