Allow skipping ending line break (`CsvGenerator.Feature.WRITE_LINEFEED_AFTER_LAST_ROW`)

[mathieu-lavigne]

In RFC-4180 it is said that CSV implementations may choose to include a line break at the end of the file or not.

It could be nice to add an option to choose. For instance something like this :

schema.withEndingLineSeparator();
schema.withoutEndingLineSeparator();

[cowtowncoder]

Ok, sounds reasonable from usage perspective. Not sure how easy it'd be to do, wrt. when determination is made regarding adding a linefeed: if it's at the time of starting a new line, this should be quite easy.

There is also the question of whether this belongs in schema, or as a CsvGenerator.Feature. I can see either being possible choice.

[tyler2cr]

2.9.9 does not include #46

should the Labels or Milestone be changed?

[tyler2cr]

This requested feature may resolve an unexpected scenario:

These tests fail with java.lang.AssertionError: expected null, but was:<>

I've been unable to determine where the <> value comes from.

import com.fasterxml.jackson.annotation.JsonPropertyOrder;
import com.fasterxml.jackson.databind.MappingIterator;
import com.fasterxml.jackson.dataformat.csv.CsvMapper;
import com.fasterxml.jackson.dataformat.csv.CsvSchema;
import org.junit.Test;

import java.io.IOException;

import static org.junit.Assert.assertEquals;
import static org.junit.Assert.assertNotNull;
import static org.junit.Assert.assertNull;
import static org.junit.Assert.assertTrue;

public class newlinetest {

    @JsonPropertyOrder({"column1", "column2"})
    static
    class Pojo {
        private String column1;
        private String column2;

        public String getColumn1() {
            return column1;
        }

        public void setColumn1(String column1) {
            this.column1 = column1;
        }

        public String getColumn2() {
            return column2;
        }

        public void setColumn2(String column2) {
            this.column2 = column2;
        }
    }

    @Test
    public void two_rows() throws IOException {
        // setup
        String csv = "value|\nvalue|\n";
        CsvMapper csvMapper = new CsvMapper();
        CsvSchema csvSchema = csvMapper.schemaFor(Pojo.class).withColumnSeparator('|');

        // execute
        MappingIterator<Pojo> mappingIterator = csvMapper.readerFor(Pojo.class).with(csvSchema).readValues(csv);

        // test
        assertTrue(mappingIterator.hasNext());
        Pojo row1 = mappingIterator.next();
        assertNotNull(row1);
        assertEquals("value", row1.column1);
        assertNull(row1.column2);

    }

    @Test
    public void one_row() throws IOException {
        // setup
        String csv = "value|\n";
        CsvMapper csvMapper = new CsvMapper();
        CsvSchema csvSchema = csvMapper.schemaFor(Pojo.class).withColumnSeparator('|');

        // execute
        Pojo pojo = csvMapper.readerFor(Pojo.class).with(csvSchema).readValue(csv);

        // test
        assertNotNull(pojo);
        assertEquals("value", pojo.column1);
        assertNull(pojo.column2);
    }
}

[cowtowncoder]

@tyler2cr Label is purely aspirational, i.e. at the time of update possible target. Updated to 2.10 as this can not be added in 2.9 any more; 2.10 is still possible. Milestone will indicate actual inclusion if and when that happens.

[rage-shadowman]

@tyler2cr the "<>" is just how JUnit describes an empty string in the assertNull error message.

[tyler2cr]

@cowtowncoder Sounds good - thanks for the quick response!

[tyler2cr]

@rage-shadowman 🤦‍♂ thank you!!

After this realization, I used .withNullValue("") on the CsvSchema, and it works like a charm

[alturkovic]

Any progress on this?

[cowtowncoder]

@alturkovic I don't think so. I definitely haven't had time to try to implement this.

But if (or anyone else) has time and interest, would be happy to help get a contributed PR merged.

EDIT: I forgot there's a PR -- #46 - that proposes a way to do it. Unfortunately I don't think it quite works reliably in that there might be a case where multi-character linefeed (\r\n f.ex) could be split on buffer boundary.
But maybe it'd be possible to rework this.

On changes to #46: I think that this should be enabled using CsvGenerator.Feature instead of CsvSchema.
That would simplify PR a bit too.

Aside from that I think ensuring "undo last linefeed" works reliably is probably doable.