StreamScheme.FSharp 0.2.2

StreamScheme

Fast, typed, streaming read and write of tabular data in xlsx format. Nothing else.

Requires .NET 10+.

Quick start

C#

var handler = Xlsx.CreateHandler();
await handler.WriteAsync(stream, rows);

foreach (var row in handler.Read(stream))
{
    // each cell is a FieldValue: Text, Number, Date, Boolean, or Empty
}

F#

rows |> Xlsx.writeAsync stream |> Async.RunSynchronously

Xlsx.read stream
|> Seq.iter handleDataRow

Each cell is a FieldValue — five types: Text, Number, Date, Boolean, Empty.

See C# examples for manual mapping, typed object writing, and reading with pattern matching.

See F# examples for idiomatic writing and reading with pattern matching.

Is StreamScheme for you?

If you need presentation, use a full Excel library.

Installation

C#:

dotnet add package StreamScheme

F#:

dotnet add package StreamScheme.FSharp

String write modes

StreamScheme lets you control how repeated text values are stored in the xlsx output:

• Off — inline every string. Fastest, no overhead. Best when values are mostly unique.
• Always — deduplicate all strings via a shared strings table. Smaller files when few distinct values repeat across many cells.
• Windowed(n) — deduplicate within a sliding window of n rows. Bounded memory, good for mixed data where some columns repeat and others don't.

Benchmarks

100,000 rows. Ratios are relative to StreamScheme (baseline = 1.00).

Writing — unique strings (10 columns, XML-escapable characters)

No repeated data — shared strings cannot help here.

Writing — sparse categories (20 columns, 70% empty, verbose status strings)

Few distinct values repeated across many cells — shared strings deduplicate effectively.

Writing — mixed data (5 unique strings + 5 repeated category enums)

Half unique, half repeated — windowed shared strings reduces file size for the repeated columns.

Reading — mixed types (6 columns, 2.04 MB file)

Notes on allocation

Allocation numbers reflect the mapping layer, not total memory — the source data must live somewhere regardless. Both benchmarks allocate a new array per row. SpreadCheetah supports an imperative API where a single DataCell[] is reused across rows, which would reduce mapping layer allocation to near zero.

StreamScheme allocates FieldValue records per cell — short-lived Gen0 objects, collected quickly. This is a deliberate tradeoff: the IEnumerable<IEnumerable<FieldValue>> API is streaming and composable (LINQ-friendly) at the cost of Gen0 churn.

Acknowledgments

StreamScheme's date format detection code is adapted from MiniExcel (Apache 2.0), which credits ExcelNumberFormat (MIT) by andersnm.

SpreadCheetah (MIT) served as inspiration and the primary performance comparison baseline.

Built by Immersus Machina