Regression - RepresenterError on dict subclass

Due to #126, I’m running pyyaml at master. When I try to run code that passes on released versions of pyyaml, it fails in master:

-> raise RepresenterError("cannot represent an object", data)
(Pdb) type(data)
<class 'pmxbot.dictlib.ConfigDict'>
(Pdb) isinstance(data, dict)
True

As the object is a subclass of dict, it would previously be treated as a dict for the purpose of serialization. Is this an intentional change?

[jaraco]

I tried using the dangerous dumper, but found that creates a YAML file that requires a dangerous loader. What I really want is to instruct YAML to serialize a dict subclass exactly as it would a dict.

[dhimmel]

I also ran into this issue trying to dump an instance of a dict subclass.

I found an ugly workaround on StackOverflow that you must run before calling yaml.dump:

yaml.add_representer(
    Dict_Subclass,
    lambda dumper, data: dumper.represent_mapping('tag:yaml.org,2002:map', data.items())
)

Limitations:

1. Dict_Subclass must be the specific dict subclass. This doesn’t fix the issue for all dict subclasses, just those explicitly overriden.
2. It works with yaml.dump but not yaml.safe_dump.