CSV is written without column separators if first column is null

[Magrath]

There seems to be a bug in CsvWriter in the code dealing with the buffer. When the first column is null, the row is output without column separators. See unit tests below. testWrite_NullFirstColumn fails (all others pass). Expected output is ,42,hello\n, actual output is 42hello\n.

Tested against jackson-dataformat-csv v2.3.2.

import com.fasterxml.jackson.core.JsonProcessingException;
import com.fasterxml.jackson.databind.ObjectWriter;
import com.fasterxml.jackson.dataformat.csv.CsvMapper;
import com.fasterxml.jackson.dataformat.csv.CsvSchema;
import com.google.common.collect.ImmutableMap;
import org.junit.Test;

import static org.junit.Assert.assertEquals;

public class CsvWriterTest {

    @Test
    public void testWrite_NoNulls() throws JsonProcessingException {
        final CsvSchema.Builder csvSchemaBuilder = new CsvSchema.Builder();
        csvSchemaBuilder.addColumn("timestamp", CsvSchema.ColumnType.STRING);
        csvSchemaBuilder.addColumn("value", CsvSchema.ColumnType.NUMBER);
        csvSchemaBuilder.addColumn("id", CsvSchema.ColumnType.STRING);
        final CsvSchema schema = csvSchemaBuilder.build();
        final ObjectWriter writer = new CsvMapper().writer().withSchema(schema);

        final String string = writer.writeValueAsString(
                ImmutableMap.of("timestamp", "2014-03-10T23:32:47+00:00",
                        "value", 42, "id", "hello"));

        assertEquals("\"2014-03-10T23:32:47+00:00\",42,hello\n", string);
    }

    @Test
    public void testWrite_NullFirstColumn() throws JsonProcessingException {
        final CsvSchema.Builder csvSchemaBuilder = new CsvSchema.Builder();
        csvSchemaBuilder.addColumn("timestamp", CsvSchema.ColumnType.STRING);
        csvSchemaBuilder.addColumn("value", CsvSchema.ColumnType.NUMBER);
        csvSchemaBuilder.addColumn("id", CsvSchema.ColumnType.STRING);
        final CsvSchema schema = csvSchemaBuilder.build();
        final ObjectWriter writer = new CsvMapper().writer().withSchema(schema);

        final String string = writer.writeValueAsString(
                ImmutableMap.of("value", 42, "id", "hello"));

        assertEquals(",42,hello\n", string);
    }

    @Test
    public void testWrite_NullSecondColumn() throws JsonProcessingException {
        final CsvSchema.Builder csvSchemaBuilder = new CsvSchema.Builder();
        csvSchemaBuilder.addColumn("timestamp", CsvSchema.ColumnType.STRING);
        csvSchemaBuilder.addColumn("value", CsvSchema.ColumnType.NUMBER);
        csvSchemaBuilder.addColumn("id", CsvSchema.ColumnType.STRING);
        final CsvSchema schema = csvSchemaBuilder.build();
        final ObjectWriter writer = new CsvMapper().writer().withSchema(schema);

        final String string = writer.writeValueAsString(
                ImmutableMap.of("timestamp", "2014-03-10T23:32:47+00:00",
                        "id", "hello"));

        assertEquals("\"2014-03-10T23:32:47+00:00\",,hello\n", string);
    }

    @Test
    public void testWrite_NullThirdColumn() throws JsonProcessingException {
        final CsvSchema.Builder csvSchemaBuilder = new CsvSchema.Builder();
        csvSchemaBuilder.addColumn("timestamp", CsvSchema.ColumnType.STRING);
        csvSchemaBuilder.addColumn("value", CsvSchema.ColumnType.NUMBER);
        csvSchemaBuilder.addColumn("id", CsvSchema.ColumnType.STRING);
        final CsvSchema schema = csvSchemaBuilder.build();
        final ObjectWriter writer = new CsvMapper().writer().withSchema(schema);

        final String string = writer.writeValueAsString(
                ImmutableMap.of("timestamp", "2014-03-10T23:32:47+00:00",
                        "value", 42));

        assertEquals("\"2014-03-10T23:32:47+00:00\",42\n", string);
    }

}

[cowtowncoder]

Thank you for reporting this, and especially for providing unit tests. It sounds like a bug indeed.

[cowtowncoder]

Thanks again -- there was a bug in code that appends (or not...) column separators, for cases where buffering is needed for out-of-order writes. Fixed it in master (for 2.4), as well as 2.3 branch for 2.3.3.

[Magrath]

Thanks for the quick fix! :)

[cowtowncoder]

Many thanks for reporting this -- I hope to get CSV module as reliable as possible, and we get there one fixed bug at a time. :)

[hkothari]

Why is the default behavior to leave out the commas for null columns at the end, ie. in the last assertion we see:

assertEquals("\"2014-03-10T23:32:47+00:00\",42\n", string);

why is it not:

assertEquals("\"2014-03-10T23:32:47+00:00\",42,\n", string);

It seems inconsistent to leave out the last comma if the provided entry has a missing value there because you then end up with rows that have different number of commas.