Part of VAULT-83.
Why
The formats are tested piece by piece:
VaultSlotFileTests pins the header offsets, the HKDF labels, the byte-flip checks and the identical slot lengths.Argon2idKeyDeriverTests has RFC 9106 known-answer tests.VaultBackupEncryptorTests has a known-answer test for an empty backup.- Encrypted items are only round-tripped (
VaultItemEncryptorTests, VaultItemDecryptorTests).
A change made on both the write side and the read side still passes all of these. Examples: the AAD layout (header with the slot size zeroed ‖ index ‖ nonce), the key box layout, a renamed payload field, or a padding change. Existing vaults, notes and backups would then stop opening. The README promises that old backups can always be restored, and vaults are on people's phones now.
Fixtures to add
Put them under the tests, with a README that says for each one: how it was made (the commit or release tag, and the script), the password, and what it should contain. The script must fix salts and nonces, or record them.
- Slot file: a whole
vault-slots.v1 file with several slots: a real vault, a duress vault and random slots. Each password opens its own slot to exactly the expected items, tags and settings. A wrong password opens nothing.
- Encrypted items: an encrypted note and a recovery phrase, as a shipped build stores them. They must decrypt to the expected text with the known password.
- Backup encryption: a known-answer test for a populated backup at a VAULT-75 bucket size, plus one from before VAULT-75's padding. Whole PDFs and QR codes belong to the restore sub-issue. This one pins the
VaultBackupEncryptor / VaultBackupDecryptor bytes.
- Calibration: a slot file made with one set of Argon2 parameters opens on a "device" that would calibrate differently. The design doc's "Test strategy" lists this test, but it doesn't exist.
Also
- Zeroing: the design doc says some buffers are zeroed with
memset_s: key-box and body plaintexts, the JSON, and the compression scratch space. Where it's practical, test that, the way Argon2idKeyDeriverTests does for the KDF's working memory. Where it isn't practical, say so in "Residual limits".
- Rule, written in the fixtures README: never regenerate a fixture to make a test pass. A format change adds a new fixture and keeps every old one.
Done when
Each kind of fixture above is in the repo and opened by a test. The README explains how to add a fixture when a format changes.
