dotnet-sep

Sep for .NET separated values

Trigger On

• delimited data needs are performance-sensitive and allocation-aware
• project needs explicit control over separator inference, escaping, trimming, and header behavior
• reading/writing large or long-lived file pipelines in ML, ETL, or analytics workloads
• startup/perf tests require AOT/trimming-friendly CSV/TSV processing

Install

• NuGet:
  • dotnet add package Sep
  • dotnet add package Sep --version <version>
• XML package reference:
  • <PackageReference Include="Sep" Version="x.y.z" />
• Verify baseline support by checking the package page:
  • NuGet: Sep
• Source:
  • GitHub: nietras/Sep

Workflow

flowchart LR
  A[Input source: file/text/stream] --> B[Sep.Reader or Sep.New(...).Reader]
  B --> C[SepReaderOptions]
  C --> D[Rows -> Cols -> Span/Parse]
  D --> E[Transform and validate]
  E --> F[SepWriter via SepWriterOptions]
  F --> G[To file/text output]

1. Decide schema shape
   • header present or no header
   • separator known (;, ,, tab, custom) or infer from first row
   • row/column quoting rules
2. Build reader with Sep.Reader(...) and explicit options only where needed:
   • Sep.Reader() for inferred separator from header-like first row
   • Sep.New(',').Reader(...) for explicit separator mode
   • Sep.Reader(o => o with { HasHeader = false }) if header is absent
3. Read rows and map columns as ReadOnlySpan<char> first, convert only when needed.
4. For output, use reader.Spec.Writer() when you need the same separator/culture as input.
5. Control writer behavior with Sep.Writer(...) and SepWriterOptions (WriteHeader, Escape, DisableColCountCheck).
6. Add async only where it brings value and your runtime is C# 13 / .NET 9+ for await foreach over async reader rows.
7. Use ParallelEnumerate for CPU-heavy transformations only after benchmarking single-threaded baseline.

Install and read patterns