Overview

This project contains Jackson extension component for reading and writing CSV encoded data, either as "raw" data (sequence of String arrays), or via data binding to/from Java Objects (POJOs).

Project is licensed under Apache License 2.0.

Status

With version 2.0.0, implementation is complete and ready to be used for "real work". Unit tests cover basic usage patterns; and there is even a small performance benchmark for evaluating approximate performance (which is within factor of 2x from equivalent JSON parsing).

What is missing:

Optimizations to make number handling as efficient as from JSON (but note: even with existing code, performance is typically limited by I/O and NOT parsing or generation)
Extensive performance measurements
Tests for various "exotic" CSV -- while basic escaping and quoting are tested, CSV is notoriously vaguely specified.

Maven dependency

To use this extension on Maven-based projects, use following dependency:

<dependency>
  <groupId>com.fasterxml.jackson.dataformat</groupId>
  <artifactId>jackson-dataformat-csv</artifactId>
  <version>2.3.0</version>
</dependency>

Usage

CSV Schema: what is that?

CSV documents are essentially rows of data, instead of JSON Objects (sequences of key/value pairs).

So one potential way to expose this data is to expose a sequence of JSON arrays; and similarly allow writing of arrays. Jackson supports this use-case (which works if you do not pass "CSV schema"), but it is not a very convenient way.

The alternative (and most commonly used) approach is to use a "CSV schema", object that defines set of names (and optionally types) for columns. This allows CsvParser to expose CSV data as if it was a sequence of JSON objects, name/value pairs.

3 ways to define Schema

So how do you get a CSV Schema instance to use? There are 3 ways:

Create schema based on a Java class
Build schema manually
Use the first line of CSV document to get the names (no types) for Schema

Here is code for above cases:

// Schema from POJO (usually has @JsonPropertyOrder annotation)
CsvSchema schema = mapper.schemaFor(Pojo.class);

// Manually-built schema: one with type, others default to "STRING"
CsvSchema schema = CsvSchema.builder()
        .addColumn("firstName")
        .addColumn("lastName")
        .addColumn("age", CsvSchema.ColumnType.NUMBER)
        .build();

// Read schema from the first line; start with bootstrap instance
// to enable reading of schema from the first line
// NOTE: reads schema and uses it for binding
CsvSchema bootstrapSchema = CsvSchema.emptySchema().withHeader();
ObjectMapper mapper = new CsvMapper();
mapper.reader(Pojo.class).with(bootstrapSchema).readValue(json);

It is important to note that the schema object is needed to ensure correct ordering of columns; schema instances are immutable and fully reusable (as are ObjectWriter instances).

Note also that while explicit type can help efficiency it is usually not required, as Jackson data binding can do common conversions/coercions such as parsing numbers from Strings.

Data-binding with schema

CSV content can be read either using CsvFactory (and parser, generators it creates) directly, or through CsvMapper (extension of standard ObjectMapper).

When using CsvMapper, you will be creating ObjectReader or ObjectWriter instances that pass CsvSchema along to CsvParser / CsvGenerator. When creating parser/generator directly, you will need to explicitly call setSchema(schema) before starting to read/write content.

The most common method for reading CSV data, then, is:

CsvMapper mapper = new CsvMapper();
Pojo value = ...;
CsvSchema schema = mapper.schemaFor(Pojo.class); // schema from 'Pojo' definition
String csv = mapper.writer(schema).writeValueAsString();
Pojo result = mapper.reader(Pojo.class).with(schema).read(csv);

Data-binding without schema

But even if you do not know (or care) about column names you can read/write CSV documents. The main difference is that in this case data is exposed as a sequence of ("JSON") Arrays, not Objects, as "raw" tabular data.

So let's consider following CSV input:

a,b
c,d
e,f

By default, Jackson CsvParser would see it as equivalent to following JSON:

["a","b"]
["c","d"]
["e","f"]

This is easy to use; in fact, if you ignore everything to do with Schema from above examples, you get working code. For example:

CsvMapper mapper = new CsvMapper();
// important: we need "array wrapping" (see next section) here:
mapper.enable(CsvParser.Feature.WRAP_AS_ARRAY);
File csvFile = new File("input.csv"); // or from String, URL etc
MappingIterator<Object[]> it = mapper.reader(Object[].class).readValues(csvFile);
while (it.hasNext()) {
  Object[] row = it.next();
}

Adding virtual Array wrapping

In addition to reading things as root-level Objects or arrays, you can also force use of virtual "array wrapping".

This means that using earlier CSV data example, parser would instead expose it similar to following JSON:

[
  ["a","b"]
  ["c","d"]
  ["e","f"]
]

This is useful if functionality expects a single ("JSON") Array; this was the case for example when using ObjectReader.readValues() functionality.

Documentation

Documentation (includes javadocs)
External articles:
CSV with Jackson 2.0

ffbit / jackson-dataformat-csv