Can I dump blank instead of null in yaml/pyyaml?

You get null because dump() uses the Representer() which subclasses SafeRepresenter() and to represent None, the following method is called:

def represent_none(self, data):
    return self.represent_scalar(u'tag:yaml.org,2002:null',
                                 u'null')

As the string null is hardcoded, there is no option to dump() to change that.

The proper way to solve this in PyYAML is to make your own Dumper subclass which has the Emitter, Serializer, and Resolver from the standard Dumper that dump() uses, but with subclass of Representer that represents None the way you want it:

import sys
import yaml

from yaml.representer import Representer
from yaml.dumper import Dumper
from yaml.emitter import Emitter
from yaml.serializer import Serializer
from yaml.resolver import Resolver


yaml_str = """\
attrs:
  first:
  second: value2
"""

class MyRepresenter(Representer):
    def represent_none(self, data):
        return self.represent_scalar(u'tag:yaml.org,2002:null',
                                 u'')

class MyDumper(Emitter, Serializer, MyRepresenter, Resolver):
    def __init__(self, stream,
            default_style=None, default_flow_style=None,
            canonical=None, indent=None, width=None,
            allow_unicode=None, line_break=None,
            encoding=None, explicit_start=None, explicit_end=None,
            version=None, tags=None):
        Emitter.__init__(self, stream, canonical=canonical,
                indent=indent, width=width,
                allow_unicode=allow_unicode, line_break=line_break)
        Serializer.__init__(self, encoding=encoding,
                explicit_start=explicit_start, explicit_end=explicit_end,
                version=version, tags=tags)
        MyRepresenter.__init__(self, default_style=default_style,
                default_flow_style=default_flow_style)
        Resolver.__init__(self)

MyRepresenter.add_representer(type(None),
                              MyRepresenter.represent_none)

data = yaml.load(yaml_str)
yaml.dump(data, stream=sys.stdout, Dumper=MyDumper, default_flow_style=False)

gives you:

attrs:
  first:
  second: value2

If that sounds like a lot of overhead just to get rid of null, it is. There are some shortcuts you can take and you can even try to graft the alternate function onto the existing Representer, but since the actual function taken is referenced in a lookup table ( populated by add_representer ) you need to handle at least that reference as well.

The far more easy solution is replace PyYAML with ruamel.yaml and use its round_trip functionality (disclaimer: I am the author of that package):

import ruamel.yaml

yaml_str = """\
# trying to round-trip preserve empty scalar
attrs:
  first:
  second: value2
"""

data = ruamel.yaml.round_trip_load(yaml_str)
assert ruamel.yaml.round_trip_dump(data) == yaml_str

apart from emitting None as the empty scalar, it also preserves order in mapping keys, comments and tag names, none of which PyYAML does. ruamel.yaml also follows the YAML 1.2 specification (from 2009), where PyYAML uses the older YAML 1.1.

---

_{The ruamel.yaml package can be installed with pip from PyPI, or with modern Debian based distributions, also with apt-get python-ruamel.yaml}

Based on @Anthon's excellent answer, I was able to craft this solution:

def represent_none(self, _):
    return self.represent_scalar('tag:yaml.org,2002:null', '')

yaml.add_representer(type(None), represent_none)

Based on my understanding of the PyYAML code, adding a representer for an existing type should simply replace the existing representer.

This is a global change and that means that all following dumps use a blank. If some unrelated other piece of code in your program relies on None to be represented in the "normal" way, e.g. a library that you import and that uses PyYAML as well, that library will no longer work in the exepected way/correctly, in that case subclassing is the correct way to go.

Extending @Jace Browning's answer while addressing @Anthon's concern, we can use a context manager which remembers the prior implementation of None:

class BlankNone(Representer):
    """Print None as blank when used as context manager"""
    def represent_none(self, *_):
        return self.represent_scalar(u'tag:yaml.org,2002:null', u'')

def __enter__(self):
    self.prior = Dumper.yaml_representers[type(None)]
    yaml.add_representer(type(None), self.represent_none)

def __exit__(self, exc_type, exc_val, exc_tb):
    Dumper.yaml_representers[type(None)] = self.prior

which can be used thus:

 with BlankNone(), open(file, 'wt') as f:
        yaml.dump(hosts, f)