Overview and supported formats

PocketCsvReader reads structured text through a forward-only data reader. Start with the core CSV package and add the extension for your input format. Schema and custom parser settings control value conversion; output APIs let you consume fields, objects, arrays, dictionaries, or tables.

Start here

  1. Install the package for your format.
  2. Read your first CSV, or choose a format below.
  3. Choose an output model and review stream ownership.

Supported inputs

Input Guide Package
CSV, TSV, other delimiters CSV and TSV PocketCsvReader
JSON objects/arrays and opt-in JSONC JSON and JSONC PocketCsvReader.Json
Independently framed JSON values NDJSON PocketCsvReader.Ndjson
Character-position records Fixed-width PocketCsvReader.FixedWidth
Labeled tab-separated and key=value records LTSV and logfmt PocketCsvReader.KeyValue
Common, Combined, and W3C logs Web logs PocketCsvReader.WebLogs
CEF and LEEF events Security logs PocketCsvReader.SecurityLogs

JSON and NDJSON support property projection and stable field layout for object records. Logfmt and LTSV support column projection with aliases and stable field layout. See JSON and NDJSON projection and logfmt and LTSV projection.

Inputs that produce Arrow batches directly

Input Guide Package
ClickHouse JSONColumns, JSONColumnsWithMetadata, and Arrow Integration JSON Columnar JSON PocketCsvReader.Arrow.ColumnarJson
Textual NetCDF datasets (.cdl) NetCDF CDL PocketCsvReader.Arrow.NetCdf

Prepare and interpret input

If the CSV structure is unknown, infer a schema and dialect, inspect diagnostics, then accept or adjust the result. Inference reads a sample; it is a discovery step before explicit configuration.

Profiles and builders combine syntax, schema, resource, parser, and optimization settings. Null and missing values and error handling differ by format and output API.

Boundaries and limitations

The readers process files and streams; they are not writers or network log receivers. Streaming avoids loading the complete input in ordinary row-reading paths, but memory also depends on the largest record, selected output model, and format. ToDataTable materializes all rows. JSON record-path selection captures a complete document before selecting rows; ordinary top-level array reading streams its elements.

See parser optimizations for the input contracts and performance considerations for stable field layout and UTF-8 CSV input.

Optional integrations expose Apache Arrow batches and Chrononuensis temporal types. Package support and framework requirements are listed in installation.


Installation and package selection