SequenceWriter#write(null) writes a single null, not an entire row of nulls

[georgewfraser]

It requires very specific conditions, it is reproduced here (along with a bunch of similar cases that DO work):

https://gist.github.com/georgewfraser/ef19edb1eca296bd3234

It would be OK to just skip writing the row, but this behavior is really bad because if you write a lot nulls you end up with a long row with too many columns like ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,

[georgewfraser]

Possible fix to CsvGenerator.java:

    public void writeNull() throws IOException {
        _verifyValueWrite("write null value");
        if (!_skipValue) {
            if (_arraySeparator >= 0) {
                _addToArray(_schema.getNullValue());
            } else {
                _writer.writeNull(_columnIndex());
            }

            // If we are told to write null, but we were never told to begin an object
            // that means the entire row is just a null, so we are now finished writing the row
            if (!_writeContext.inObject()) {
                finishRow();
            }
        }
    }

The last part just guesses that we're in the row = null case if we're writing a null and not inside an object. I'm not sure whether it belongs inside or outside if(!_skipValue) because I don't know what that is about.

[cowtowncoder]

I agree, exact semantics are bit tricky. Writing an empty row sounds like a reasonable way to go, although this may cause problems when using custom null representation. But we can cross that bridge when we get there.

[georgewfraser]

Until a fix makes it into the next Jackson release, here's a workaround for anyone who needs it. Usage is new CsvMapper(new PatchCsvFactory())

https://gist.github.com/georgewfraser/52780f7d82d2b5ec80dc

[cowtowncoder]

Interestingly it is quite easy to add one of two possibilities:

1. Simply output nothing (ignore)
2. Output "empty Object" (all commas)

so it'd even be possible to configure behavior.

I'll go ahead and implement (1) first.