Pyyaml: Yaml-4.1 failed to serialize or deserialize numpy objects.

[lyuwen]

```>>> import yaml

import numpy as np
yaml.dump(np.random.random((100,100)))
Traceback (most recent call last):
File "", line 1, in
File "/home/lf2507/.local/lib/python2.7/site-packages/yaml/__init__.py", line 219, in dump
return dump_all([data], stream, Dumper=Dumper, **kwds)
File "/home/lf2507/.local/lib/python2.7/site-packages/yaml/__init__.py", line 198, in dump_all
dumper.represent(data)
File "/home/lf2507/.local/lib/python2.7/site-packages/yaml/representer.py", line 28, in represent
node = self.represent_data(data)
File "/home/lf2507/.local/lib/python2.7/site-packages/yaml/representer.py", line 67, in represent_data
node = self.yaml_representersNone
File "/home/lf2507/.local/lib/python2.7/site-packages/yaml/representer.py", line 249, in represent_undefined
raise RepresenterError("cannot represent an object", data)
yaml.representer.RepresenterError: ('cannot represent an object', array([[0.78867287, 0.1145689 , 0.73455115, ..., 0.03162476, 0.06458778,
0.02051163],
[0.0648247 , 0.44383162, 0.80787024, ..., 0.07586477, 0.47848779,
0.05648012],
[0.84362173, 0.31730949, 0.99826481, ..., 0.8951278 , 0.87577464,
0.64970616],
...,
[0.98758733, 0.26062501, 0.68675972, ..., 0.82339569, 0.59255629,
0.24641607],
[0.11395974, 0.41112024, 0.64924667, ..., 0.33038519, 0.8491808 ,
0.87419596],
[0.44593807, 0.15025162, 0.16101452, ..., 0.25134148, 0.30600472,
0.38231974]]))

```

[alex]

This is a deliberately change. Previously PyYAML's default for serialization/deserialization was to use a dangerous subset of yaml that allowed handling arbitrary python objects - meaning that deserialization of untrusted input could lead to arbitrary code execution.

As a result, PyYAML 4 is now safe by default. Those wanting the previous behavior can use yaml.danger_load and yaml.danger_dump.

[lyuwen]

But then do you mean dumping a numpy object is dangerous? And if I want to serialize it I'll have to use the yaml.danger_dump?

[alex]

Loading an object using yaml.load() in pyyaml < 4 was dangerous, yes.
Now the naming reflects this.

On Tue, Jun 26, 2018 at 8:12 PM LFu [e-mail] wrote:

But then do you mean dumping a numpy object is dangerous? And if I want to
serialize it I'll have to use the yaml.danger_dump?

—
You are receiving this because you modified the open/close state.
Reply to this email directly, view it on GitHub
https://github.com/yaml/pyyaml/issues/177#issuecomment-400501913, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAADBB_ZKfjbin_yBFVJeyz9BZSnQM2Iks5uAs3ZgaJpZM4U42r3
.

--
"I disapprove of what you say, but I will defend to the death your right to
say it." -- Evelyn Beatrice Hall (summarizing Voltaire)
"The people's good is the highest law." -- Cicero
GPG Key fingerprint: D1B3 ADC0 E023 8CA6

[lyuwen]

Why can't you 'safely' serialize a numpy object?

[alex]

Because danger_load (yaml.load in yaml < 4) can load arbitrary Python objects, which leads to arbitrary code execution.

Someone could write a wrapper around yaml that special cases numpy objects to handle them safely.

[mithrandi]

To be precise, I think danger_dump itself is not dangerous, but it produces output that (potentially) can only be loaded by danger_load, hence the naming.

[ingydotnet]

I am planning to slightly change the top level API very quickly. As soon as some issues with the 4.1 release have been resolved.

I didn't happen to see this PR go in, and I think it should be done better.

In essence, I agree that safe loading is a good default behavior (schema, in yaml terms), but not a good default schema for dumping. Also not compatible with where the API intends to go in the future.

Stay, tuned. Coming soon...

[ingydotnet]

PyYAML 5.1 is released and addresses this issue. Closing.