New PyYAML version breaks on most custom python ob

Question

About 5 hours ago, version 4.1.0 was released. It is breaking my unit tests. Here is a clean MVCE displaying this:

Version 3.12:

>>> import numpy as np
>>> import yaml
>>> x = np.int64(2)
>>> yaml.dump(x, Dumper=yaml.Dumper)
'!!python/object/apply:numpy.core.multiarray.scalar\n- !!python/object/apply:numpy.dtype\n  args: [i8, 0, 1]\n  state: !!python/tuple [3, <, null, null, null, -1, -1, 0]\n- !!binary |\n  AgAAAAAAAAA=\n'

Version 4.1.0:

>>> import numpy as np
>>> import yaml
>>> x = np.int64(2)
>>> yaml.dump(x, Dumper=yaml.Dumper)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/foo/anaconda3/envs/bar/lib/python3.6/site-packages/yaml/__init__.py", line 217, in dump
    return dump_all([data], stream, Dumper=Dumper, **kwds)
  File "/foo/anaconda3/envs/bar/lib/python3.6/site-packages/yaml/__init__.py", line 196, in dump_all
    dumper.represent(data)
  File "/foo/anaconda3/envs/bar/lib/python3.6/site-packages/yaml/representer.py", line 26, in represent
    node = self.represent_data(data)
  File "/foo/anaconda3/envs/bar/lib/python3.6/site-packages/yaml/representer.py", line 57, in represent_data
    node = self.yaml_representers[None](self, data)
  File "/foo/anaconda3/envs/bar/lib/python3.6/site-packages/yaml/representer.py", line 229, in represent_undefined
    raise RepresenterError("cannot represent an object", data)
yaml.representer.RepresenterError: ('cannot represent an object', 2)

Is there a clear reason for why PyYAML no longer supports these object types?

Answer 1

dump is now safe_dump, which won't handle arbitrary objects:

>>> yaml.dump is yaml.safe_dump
True

Use danger_dump for the old behaviour.

>>> yaml.danger_dump(x)
'!!python/object/apply:numpy.core.multiarray.scalar\n- !!python/object/apply:numpy.dtype\n  args: [i8, 0, 1]\n  state: !!python/tuple [3, <, null, null, null, -1, -1, 0]\n- !!binary |\n  AgAAAAAAAAA=\n'

The same goes for load/safe_load. Can't find the docs or release notes for 4.1.0, I only found out by digging through the commits (here).

Is there a clear reason for why PyYAML no longer supports these object types?

Yes. yaml.load was allowing arbitrary code execution, and such a dangerous feature should be opt-in only - not possible to use by accident. Arguably, it should have been this way from the beginning, and I'm glad the new maintainers of PyYAML have corrected that.