We make Data fit for C#.

From local files to cloud scale — LINQ all the way down.
Let IntelliSense and the compiler do the work.

- df.filter(pl.col("ammount") > 1000)   # Typo? Runtime error.
+ .Where(o => o.Amount > 1000)          // Typo? Won't compile. ✓

# Install via NuGet
dotnet add package DataFlow.Net --version 1.2.1

1. Sound Familiar?
2. Three Simple Rules
3. Everything is a Stream
4. Quick Start
5. Documentation
6. Community & Support

.NET developers know the story — You write a clean, type-safe data processor in C# — It works perfectly on your dev machine — Then reality hits:

1. The Data Grows:

   • 10 MB: List<T> works fine.
   • 10 GB: OutOfMemoryException. You rewrite using StreamReader.
   • 10 TB: You abandon C# for Spark/SQL. You lose type safety and duplicate logic.

2. The Logic Tangles:

   • New requirements mean new if/else branches.
   • You loop over the same data 5 times to handle 5 different cases.
   • The code becomes spaghetti, and the data lifecycle becomes a black box.

3. The Source Fragments:

   • Today it's a CSV file. Tomorrow it's a REST API. Next week it's a Kafka Stream.
   • For each source, you write different adapter code.
   • You end up with a "Code Salad": mixed abstractions, different error handling, and no reuse.

DataFlow.NET was built to stop this cycle:

• ✅ Unified API — Same code for CSV, JSON, Kafka, Spark
• ✅ Constant memory — Stream billions of rows without OutOfMemoryException (see benchmarks)
• ✅ No spaghetti — Declarative Cases pattern replaces nested if/else
• ✅ Pure C# — LINQ all the way down

DataFlow.NET is more than a framework — it defines a pattern to process data.

graph LR
    S[**S**ink] --> U[**U**nify]
    U --> P[**P**rocess]
    P --> R[**R**oute]
    R --> A[**A**pply]
    
    style S fill:#f9f,stroke:#333,stroke-width:2px
    style A fill:#bbf,stroke:#333,stroke-width:2px

We call this the SUPRA pattern — the name comes from gathering the first letter of each stage: Sink, Unify, Process, Route, Apply.

To achieve the SUPRA pattern, you'll have to follow these rules:

1. Sink First — Buffer and normalize at the edge, never in the middle.
2. Flow Lazy — Items stream one by one. Constant memory.
3. Route Declaratively — No more if/else spaghetti.

DataFlow.NET provides all the ready-to-use blocks to natively apply these rules.

DataFlow.NET provides tools to abstract the source of data from the processing. Use these to make every data source an IAsyncEnumerable<T> stream — the essence of the "Unified API" — same LINQ operators, same processing logic, regardless of origin.

See Integration Patterns Guide →

Already using Entity Framework Core? DataFlow.NET plugs right in:

// EF Core — Native support
await dbContext.Orders.AsAsyncEnumerable()
    .Where(o => o.Amount > 100)
    .WriteCsv("orders.csv");

• ✅ EF Core handles database access
• ✅ DataFlow.NET handles processing logic
• ✅ Works with SQL Server, PostgreSQL, MySQL, SQLite

Need to integrate REST APIs or message queues? Use polling and buffering:

// REST API — Poll and flatten
var orders = (() => httpClient.GetFromJsonAsync<Order[]>("/api/orders"))
    .Poll(TimeSpan.FromSeconds(5), token)
    .SelectMany(batch => batch.ToAsyncEnumerable());

// Kafka/WebSocket — Wrap in async iterator + buffer
var kafkaStream = ConsumeKafka(token).WithBoundedBuffer(1024);

DataFlow.NET provides high-performance file readers: no Reflection on the hot path; expression trees are compiled once and cached.

• 4x faster than standard reflection-based creation (benchmark results →)
• Zero allocation overhead — same 48 bytes as native new() instantiation
• Handles CSV, JSON, and YAML files generically.

We carefully crafted an intuitive, fully-featured readers API with advanced error handling — all while streaming row-by-row.

Materialization Quick Reference → | Data Reading Guide →

DataFlow.NET implements additional LINQ extensions to make every data loop composable—even side-effect loops.

• Independent implementation — Re-implemented IAsyncEnumerable methods without depending on System.Linq.Async
• Clear terminal vs non-terminal separation — Terminal methods (Do(), Display()) force execution; non-terminal methods (ForEach(), Select(), Where()) stay lazy

See Extension Methods API Reference →

We've extended standard LINQ with custom operators for declarative branching. Using Cases, SelectCase, and ForEachCase, you can replace complex nested if/else blocks with an optimized, single-pass dispatch tree — while remaining fully composable.

See Cases Pattern Guide →

This is the "U" (Unify) step of the SUPRA pattern — "absorb many sources into one stream."

var unifiedStream = new UnifiedStream<Log>()
    .Unify(fileLogs, "archive")
    .Unify(apiLogs, "live")
    .Unify(dbLogs, "backup");
// Result: A single IAsyncEnumerable<Log> you can query

See Stream Merging Guide →

Insert observation points anywhere in your pipeline without changing data flow. Because Spy() is fully composable, you can add or remove traces by simply commenting a line — no code rewriting required.

await data
    .Where(...)
    .Spy("After filtering")       // 👈 See items flow through
    .Select(...)
    .Spy("After transformation")
    .ForEach(...)                 // 👈 Side-effect iteration, still composable
    .Do();                        // 👈 Force execution (no output needed)

⚠️ Note: Due to lazy execution, output from multiple Spy() calls appears interleaved (item-by-item), not grouped by stage. This preserves the streaming nature of the pipeline.

Need to parallelize CPU-intensive or I/O-bound work? DataFlow.NET provides parallel counterparts that work just like their sequential equivalents — still lazy, still composable:

// Parallel sync processing
await data.AsParallel()
    .Select(item => ExpensiveCompute(item))
    .ForEach(item => WriteToDb(item))
    .Do();

// Parallel async processing
await asyncStream.AsParallel()
    .WithMaxConcurrency(8)
    .Select(async item => await FetchAsync(item))
    .Do();

See ParallelAsyncQuery API Reference → | Parallel Processing Guide → | Extension Methods →

If you hit the limit of local computing power, DataFlow.NET lets you seamlessly scale to the cloud with LINQ-to-Spark & Snowflake. Your C# lambda expressions are decompiled at runtime and translated into native Spark/SQL execution plans.

• ✅ No data transfer to client
• ✅ Execution happens on the cluster
• ✅ Full type safety

LINQ-to-Spark Guide → | LINQ-to-Snowflake Guide →

• .NET 8.0 SDK or later

Via NuGet (Recommended):

dotnet add package DataFlow.Net --version 1.2.1

Or clone the repository:

git clone https://github.com/improveTheWorld/DataFlow.NET
cd DataFlow.NET

dotnet run --project DataFlow.Test.UsageExamples/DataFlow.App.UsageExamples.csproj

Or open the full solution in Visual Studio 2022:

DataFlow.Net.sln

using DataFlow;

// A complete, memory-efficient pipeline in 10 lines
await Read.Csv<Order>("orders.csv")
    .Cases(
        o => o.Amount > 1000, 
        o => o.CustomerType == "VIP"
    )
    .SelectCase(
        highValue => ProcessHighValue(highValue),
        vip => ProcessVip(vip)
    )
    .AllCases()
    .WriteJson("output.json");

Your business rule is: "Flag high-value transactions from international customers."

// 1. DEVELOPMENT: Read from a local CSV file
await Read.Csv<Order>("orders.csv")
    .Cases(o => o.Amount > 10000, o => o.IsInternational) // 👈 Your Logic
    .SelectCase(...) 
    .AllCases()
    .WriteCsv("output.csv");

// 2. PRODUCTION: Merge multiple async streams
await new UnifiedStream<Order>()
    .Unify(ordersApi, "api")
    .Unify(ordersDb, "db")
    .Cases(o => o.Amount > 10000, o => o.IsInternational) // 👈 SAME Logic
    .SelectCase(...)
    .AllCases()
    .WriteJson("output.json");

// 3. CLOUD: Query Snowflake Data Warehouse
// Filters and aggregations execute on the server
using var sfContext = Snowflake.Connect(account, user, password, database, warehouse);
await sfContext.Read.Table<Order>("orders")
    .Where(o => o.Year == 2024)
    .Cases(o => o.Amount > 10000, o => o.IsInternational) // 👈 SAME Logic
    .SelectCase(...)
    .ToList();

// 4. SCALE: Run on Apache Spark (Petabyte Scale)
// Translates your C# Expression Tree to native Spark orchestration
using var sparkContext = Spark.Connect("spark://master:7077", "MyApp");
sparkContext.Read.Table<Order>("sales.orders")
    .Where(o => o.Amount > 10000)
    .Cases(o => o.Amount > 50000, o => o.IsInternational) // 👈 SAME Logic
    .SelectCase(...)
    .AllCases()
    .WriteParquet("s3://data/output");

• Issues: GitHub Issues
• Discord: Join the Community
• Email: tecnet.paris@gmail.com

DataFlow.NET — Sink the chaos. Let the rest flow pure. 🚀