Allow dumpdata in YAML using ruamel.yaml rather than PyYaml.

[Rémy Hubscher]

For some reason, ruamel.yaml is a recommanded lib for loading Yaml in Python.
However Django dumpdata command ask for PyYaml.

Could we consider having trying to use ruamel.yaml if PyYaml is not installed but ruamel.yaml is?

[Tim Graham]

Recommended by who? What are the advantages of switching? Is it backwards compatible?

[Daniel Rios]

Hi.
First, reasons should be provided as to why something is recommended. In this case, ruaml is recommended because it is, as far as my research has taken me, the only other python yaml library out there.

The preference for the pyyaml module is baked into django/core/serializers/__init__.py in the BUILTIN_SERIALIZERS constant at ​https://github.com/django/django/blob/5cc81cd9eb69f5f7a711412c02039b435c393135/django/core/serializers/__init__.py#L26

I don't think it makes sense to have another built-in serializer. However, there is a workaround:

Django does offer the possibility to use your own serializer in the docstring of the same file (​https://github.com/django/django/blob/5cc81cd9eb69f5f7a711412c02039b435c393135/django/core/serializers/__init__.py#L10), but the actual user-facing documentation seems to be nonexistent.

I was able to, however, overwrite the preference for pyyaml by installing ruamel and with SERIALIZATION_MODULES in settings.py as follows:

SERIALIZATION_MODULES = {
    'yaml': 'ruamel.yaml.serializer'
}

I was unable to get it running because ruaml (even though based on pyyaml) is incompatible to the way Django handles serializers. However, the fact that Django does support other serializers means that it would be possible to get ruaml to work.

An idea would be to create a third party package django-ruaml-serializer that patches the differences in ruaml to work with the existing django serializer.