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

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

csharpCopy code
using var reader = Sep.Reader(o => o with
{
    HasHeader = true,
    Unescape = true,
    Trim = SepTrim.Both
}).FromText(data);

foreach (var row in reader)
{
    var id = row["Id"].Parse<int>();
    var name = row[1].ToString();
    // process row
}

Write patterns

csharpCopy code
using var reader = Sep.Reader().FromFile("input.csv");
using var writer = reader.Spec.Writer().ToFile("output.csv");

foreach (var row in reader)
{
    using var writeRow = writer.NewRow(row);
    writeRow["Amount"].Format(row["Amount"].Parse<double>() * 1.2);
}

Async reading and writing

csharpCopy code
var text = "A;B\n1;hello\n";

using var reader = await Sep.Reader().FromTextAsync(text);
await using var writer = reader.Spec.Writer().ToText();

await foreach (var row in reader)
{
    await using var writeRow = writer.NewRow(row);
    var normalized = row["B"].ToString().ToUpperInvariant();
    writeRow["B"].Set(normalized);
}

Common configuration patterns

• Header-driven read
  • default HasHeader = true
  • query by name: row["ColName"]
• Headerless pipelines
  • HasHeader = false
  • use index-based access: row[0], row[1]
• Round-trip output
  • start writer with reader.Spec.Writer() to preserve inference and formatting contract
• Speed-first processing
  • keep default buffer + culture unless profiling proves a need to tune

Best practices

• Parse to primitive types with Parse<T> in hot paths to avoid extra allocations.
• Keep ToString/format conversions at the edge (presentational layers), not in inner loops.
• Prefer Unescape, Trim, and DisableQuotesParsing settings deliberately and test with realistic samples.
• For large transforms, isolate heavy CPU work after enumeration and then apply ParallelEnumerate where appropriate.

Limitations to check before production

• SepReader.Row and SepWriter.Row are ref structs:
  • avoid patterns that store rows beyond immediate scope
  • materialize if you truly need random async/LINQ-style buffering
• SepReader row iteration is row-by-row by design; it is intentionally not the same as a classic collection model.

Deliver

• installation and usage guide that is ready to copy into a .NET repo
• practical reader/writer configuration patterns
• clear notes on defaults, tradeoffs, and constraints

Validate

• dotnet add package Sep installs correctly and project compiles
• one file-read sample and one file-write sample execute successfully
• header/no-header and explicit-separator cases are covered
• at least one validation sample for quoting/unescaping or async path exists if required by task

Load References

• references/overview.md - official links and practical decision notes.