jackson-dataformat-csv：没有 POJO 的映射数字值

Question

I'm trying to parse a CSV file using jackson-dataformat-csv and I want to map the numeric column to the Number java type.

CsvSchema schema = CsvSchema.builder().setUseHeader(true)
    .addColumn("firstName", CsvSchema.ColumnType.STRING)
    .addColumn("lastName", CsvSchema.ColumnType.STRING)
    .addColumn("age", CsvSchema.ColumnType.NUMBER)
    .build();

CsvMapper csvMapper = new CsvMapper();  

MappingIterator<Map<String, Object>> mappingIterator = csvMapper
        .readerFor(Map.class)
        .with(schema)
        .readValues(is);        

while (mappingIterator.hasNext()) {
    Map<String, Object> entryMap = mappingIterator.next();
    Number age = (Number) entryMap.get("age");
}       

I'm expecting entryMap.get("age") should be a Number , but I get String instead.

My CSV file:

firstName,lastName,age
John,Doe,21
Error,Name,-10

I know that CsvSchema works fine with POJOs, but I need to process arbitrary CSV schemas, so I can't create a new java class for every case.

Any way to parse CSV into a typed Map or Array ?

Answer 1

Right now it is not possible to configure Map deserialisation using CsvSchema . Process uses com.fasterxml.jackson.databind.deser.std.MapDeserializer which right now does not check schema. We could write custom Map deserialiser. There is a question on GitHub:CsvMapper does not respect CsvSchema.ColumnType when using @JsonAnySetter where cowtowncoder answered:

At this point schema type is not used much for anything, but I agree it should.

EDIT

I decided to take a look closer what we can do with that fact that com.fasterxml.jackson.databind.deser.std.MapDeserializer is used behind the scene. Implementing custom Map deserialiser which will take care about types would be tricky to implement and register but we can use knowledge about ValueInstantiator . Let's define new Map type which knows what to do with ColumnType info:

class CsvMap extends HashMap<String, Object> {

    private final CsvSchema schema;
    private final NumberFormat numberFormat = NumberFormat.getInstance();

    public CsvMap(CsvSchema schema) {
        this.schema = schema;
    }

    @Override
    public Object put(String key, Object value) {
        value = convertIfNeeded(key, value);
        return super.put(key, value);
    }

    private Object convertIfNeeded(String key, Object value) {
        CsvSchema.Column column = schema.column(key);
        if (column.getType() == CsvSchema.ColumnType.NUMBER) {
            try {
                return numberFormat.parse(value.toString());
            } catch (ParseException e) {
                // leave it as it is
            }
        }

        return value;
    }
}

For new type without no-arg constructor we should create new ValueInstantiator :

class CsvMapInstantiator extends ValueInstantiator.Base {

    private final CsvSchema schema;

    public CsvMapInstantiator(CsvSchema schema) {
        super(CsvMap.class);
        this.schema = schema;
    }

    @Override
    public Object createUsingDefault(DeserializationContext ctxt) {
        return new CsvMap(schema);
    }

    @Override
    public boolean canCreateUsingDefault() {
        return true;
    }
}

Example usage:

import com.fasterxml.jackson.databind.DeserializationContext;
import com.fasterxml.jackson.databind.MappingIterator;
import com.fasterxml.jackson.databind.ObjectReader;
import com.fasterxml.jackson.databind.deser.ValueInstantiator;
import com.fasterxml.jackson.databind.module.SimpleModule;
import com.fasterxml.jackson.dataformat.csv.CsvMapper;
import com.fasterxml.jackson.dataformat.csv.CsvSchema;

import java.io.File;
import java.io.IOException;
import java.text.NumberFormat;
import java.text.ParseException;
import java.util.HashMap;

public class CsvApp {

    public static void main(String[] args) throws IOException {
        File csvFile = new File("./resource/test.csv").getAbsoluteFile();

        CsvSchema schema = CsvSchema.builder()
                .addColumn("firstName", CsvSchema.ColumnType.STRING)
                .addColumn("lastName", CsvSchema.ColumnType.STRING)
                .addColumn("age", CsvSchema.ColumnType.NUMBER)
                .build().withHeader();

        // Create schema aware map module
        SimpleModule csvMapModule = new SimpleModule();
        csvMapModule.addValueInstantiator(CsvMap.class, new CsvMapInstantiator(schema));

        // register map
        CsvMapper csvMapper = new CsvMapper();
        csvMapper.registerModule(csvMapModule);

        // get reader for CsvMap + schema
        ObjectReader objectReaderWithSchema = csvMapper
                .readerWithSchemaFor(CsvMap.class)
                .with(schema);

        MappingIterator<CsvMap> mappingIterator = objectReaderWithSchema.readValues(csvFile);

        while (mappingIterator.hasNext()) {
            CsvMap entryMap = mappingIterator.next();

            Number age = (Number) entryMap.get("age");
            System.out.println(age + " (" + age.getClass() + ")");
        }
    }
}

Above code for below CSV payload:

firstName,lastName,age
John,Doe,21
Error,Name,-10.1

prints:

21 (class java.lang.Long)
-10.1 (class java.lang.Double)

It looks like a hack but I wanted to show this possibility.

Answer 2

You can use univocity-parsers for this sort of thing. It's faster and way more flexible:

CsvParserSettingssettings = new CsvParserSettings(); //configure the parser if needed
CsvParser parser = new CsvParser(settings);

for (Record record : parser.iterateRecords(is)) {
    Short age = record.getShort("age");
}

To get a typed map, tell the parser what is the type of the columns you are working with:

parser.getRecordMetadata().setTypeOfColumns(Short.class, "age" /*, and other column names*/);

//to get 0 instead of nulls when the field is empty in the file:
parser.getRecordMetadata().setDefaultValueOfColumns("0", "age", /*, and other column names*/);

// then parse
for (Record record : parser.iterateRecords(is)) {
    Map<String,Object> map = record.toFieldMap();
}

Hope this helps

Disclaimer: I'm the author of this library. It's open source and free (Apache 2.0 license)