test: add rust-hook migration test for tags

[swarna1101]

@igamigo @gabrielbosio , follow-up to #2318. fixture v1->v2 on real tags; v2 uses M::up_with_hook to migrate legacy account-only source blobs to NoteTagSource wire format, via apply_migrations_with, checked with get_note_tags.

[juan518munoz]

@swarna1101 I have pushed some tests that possibly raise issues with the hook implementation rather than the tests themselves, could you look into it?

[swarna1101]

@swarna1101 I have pushed some tests that possibly raise issues with the hook implementation rather than the tests themselves, could you look into it?

the hook was using NoteTagSource::read_from_bytes as a skip probe, which misread 15-byte account ids starting with discriminant 2 as User. Fixed by checking len == AccountId::SERIALIZED_SIZE first; all three tests pass now.

[swarna1101]

hi @juan518munoz , let me know if you get a chance to take a look at this

[juan518munoz]

The happy path looks good. One gap worth naming before this lands: the schema fingerprint only covers sqlite_schema, so it cannot cross-check that a recorded user_version matches the data. Any migration with no footprint is invisible to it.

Concretely, a db stamped at v2 that never ran the hook is accepted and the re-encode is silently skipped:

#[test]
fn unencoded_rows_at_the_latest_version_go_undetected() {
    let mut conn = open_tags_db_at_v1();
    let account_id = test_account_id();
    insert_tag_row(&conn, NoteTag::with_account_target(account_id), &account_id.to_bytes());

    // Claim v2 without running the hook.
    conn.pragma_update(None, "user_version", 2).unwrap();

    // v2 has empty SQL, so both versions fingerprint identically and there is nothing to catch.
    assert_eq!(TAGS_FIXTURE_EXPECTED_SCHEMA_HASHES[0], TAGS_FIXTURE_EXPECTED_SCHEMA_HASHES[1]);
    apply_tags_fixture_migrations(&mut conn).expect("accepted despite legacy rows");
    assert_eq!(untransformed_source_count(&conn), 1);
}

Let's fix it here rather than leave it as a known hole.

[swarna1101]

The happy path looks good. One gap worth naming before this lands: the schema fingerprint only covers sqlite_schema, so it cannot cross-check that a recorded user_version matches the data. Any migration with no footprint is invisible to it.

Concretely, a db stamped at v2 that never ran the hook is accepted and the re-encode is silently skipped:

#[test]
fn unencoded_rows_at_the_latest_version_go_undetected() {
    let mut conn = open_tags_db_at_v1();
    let account_id = test_account_id();
    insert_tag_row(&conn, NoteTag::with_account_target(account_id), &account_id.to_bytes());

    // Claim v2 without running the hook.
    conn.pragma_update(None, "user_version", 2).unwrap();

    // v2 has empty SQL, so both versions fingerprint identically and there is nothing to catch.
    assert_eq!(TAGS_FIXTURE_EXPECTED_SCHEMA_HASHES[0], TAGS_FIXTURE_EXPECTED_SCHEMA_HASHES[1]);
    apply_tags_fixture_migrations(&mut conn).expect("accepted despite legacy rows");
    assert_eq!(untransformed_source_count(&conn), 1);
}

Let's fix it here rather than leave it as a known hole.

Gave v2 a tags_encoding schema marker so a database stamped at v2 with a v1 schema now fails the fingerprint check instead of silently skipping the hook.