Introduction to Kiwi DI
Kiwi DI (Kiwify.Kiwi.DependencyInjection) is an attribute-driven dependency injection framework for .NET. It moves service registration from startup code to the classes themselves, enabling config-driven activation, generic expansion, and declarative construction - all with a single AddKiwiServices call.
The Problem Kiwi DI Solves
Standard .NET DI registration works well at small scale. Five services, ten calls to AddScoped and AddSingleton, one startup entry point - manageable. But as an application grows, that startup code grows with it. At fifty services it is a maintenance problem. At one hundred, it is a liability.
The deeper issue is not the line count. It is the disconnection. A class exists in one file. Its DI registration exists in another. To understand how a class is wired-its lifetime, conditions, and interfaces-you have to look in multiple places. To add a feature flag that controls whether a service is active, you edit both the service and the startup code. To swap an implementation, you find and modify the startup block that wires it.
Kiwi DI inverts this relationship: the class declares its own registration rules. An attribute on the class says what its lifetime is, what conditions it depends on, and how it should be constructed. The framework reads those declarations at startup and wires everything automatically. Startup code becomes trivially short - one call - and stays that way regardless of how many services are added.
Source & Repository
Kiwi Config is open-source and available on GitHub:
https://github.com/kiwifylabs/kiwi-foundation-config-di
Installation
NuGet Package
<PackageReference Include="Kiwify.Kiwi.DependencyInjection" Version="1.0.0" />
<PackageReference Include="Microsoft.Extensions.DependencyInjection" Version="10.0.7" />
Kiwify.Kiwi.Configuration is included transitively.
.NET CLI
dotnet add package Kiwify.Kiwi.DependencyInjection --version 1.0.0
Related Articles
Source Code
https://github.com/kiwifylabs/kiwi-foundation-config-di
Quick Start
var services = new ServiceCollection();
services.AddKiwiServices(configuration);
var provider = services.BuildServiceProvider();
That single call scans your assembly, loads all [ConfigService] config classes, and registers all [Service] classes - conditionally where specified.
Mental Model
Kiwi DI follows a simple model:
- Classes declare what they are - lifetime, interfaces, and construction rules, expressed as attributes on the class itself.
- Configuration declares when they are active - a config flag or named predicate controls whether a service enters the container.
-
The framework decides what gets wired -
AddKiwiServicesreads those declarations at startup and builds the container automatically.
Everything in this article maps back to one of these three responsibilities.
How AddKiwiServices Works
Before looking at individual attributes, it is worth understanding what AddKiwiServices actually does when you call it. Everything else flows from this.
services.AddKiwiServices(configuration);
Internally, this runs four deterministic phases:
Phase 0 (pre-registration):
Register any pre-loaded config instances passed via preLoadedConfigs
Phase 1 (config discovery):
Scan assembly for classes carrying both [ConfigSection] and [ConfigService]
For each: call LoadConfiguration<T>() and register the instance as a singleton
Phase 2 (temporary provider):
Build a temporary IServiceProvider
(All config singletons from Phase 1 are now resolvable)
Phase 3 (service discovery):
Scan assembly for classes carrying [Service]
For each: evaluate any conditions against the temporary provider
Register passing services into the real container
Done - call BuildServiceProvider() normally
The temporary provider built in Phase 2 exists only to evaluate conditions. It is never used for actual service resolution. It is discarded when AddKiwiServices returns. The final container is a standard IServiceProvider - Kiwi DI has no presence at runtime.
The Attributes
Kiwi DI uses four attributes. Each is placed on the class that it describes.
Attributes live in Kiwify.Kiwi.Platform.DependencyInjection.Attributes. Extension methods (AddKiwiServices, AddKiwiConfig, AddKiwiService) live in Kiwify.Kiwi.Platform.DependencyInjection.
using Kiwify.Kiwi.Platform.DependencyInjection; // AddKiwiServices and friends
using Kiwify.Kiwi.Platform.DependencyInjection.Attributes; // [Service], [ConfigService], [RegistersFor], [ConstructFrom]
| Attribute | Purpose |
|---|---|
[ConfigService] |
Marks a config class for auto-loading and singleton registration |
[Service(Lifetime)] |
Marks a service class for auto-registration |
[RegistersFor(typeof(T))] |
Expands a generic class into concrete closed-generic registrations |
[ConstructFrom(typeof(C), "Prop")] |
Injects scalar config property values as constructor arguments |
[ConfigService] - Auto-Loading Configuration
[ConfigService] comes from this library but works together with [ConfigSection] from Kiwi Config. A class carrying both is discovered during Phase 1 and loaded as a config singleton.
using Kiwify.Kiwi.Platform.Configuration.Attributes;
using Kiwify.Kiwi.Platform.DependencyInjection;
using Kiwify.Kiwi.Platform.DependencyInjection.Attributes;
[ConfigSection("database")] // from Kiwi Config - declares the config key
[ConfigService] // from Kiwi DI - opts into auto-loading
public class DatabaseConfig
{
[ConfigKey("host", "localhost")]
public string Host { get; private set; } = string.Empty;
[ConfigKey("port", 5432)]
public int Port { get; private set; }
[ConfigKey("name", Required = true)]
public string Name { get; private set; } = string.Empty;
}
After AddKiwiServices runs, DatabaseConfig is in the container as a singleton and can be injected normally:
[Service(ServiceLifetime.Scoped)]
public class OrderRepository
{
private readonly DatabaseConfig _db;
public OrderRepository(DatabaseConfig db) => _db = db;
}
What [ConfigSection] alone does not do
[ConfigSection] tells Kiwi Config how to load the class. It says nothing about DI. Without [ConfigService], a class is invisible to AddKiwiServices - it still works with configuration.LoadConfiguration<T>() but is not auto-loaded into the container.
This separation matters: you can have config classes that are only ever loaded manually (e.g. before the container is built) without them appearing as singletons in the container.
[Service] - Registering a Service
[Service] marks a class for automatic registration. The required argument is the service lifetime.
[Service(ServiceLifetime.Singleton)]
public class GreetingService : IGreetingService { ... }
[Service(ServiceLifetime.Scoped)]
public class OrderService : IOrderService { ... }
[Service(ServiceLifetime.Transient)]
public class EmailFormatter : IEmailFormatter { ... }
Interface resolution
By default, all non-System.* interfaces implemented by the class are registered automatically.
If a class implements no non-system interfaces, it is registered as its concrete type.
public interface IPaymentGateway { ... }
public interface IRetryable { ... }
[Service(ServiceLifetime.Singleton)]
public class StripeGateway : IPaymentGateway, IRetryable { ... }
// Registered as: IPaymentGateway and IRetryable
Registering as self too
RegisterAsSelf = true adds the concrete type registration alongside the interface registrations. Useful when code needs to inject the concrete type directly - for example, when calling methods not on the interface.
[Service(ServiceLifetime.Singleton, RegisterAsSelf = true)]
public class MetricsCollector : IMetricsCollector { ... }
// Registered as: IMetricsCollector AND MetricsCollector
Service lifetimes
| Lifetime | Created | Disposed |
|---|---|---|
Singleton |
Once per container | When the container is disposed |
Scoped |
Once per scope (per request in web apps) | When the scope is disposed |
Transient |
Every time it is resolved | When the scope that created them is disposed (if tracked by the container) |
Choose based on how much state the service holds and how long that state should live. Stateless services can be Singleton. Services that hold per-request state should be Scoped. Very lightweight, stateless objects can be Transient.
Multiple Implementations of the Same Interface
When two services implement the same interface and neither uses Key, both are registered:
[Service(ServiceLifetime.Singleton)]
public class ConsoleLogger : ILogger { ... }
[Service(ServiceLifetime.Singleton)]
public class FileLogger : ILogger { ... }
-
GetRequiredService<ILogger>()returns the last registration (determined by discovery order, which depends on assembly metadata and should not be relied upon). -
GetServices<ILogger>()returns all of them.
If you need deterministic resolution, use Key to give each implementation a distinct identity:
[Service(ServiceLifetime.Singleton, Key = "console")]
public class ConsoleLogger : ILogger { ... }
[Service(ServiceLifetime.Singleton, Key = "file")]
public class FileLogger : ILogger { ... }
var console = provider.GetRequiredKeyedService<ILogger>("console");
var file = provider.GetRequiredKeyedService<ILogger>("file");
Conditional Registration: Config-Key Conditions
A service can declare a condition under which it should be registered. The most common form checks a configuration value.
[Service(ServiceLifetime.Singleton, ConfigKey = "Features:Redis")]
public class RedisCache : ICache { ... }
This service is registered only when configuration["Features:Redis"] equals "true" (evaluated as a case-insensitive string comparison). The condition is evaluated in Phase 3, after all config singletons are loaded, using the temporary provider.
A different expected value:
[Service(ServiceLifetime.Singleton, ConfigKey = "Database:Type", ConfigValue = "postgres")]
public class PostgresRepository : IRepository { ... }
This registers only when configuration["Database:Type"] == "postgres".
appsettings.json:
{
"Features": { "Redis": "true" },
"Database": { "Type": "postgres" }
}
The Fallback Pattern: Negate = true
Negate = true inverts the condition. Combine it with the same condition on a second class to create a primary/fallback pair - exactly one of which is always registered.
// Primary: registered when Redis is enabled
[Service(ServiceLifetime.Singleton, ConfigKey = "Features:Redis")]
public class RedisCache : ICache { ... }
// Fallback: registered when Redis is NOT enabled
[Service(ServiceLifetime.Singleton, ConfigKey = "Features:Redis", Negate = true)]
public class InMemoryCache : ICache { ... }
With Features:Redis = "true": only RedisCache is in the container.
With Features:Redis = "false" (or absent): only InMemoryCache is in the container.
The swap costs one config change. No startup code changes. No recompilation.
This pattern works for any number of implementations across any condition. You can also nest conditions or use the same flag for multiple service pairs.
Named Conditions: Complex Logic
Config flag checks cover common cases, but some conditions go beyond a single config value. Maybe a service should only be registered in the production environment. Maybe it requires credentials that might not be present. Maybe it checks a combination of conditions.
Named conditions are registered as predicates and referenced by name in the attribute.
services.AddKiwiServices(configuration, options =>
{
options.AddCondition("IsProduction", sp =>
sp.GetRequiredService<IHostEnvironment>().IsProduction());
options.AddCondition("HasAzureCredentials", sp =>
{
var cfg = sp.GetRequiredService<IConfiguration>();
return !string.IsNullOrEmpty(cfg["Azure:ConnectionString"]);
});
options.AddCondition("HighTrafficMode", sp =>
{
var appConfig = sp.GetRequiredService<AppConfig>();
return appConfig.MaxConcurrency > 100;
});
});
The IServiceProvider passed to each predicate is the Phase 2 temporary provider - all config singletons are already available, so you can resolve them to inspect their values.
Referencing a named condition:
[Service(ServiceLifetime.Singleton, Condition = "IsProduction")]
public class ProductionMetrics : IMetrics { ... }
[Service(ServiceLifetime.Singleton, Condition = "IsProduction", Negate = true)]
public class NoOpMetrics : IMetrics { ... }
[Service(ServiceLifetime.Singleton, Condition = "HasAzureCredentials")]
public class AzureBlobStorage : IStorageService { ... }
[Service(ServiceLifetime.Scoped, Condition = "HighTrafficMode")]
public class BulkOrderProcessor : IOrderProcessor { ... }
If both ConfigKey and Condition are set on the same [Service], Condition takes precedence. In practice, use one or the other - setting both invites ambiguity.
Keyed Services
Keyed services let multiple implementations of the same interface coexist, each retrievable by a string key.
[Service(ServiceLifetime.Scoped, Key = "primary")]
public class PrimaryDatabase : IDatabase { ... }
[Service(ServiceLifetime.Scoped, Key = "replica")]
public class ReplicaDatabase : IDatabase { ... }
Resolution:
var primary = provider.GetRequiredKeyedService<IDatabase>("primary");
var replica = provider.GetRequiredKeyedService<IDatabase>("replica");
Keyed and conditional can be combined:
// Keyed primary/fallback - the key stays the same, the implementation swaps
[Service(ServiceLifetime.Scoped, Key = "cache", ConfigKey = "Features:Redis")]
public class RedisCache : ICache { ... }
[Service(ServiceLifetime.Scoped, Key = "cache", ConfigKey = "Features:Redis", Negate = true)]
public class InMemoryCache : ICache { ... }
Callers always resolve by "cache" key - which implementation they get depends on the config flag.
Generic Services: [RegistersFor]
When the same pattern applies to multiple concrete types - event handlers, message processors, pipeline stages, notification senders - writing a separate registration for each concrete instantiation is repetitive.
[RegistersFor] lets one generic class declaration produce multiple concrete closed-generic registrations. Each attribute is one registration.
[Service(ServiceLifetime.Singleton)]
[RegistersFor(typeof(OrderHandler))]
[RegistersFor(typeof(PaymentHandler), Key = "payment")]
[RegistersFor(typeof(AuditHandler), Key = "audit", ConfigKey = "Features:Audit")]
public class EventManager<THandler> where THandler : class
{
private readonly THandler _handler;
public EventManager(THandler handler) => _handler = handler;
public void Handle(object @event) { /* delegate to _handler */ }
}
This single class produces:
-
EventManager<OrderHandler>- singleton, no key, always registered -
EventManager<PaymentHandler>- singleton, keyed"payment", always registered -
EventManager<AuditHandler>- singleton, keyed"audit", only ifFeatures:Audit == "true"
Adding support for a new handler means adding one [RegistersFor] line. The startup code does not change.
Note: If a concrete type used in
[RegistersFor]is itself conditionally registered, apply the same condition to the[RegistersFor]attribute. Otherwise, the generic wrapper can be registered while the dependency it wraps is absent from the container, causing a resolution failure at runtime.// Good - condition on the generic registration matches the service it wraps [RegistersFor(typeof(SmsHandler), Key = "sms", ConfigKey = "Features:Sms")]
Per-registration options
Each [RegistersFor] attribute can override the lifetime from [Service] and specify its own key, condition, and negate:
| Property | Purpose |
|---|---|
concreteType |
The closed type argument (must not be a generic type definition) |
Key |
Keyed registration key |
Lifetime |
Override the parent [Service] lifetime for this registration |
ConfigKey |
Config-flag condition |
ConfigValue |
Expected value (default "true") |
Condition |
Named condition reference |
Negate |
Invert the condition |
Generic type constraints
If the class has a type parameter constraint (: class, : new(), : ISomeInterface), the concrete type provided to [RegistersFor] is validated against it. An invalid type throws InvalidOperationException at startup.
public class EventManager<THandler> where THandler : class, IEventHandler
{
...
}
[RegistersFor(typeof(string))] // throws - string does not satisfy IEventHandler constraint
[ConstructFrom] - Scalar Values from Config
Standard constructor injection resolves entire services. Some components need raw scalar values from configuration - a port number, a connection string, a timeout - not a reference to the full config object.
Without [ConstructFrom], the only way to provide these is a factory lambda in startup code:
// Startup code - disconnected from the class that needs these values
services.AddSingleton(sp =>
{
var cfg = sp.GetRequiredService<AppConfig>();
return new TcpServer(cfg.Host, cfg.Port);
});
[ConstructFrom] moves the extraction rule to the class itself:
[Service(ServiceLifetime.Singleton)]
[ConstructFrom(typeof(AppConfig), nameof(AppConfig.Host), nameof(AppConfig.Port))]
public class TcpServer
{
private readonly string _host;
private readonly int _port;
public TcpServer(string host, int port)
{
_host = host;
_port = port;
}
}
At resolve time, the framework:
- Resolves
AppConfigfrom the container. - Reads
AppConfig.HostandAppConfig.Portin the order listed. - Calls
new TcpServer(host, port).
The startup code stays untouched.
Single property from one source
[Service(ServiceLifetime.Singleton)]
[ConstructFrom(typeof(DatabaseConfig), nameof(DatabaseConfig.ConnectionString))]
public class ConnectionPool
{
public ConnectionPool(string connectionString) { ... }
}
Multiple properties from one source
Properties are passed to the constructor in the order they are listed in the attribute:
[Service(ServiceLifetime.Singleton)]
[ConstructFrom(typeof(DatabaseConfig),
nameof(DatabaseConfig.Host),
nameof(DatabaseConfig.Port),
nameof(DatabaseConfig.Name))]
public class DatabaseConnection
{
public DatabaseConnection(string host, int port, string dbName) { ... }
}
Multiple sources
Multiple [ConstructFrom] attributes are processed in declaration order. All properties from the first attribute become the first constructor arguments, then the second, and so on:
[Service(ServiceLifetime.Scoped)]
[ConstructFrom(typeof(DatabaseConfig), nameof(DatabaseConfig.ConnectionString))] // arg 0
[ConstructFrom(typeof(CacheConfig), nameof(CacheConfig.RedisEndpoint))] // arg 1
[ConstructFrom(typeof(AppConfig), nameof(AppConfig.InstanceId))] // arg 2
public class DataCoordinator
{
public DataCoordinator(string connStr, string redisEndpoint, string instanceId) { ... }
}
When to use vs regular injection
| Scenario | Approach |
|---|---|
| Service needs specific scalar values (port, timeout, connection string) | [ConstructFrom] |
| Service needs to call methods on the config object | Regular injection - inject the config class |
| Service needs both scalar config values and other services | Regular injection - let the service extract what it needs |
| Factory logic is too complex for property extraction | Manual services.Add*(sp => ...)
|
Constraint to know
Important: A class using
[ConstructFrom]cannot also use regular DI constructor injection. All constructor arguments must come from property extraction. If a class needs both scalar config values and injected services, inject the full config object instead.
Pre-Loaded Configs
Sometimes you need configuration values before the DI container is built - to configure the logging pipeline, set up middleware, or make decisions before AddKiwiServices is called.
Load the config manually and use it immediately:
// Load before the container exists
var appConfig = configuration.LoadConfiguration<AppConfig>();
// Use the values pre-container
SetupLogging(appConfig.LogLevel);
ConfigureListenAddress(appConfig.Port);
// Pass the already-loaded instance to AddKiwiServices
// It skips auto-loading for that type and registers the provided instance as the singleton
services.AddKiwiServices(configuration, preLoadedConfigs: appConfig);
Services that inject AppConfig receive the same instance loaded before the container was built.
Multiple pre-loaded configs:
var appConfig = configuration.LoadConfiguration<AppConfig>();
var dbConfig = configuration.LoadConfiguration<DatabaseConfig>();
services.AddKiwiServices(configuration,
preLoadedConfigs: new object[] { appConfig, dbConfig });
All other [ConfigService] types in the assembly are still auto-loaded normally. Only the pre-loaded types are skipped (since they are already registered).
Manual Registration
For selective control without a full assembly scan, use the single-type registration methods:
// Register a single config class
services.AddKiwiConfig<DatabaseConfig>(configuration);
// Register a single service
services.AddKiwiService<OrderService>();
AddKiwiConfig<T> validates [ConfigSection] and [ConfigService], calls LoadConfiguration<T>, and registers the result as a singleton.
AddKiwiService<T> validates [Service], resolves constructor dependencies recursively, and registers the service.
These are useful in tests, in tools that do not scan entire assemblies, or when integrating Kiwi DI incrementally into an existing codebase.
Using Kiwi Config and Kiwi DI Together
The two libraries are designed to work together, but each can be used independently. When used together, AddKiwiServices handles the entire coordination:
// Program.cs
var configuration = new ConfigurationBuilder()
.AddJsonFile("appsettings.json")
.AddEnvironmentVariables()
.Build();
var services = new ServiceCollection();
services.AddKiwiServices(configuration);
var provider = services.BuildServiceProvider();
Config classes annotated with both [ConfigSection] and [ConfigService] are loaded from configuration and become injectable singletons. Service classes annotated with [Service] can inject those config singletons through normal constructor injection because the config singletons are registered first (Phase 1 before Phase 3).
[ConfigSection("app")]
[ConfigService]
public class AppConfig
{
[ConfigKey("name", "MyApp")]
public string Name { get; private set; } = string.Empty;
[ConfigKey("port", 5000)]
public int Port { get; private set; }
}
[Service(ServiceLifetime.Singleton)]
public class AppServer : IAppServer
{
private readonly AppConfig _config;
public AppServer(AppConfig config) => _config = config;
public void Start() => Console.WriteLine($"Starting {_config.Name} on port {_config.Port}");
}
services.AddKiwiServices(configuration);
var provider = services.BuildServiceProvider();
provider.GetRequiredService<IAppServer>().Start();
// Starting MyApp on port 5000
No factory, no registration block, no startup code changes when new services are added.
Complete End-to-End Example
A realistic notification system demonstrating config loading, conditional services, the fallback pattern, generic expansion, and [ConstructFrom] - all wired together with a single AddKiwiServices call.
Configuration
appsettings.json (all configuration values are treated as strings):
{
"notifications": {
"smtpHost": "mail.example.com",
"retryCount": 3
},
"Features": {
"Sms": "false",
"Audit": "true"
}
}
Config class
[ConfigSection("notifications")]
[ConfigService]
public class NotificationConfig
{
[ConfigKey("smtpHost", Required = true)]
public string SmtpHost { get; private set; } = string.Empty;
[ConfigKey("retryCount", 3)]
public int RetryCount { get; private set; }
}
Conditional services - SMS or email based on config
public interface INotificationSender
{
void Send(string to, string body);
}
// Active only when Features:Sms = "true"
[Service(ServiceLifetime.Scoped, ConfigKey = "Features:Sms")]
public class SmsNotificationSender : INotificationSender
{
public void Send(string to, string body)
=> Console.WriteLine($"[SMS] {to}: {body}");
}
// Active when Features:Sms is not "true" (the fallback)
[Service(ServiceLifetime.Scoped, ConfigKey = "Features:Sms", Negate = true)]
public class EmailNotificationSender : INotificationSender
{
private readonly string _smtpHost;
public EmailNotificationSender(NotificationConfig config)
=> _smtpHost = config.SmtpHost;
public void Send(string to, string body)
=> Console.WriteLine($"[Email via {_smtpHost}] {to}: {body}");
}
With Features:Sms = "false", only EmailNotificationSender enters the container.
Retry handler - scalar value from config via [ConstructFrom]
[Service(ServiceLifetime.Singleton)]
[ConstructFrom(typeof(NotificationConfig), nameof(NotificationConfig.RetryCount))]
public class RetryHandler
{
private readonly int _retryCount;
public RetryHandler(int retryCount) => _retryCount = retryCount;
public void Execute(Action action)
{
for (var i = 0; i <= _retryCount; i++)
{
try { action(); return; }
catch when (i < _retryCount) { }
}
}
}
RetryHandler receives the RetryCount = 3 value extracted directly from NotificationConfig - no reference to the full config object needed.
Generic processor - one class, two registrations
[Service(ServiceLifetime.Scoped)]
[RegistersFor(typeof(SmsNotificationSender), Key = "sms", ConfigKey = "Features:Sms")]
[RegistersFor(typeof(EmailNotificationSender), Key = "email", ConfigKey = "Features:Sms", Negate = true)]
public class NotificationProcessor<TSender>
where TSender : class, INotificationSender
{
private readonly TSender _sender;
private readonly RetryHandler _retry;
public NotificationProcessor(TSender sender, RetryHandler retry)
{
_sender = sender;
_retry = retry;
}
public void Notify(string to, string body)
=> _retry.Execute(() => _sender.Send(to, body));
}
This produces:
-
NotificationProcessor<SmsNotificationSender>registered under key"sms" -
NotificationProcessor<EmailNotificationSender>registered under key"email"
Audit service - named condition example
services.AddKiwiServices(configuration, options =>
{
options.AddCondition("AuditEnabled", sp =>
{
var cfg = sp.GetRequiredService<IConfiguration>();
return cfg["Features:Audit"] == "true";
});
});
[Service(ServiceLifetime.Singleton, Condition = "AuditEnabled")]
public class AuditLogger : IAuditLogger
{
public void Log(string action) => Console.WriteLine($"[AUDIT] {action}");
}
[Service(ServiceLifetime.Singleton, Condition = "AuditEnabled", Negate = true)]
public class NoOpAuditLogger : IAuditLogger
{
public void Log(string action) { }
}
Startup and usage
services.AddKiwiServices(configuration, options =>
{
options.AddCondition("AuditEnabled", sp =>
sp.GetRequiredService<IConfiguration>()["Features:Audit"] == "true");
});
var provider = services.BuildServiceProvider();
// With Features:Sms = "false" and Features:Audit = "true":
var emailProcessor = provider.GetRequiredKeyedService<NotificationProcessor<EmailNotificationSender>>("email");
emailProcessor.Notify("user@example.com", "Your order has shipped.");
// [Email via mail.example.com] user@example.com: Your order has shipped.
var audit = provider.GetRequiredService<IAuditLogger>();
audit.Log("Notification sent");
// [AUDIT] Notification sent
What AddKiwiServices did:
- Loaded
NotificationConfig→ registered as singleton (SmtpHost =mail.example.com, RetryCount =3). - Evaluated
Features:Sms = "false"→ registeredEmailNotificationSender; skippedSmsNotificationSender. - Registered
RetryHandleras singleton - extractedRetryCount = 3fromNotificationConfig. - Evaluated
Features:Sms = "false"→ registeredNotificationProcessor<EmailNotificationSender>(key"email"); skippedNotificationProcessor<SmsNotificationSender>(key"sms") because its condition mirrors that ofSmsNotificationSender. - Evaluated named condition
AuditEnabled→ registeredAuditLogger; skippedNoOpAuditLogger.
To switch to SMS: set Features:Sms = "true". To disable auditing: set Features:Audit = "false". The swap is a config change. No startup edits. No recompilation.
Performance and Startup Cost
Kiwi DI shifts all discovery and wiring work to startup so runtime remains identical to standard .NET DI. All reflection and assembly scanning happens exactly once - at startup - and contributes nothing to runtime latency.
| Step | When | Notes |
|---|---|---|
| Assembly type scanning | Startup | Proportional to assembly size; typically well under 100ms for a few hundred types (depending on environment) |
| Config loading (per class) | Startup | Reflection-based; negligible for typical class sizes |
| Temporary provider build | Startup | One extra BuildServiceProvider() call |
| Condition evaluation | Startup | One predicate call per conditional service |
| Service resolution at runtime | Runtime | Standard .NET DI - no Kiwi reflection involved |
If assembly scanning of very large assemblies causes noticeable startup delay, pass explicit assemblies to narrow the scope:
services.AddKiwiServices(configuration, assemblies: new[]
{
typeof(OrderService).Assembly,
typeof(PaymentService).Assembly
});
When to Use Kiwi DI
Use it when:
- Your application has a meaningful number of services and startup registration has become a maintenance concern.
- You want config-driven feature flags to control service activation without touching startup code.
- You use generic processor or handler patterns where one class is instantiated for many concrete types.
- You value co-location: the DI role of a class is visible at the class, not buried in startup code.
Consider alternatives when:
- Your application is small (roughly fewer than 20-30 services). The attribute indirection adds complexity that manual registration does not, and the savings are minimal.
- You need fine-grained control over registration order or multi-step factory chains that do not map to attribute declarations.
- Your team prefers fully explicit, line-by-line startup code as a policy.
- You are debugging complex DI graph issues and want the simplest possible dependency graph.
Feature Summary
| Goal | How |
|---|---|
| Auto-load a config class into DI |
[ConfigSection] + [ConfigService] on the class |
| Register a service |
[Service(Lifetime)] on the class |
| Also register as concrete type | RegisterAsSelf = true |
| Register only when a config flag is set | ConfigKey = "Section:Key" |
| Register only when a flag is NOT set |
ConfigKey = "..." + Negate = true
|
| Register based on complex logic |
options.AddCondition("name", sp => ...) + Condition = "name"
|
| Fallback when condition fails |
Negate = true (same condition) |
| Keyed registration | Key = "myKey" |
| Resolve keyed service | provider.GetRequiredKeyedService<T>("myKey") |
| One generic class → many registrations |
[RegistersFor(typeof(T))] (multiple attributes) |
| Scalar config values as constructor args | [ConstructFrom(typeof(Config), "Prop1", "Prop2")] |
| Load config before container is built |
configuration.LoadConfiguration<T>() then preLoadedConfigs: instance
|
| Register one class manually |
services.AddKiwiConfig<T>() or services.AddKiwiService<T>()
|
| Named conditions from external logic | options.AddCondition(name, sp => bool) |
Top comments (0)