Attributes Overview
Tayra uses attributes to declaratively mark personal data on your model classes. At runtime, the metadata cache scans your types via reflection and drives the encryption engine automatically.
Fluent API Alternative
All attributes have fluent API equivalents that can be used instead of (or alongside) attributes. The fluent API is useful when you cannot modify the model classes directly, or when you prefer centralized configuration. The expression selectors accept either a property expression (x => x.SomeProperty) or a field expression (x => x.SomeField).
| Attribute | Fluent Equivalent | Builder Options |
|---|---|---|
[DataSubjectId] | e.DataSubjectId(x => x.Prop) | .WithGroup(), .WithPrefix() |
[PersonalData] | e.PersonalData(x => x.Prop) | .WithGroup(), .WithMaskValue(), .WithMaskAfter(), .WithMaskBefore(), .WithMaskEmailDomain(), .WithMask() |
[DeepPersonalData] | e.DeepPersonalData(x => x.Prop) | - |
[SerializedPersonalData] | e.SerializedPersonalData(x => x.Prop) | .WithGroup(), .StoredIn() |
[BlindIndex] | e.BlindIndex(x => x.Prop) | .WithLowercase(), .WithTrim(), .WithTransform(), .StoredIn(), .WithScope(), .WithBitLength() |
[CompoundBlindIndex] | e.CompoundBlindIndex("name") | .Field(), .StoredIn(), .WithScope(), .WithBitLength() |
[ArrayBlindIndex] | - | (attribute mirrors [BlindIndex]: IndexName, IndexPropertyName, Scope, BitLength, Transforms) |
See Fluent API for complete documentation.
Attribute Summary
| Attribute | Target | Purpose |
|---|---|---|
[PersonalData] | string properties or fields | Marks a member for AES-256-GCM encryption |
[DataSubjectId] | Guid or string properties or fields | Identifies the data owner; derives the encryption key ID |
[DeepPersonalData] | Class-type properties or fields | Recurses into nested objects to encrypt their [PersonalData] members |
[SerializedPersonalData] | Non-string properties or fields (int, DateTime, etc.) | Serializes to binary, encrypts, and stores in a companion byte[] member |
[BlindIndex] | string properties or fields with [PersonalData] | Computes an HMAC hash for equality queries on encrypted fields |
[CompoundBlindIndex] | Classes | Combines multiple members into a single HMAC hash for multi-field lookups |
[ArrayBlindIndex] | String collection properties or fields (string[], List<string>, HashSet<string>, ...) with [PersonalData] | Computes one HMAC hash per element into a same-shape companion collection for Contains / Any queries |
Example
Here is a typical annotated model:
public class Customer
{
[DataSubjectId]
public Guid Id { get; set; }
[PersonalData]
public string Name { get; set; } = "";
[PersonalData(MaskValue = "redacted@example.com")]
public string Email { get; set; } = "";
/// <summary>
/// Not annotated - stored and retrieved as plaintext.
/// </summary>
public string AccountType { get; set; } = "";
}In this example:
Idis the data subject identifier. Tayra derives the encryption key ID from this property's value.NameandEmailare personal data fields that will be encrypted in-place.Emailspecifies a custom replacement value for after crypto-shredding.AccountTypehas no annotation and is never encrypted.
How Metadata Discovery Works
When Tayra first encounters a type (via EncryptAsync<T> or DecryptAsync<T>), the PersonalDataMetadataCache scans the type using reflection:
- Member scan - All public instance properties and public instance fields are inspected for attributes.
- Subject resolution - Members with
[DataSubjectId]are recorded, including their group and prefix. - Field classification - Members with
[PersonalData],[DeepPersonalData], or[SerializedPersonalData]are classified by kind:Text,Deep,Serialized,TextCollection, orDeepCollection. - Caching - The result is stored in a
ConcurrentDictionarykeyed byType. Subsequent calls for the same type return the cached metadata with no reflection overhead.
The metadata cache is thread-safe and is registered as a singleton. The reflection cost is paid only once per type for the lifetime of the application.
Properties or Public Instance Fields
All seven attributes ([PersonalData], [DataSubjectId], [DeepPersonalData], [SerializedPersonalData], [BlindIndex], [CompoundBlindIndex], [ArrayBlindIndex]) can be applied to a property or a public instance field. Only public instance members are scanned - private and static members are ignored, exactly like the property scan rule.
Members that Tayra writes back to ([PersonalData] strings, [SerializedPersonalData] values, and their companions) must be writable: a settable property (init-only is fine) or a non-readonly, non-const field. A get-only property or a readonly/const annotated field fails fast with a clear InvalidOperationException the first time the type is encrypted or decrypted. Deep objects and collections are mutated in place and do not need a setter.
Properties are the default idiom in the examples throughout these docs, but a field works identically:
public class Customer
{
[DataSubjectId]
public Guid Id; // public instance field
[PersonalData]
public string Email = ""; // encrypted in place, just like a property
}Roslyn Analyzers
The Tayra.Core package includes Roslyn analyzers that validate attribute usage at compile time:
| Rule | Severity | Description |
|---|---|---|
| TAYRA001 | Warning | Entity with [PersonalData] must have a [DataSubjectId] property |
| TAYRA002 | Info | [DataSubjectId] without [PersonalData] fields is unused |
| TAYRA003 | Error | [DeepPersonalData] must be on a class or record type |
| TAYRA004 | Warning | Multiple [DataSubjectId] properties require Group |
| TAYRA005 | Warning | [BlindIndex] without companion property (e.g. EmailIndex) |
| TAYRA006 | Warning | [BlindIndex] on a non-string property |
| TAYRA011 | Warning | [BinaryEvent] type has [PersonalData] fields |
| TAYRA007 | Warning | [PersonalData] member mapped into a flat-table column |
| TAYRA008 | Warning | [PersonalData] member configured as a duplicated field |
| TAYRA009 | Warning | [PersonalData] member on a Wolverine saga |
| TAYRA010 | Warning | [ArrayBlindIndex] companion shape mismatch or missing companion |
These analyzers catch common mistakes before your code runs. No separate package installation is needed.
See Also
[PersonalData]- Encryption of string fields[DataSubjectId]- Key derivation and grouping[DeepPersonalData]- Nested object encryption[SerializedPersonalData]- Non-string type encryption[BlindIndex]- HMAC blind indexes for searchable encryption[ArrayBlindIndex]- Blind indexes over encrypted string collections- Getting Started - End-to-end tutorial
