CSV can expose several inputs as one forward-only CsvBatchDataReader. Supply filenames, already-open streams, or functions that open streams. Stream factories avoid opening every file up front.
Example with stream factories
using System.Text;
using PocketCsvReader.Configuration;
var csv = new CsvReaderBuilder()
.WithDialect(d => d.WithHeader().WithLineTerminator("\n"))
.WithRowNumber()
.Build();
Func<Stream>[] inputs =
[
() => new MemoryStream(Encoding.UTF8.GetBytes("Name\nAda\n")),
() => new MemoryStream(Encoding.UTF8.GetBytes("Name\nGrace\n"))
];
using var reader = csv.ToDataReader(inputs);
while (reader.Read())
Console.WriteLine($"{reader.GetInt32(0)}: {reader.GetString(1)}");
1: Ada
2: Grace
For files, call csv.ToDataReader(new[] { "part1.csv", "part2.csv" }). ToDbDataReader has corresponding batch overloads; use ReadAsync(token) for async advancement. Factories are invoked synchronously when moving to an input; they are not asynchronous stream factories.
Headers and schema
By default, each file has its own header (HeaderRepeat = true). Use WithHeaderRepeat(false) when only the first file supplies a header; the batch carries those names forward. All inputs use the same profile. Keep column order, conversion schema, and header expectations compatible across files.
Ownership
Batch construction opens its first input. Advancing closes the previous stream; disposing closes the current stream and releases the batch enumerator. The batch owns provided streams and streams returned by factories. Prefer factories for many files and dispose a batch even if no records are read. Do not reuse a consumed stream sequence without resetting/reopening it.
Generated row numbers
WithRowNumber() prepends an int column named rowNumber; WithRowNumber("recordId") chooses its name, and WithoutRowNumber() disables it. The last call wins. Null, empty, or whitespace names are rejected. A name exactly colliding with a source column throws InvalidDataException.
Numbers start at one for each reader and continue across batch inputs. Headers, comments, and records skipped by malformed-data recovery do not consume a generated row number. Reader filters run after numbering, so filtered-out data records consume a number and leave gaps. Multiline CSV records receive one number. Overflow beyond Int32 throws.
Constructor mapping requires an additional first parameter for the generated value; custom CSV span mappers receive its field first. Tables and parsed row outputs expose an integer; string arrays expose a decimal string. Source schema indexes remain indexes of the original columns. Generated row numbers differ from source positions, which identify source records and offsets and can count skipped malformed records differently.
