ValueTask et réduction des allocations

Le problème d’allocation avec Task

Qu’est-ce que ValueTask ?

// Task<T>: always allocates, even when result is cached
public async Task<string> GetCached_Task() =>
    _cache.TryGetValue("key", out var v) ? v : await FetchAsync();

// ValueTask<T>: zero allocation when value is cached
public ValueTask<string> GetCached_ValueTask() =>
    _cache.TryGetValue("key", out var v)
        ? ValueTask.FromResult(v)  // no allocation
        : new ValueTask<string>(FetchAsync()); // allocation only when needed

Renvoyer ValueTask depuis une interface

public interface IProductRepository
{
    // ValueTask: often completes synchronously from cache
    ValueTask<Product?> GetByIdAsync(int id);

    // Task: always async (DB query)
    Task<List<Product>> SearchAsync(string query);
}

// Consumption is identical to Task:
var product = await _repo.GetByIdAsync(42);
var results = await _repo.SearchAsync("keyboard");

ValueTask avec IValueTaskSource

// In most cases, use the simple ValueTask approach.
// IValueTaskSource is for library authors needing zero-alloc async at extreme scale.

// Example: System.IO.Pipelines uses this internally
var result = await pipe.Reader.ReadAsync(ct);
// ReadAsync returns ValueTask<ReadResult> backed by IValueTaskSource
// -- zero allocation for every read even under high throughput

Contraintes de ValueTask : un seul await

var vt = repo.GetByIdAsync(42);

// BAD: await multiple times
var a = await vt; // OK
var b = await vt; // UNDEFINED BEHAVIOR — don't do this!

// BAD: two concurrent awaits
var t1 = vt.AsTask(); // converts to Task (safe to share)
var t2 = vt.AsTask(); // second AsTask() on same ValueTask — WRONG

// GOOD: if you need multiple awaits, convert to Task
var task = repo.GetByIdAsync(42).AsTask();
var r1 = await task;
var r2 = await task; // OK — Task can be awaited multiple times

Guide de décision : ValueTask<T> ou Task<T>

// USE ValueTask when:
// 1. The method often completes synchronously (cache hit)
// 2. The method is on a hot path called millions of times/sec
// 3. You're writing a library with high-throughput I/O

// USE Task when:
// 1. The result is always async
// 2. The method might be awaited multiple times
// 3. You use Task.WhenAll / Task.WhenAny
// 4. You store the task in a variable and share it

ValueTask non générique

public interface ICache
{
    // Often a no-op when value exists — ValueTask avoids allocation
    ValueTask SetAsync(string key, string value);
    ValueTask InvalidateAsync(string key);
}

public class InMemoryCache : ICache
{
    private readonly Dictionary<string, string> _data = new();

    public ValueTask SetAsync(string key, string value)
    {
        _data[key] = value;
        return ValueTask.CompletedTask; // no allocation
    }
}

Mesurer les allocations avec BenchmarkDotNet

[MemoryDiagnoser]
public class CacheBenchmark
{
    private readonly CachedRepo _repo = new();

    [Benchmark]
    public async Task<Product?> Task_GetById()
        => await _repo.GetByIdTask(1);

    [Benchmark]
    public async ValueTask<Product?> ValueTask_GetById()
        => await _repo.GetByIdValueTask(1);
}

// Run: dotnet run -c Release
// Compare Gen0 allocations column — ValueTask should show 0B on cache hit

ValueTask dans SocketsHttpHandler

// HttpClient uses ValueTask-based sockets internally.
// When using HttpClient.GetAsync(), the response is awaited with Task<HttpResponseMessage>.
// For streaming large bodies, use:
using var response = await client.GetAsync(url, HttpCompletionOption.ResponseHeadersRead);
await using var stream = await response.Content.ReadAsStreamAsync();
// Stream reads internally use ValueTask<int> for zero-allocation I/O

Cas réel : service de cache sur un chemin d’exécution fréquent

public class ProductCache
{
    private readonly Dictionary<int, Product> _cache = new();
    private readonly IProductRepository _repo;

    public ProductCache(IProductRepository repo) => _repo = repo;

    public ValueTask<Product?> GetAsync(int id)
    {
        // Cache hit: no allocation
        if (_cache.TryGetValue(id, out var cached))
            return ValueTask.FromResult<Product?>(cached);

        // Cache miss: async fetch
        return new ValueTask<Product?>(FetchAndCacheAsync(id));
    }

    private async Task<Product?> FetchAndCacheAsync(int id)
    {
        var product = await _repo.GetByIdAsync(id);
        if (product is not null) _cache[id] = product;
        return product;
    }
}

Vérification rapide

Récapitulatif : ValueTask et prévention des allocations