UnhashableTypeError: Cannot hash object of type _json.Scanner...

Summary

When attempting to use st.cache decorator (with streamlit 1.10.0) on a json data loader method, streamlit returns an unhashable error for a _json.Scanner. This behavior does not exist in the previous version (1.9.2).

Steps to reproduce

1. Update streamlit to version 1.10.0
2. Use st.cache on any method that loads data from a json file

For example:

@st.cache(allow_output_mutation=True)
def fetch_data_json(filename: str) -> Dict[str, Any]:
    with open(filename, "r") as f:
        config = json.load(f)
    return config

Returns:

UnhashableTypeError: Cannot hash object of type _json.Scanner, found in the body of fetch_data_json().

While caching the body of fetch_data_json(), Streamlit encountered an object of type _json.Scanner, which it does not know how to hash.

To address this, please try helping Streamlit understand how to hash that type by passing the hash_funcs argument into @st.cache. For example:


@st.cache(hash_funcs={_json.Scanner: my_hash_func})
def my_func(...):
  
If you don't know where the object of type _json.Scanner is coming from, try looking at the hash chain below for an object that you do recognize, then pass that to hash_funcs instead:


Object of type _json.Scanner: <_json.Scanner object at 0x7fa22470b160>
Object of type builtins.method: <bound method JSONDecoder.raw_decode of <json.decoder.JSONDecoder object at 0x7fa2247b0760>>
Object of type builtins.method: <bound method JSONDecoder.decode of <json.decoder.JSONDecoder object at 0x7fa2247b0760>>
Object of type builtins.function: <function loads at 0x7fa22470e3a0>
Object of type builtins.function: <function load at 0x7fa22470e310>
Object of type builtins.function: <function fetch_data_json at 0x7fa22128cdc0>
Please see the hash_funcs [documentation] (https://docs.streamlit.io/library/advanced-features/caching#the-hash_funcs-parameter) for more details.

Traceback:
File "app.py", line 23, in <module>
    json_file = fetch_data_json(TEST_CONFIG_FOLDER + "/test_config.json")
File "/usr/lib/python3.8/copyreg.py", line 66, in _reduce_ex
    raise TypeError(f"cannot pickle {cls.__name__!r} object")

Expected behavior:

In version 1.9.2 the same code would produce valid parsed json data without @st.cache attempting to has an unhashable object. This image below is the method outlined above implemented in 1.10.0 and 1.9.2:

Debug info

• Streamlit version: 1.10.0
• Python version: 3.8.12
• Using Conda? PipEnv? PyEnv? Pex? Pip via Poetry
• OS version: Ubuntu 20.04 (via Windows 11)
• Browser version: Chrome 102.0.5005.115

[LexABzH]

Hello.

I’m having exactly the same issue here. I have updated Streamlit from 1.7.0 to 1.10.0 and I’m now having an UnhashableTypeError without changing anything in my code.

Weird thing is that sometimes it randomly work (let’s say 1 time out of 10).

I was able to find a temporary fix by adding hash_funcs={'_json.Scanner': hash} to each @st.cache decorator that triggers this error, as stated in an old issue -> https://github.com/streamlit/streamlit/issues/1627#issuecomment-779293099. So far it seems to work.

[vdonato]

Hey @amhrasmussen,

Release 1.11.1 should be more or less the same release as 1.11.0, except with a security patch cherry-picked onto it. I believe #5023 was merged after 1.11.0 was released, so I wouldn’t expect it to appear in the patch release.

Somewhat conveniently, release 1.12.0 is being cut today (as far as I know), so we should have an official release including #5023 out on Thursday.