st.cache always shows a warning for mutated inputs, even if ignore_hash is True

Steps to reproduce

Get some data files:

python -m spacy download en_core_web_sm

Then run this:

import streamlit as st
import spacy

@st.cache(ignore_hash=True)
def load_model(name):
    return spacy.load(name)

@st.cache(ignore_hash=True)
def process_text(nlp, text):
    return nlp(text)  # This mutates the nlp object. So we raise a warning.

nlp = spacy.load("en_core_web_sm")
doc = process_text(nlp, "Hello world")

What you see:

Cached function mutated its input arguments

When decorating a function with @st.cache, the arguments should not be mutated inside the function body, as that breaks the caching mechanism. Please update the code of process_text to bypass the mutation.

See the Streamlit docs for more info.

Expected

It’s OK if we have a warning for this case, but there should be a way to turn it off too.

Debug info

• Streamlit version: 0.45.0

Possible solutions

1. Add kwarg allow_input_mutation=True that suppresses this warning and doesn’t check inputs when pulling from the cache.
2. This should actually be a different feature request, but if we added kwargs clone_inputs=True and clone_output=True, which allows mutation.

Addendum: should we rename or change the semantics of ignore_hash? There are different hashes in st.cache, and ignore_hash only ignores one of them.

[treuille]

@sam-qordoba : I believe this is the issue you’re seeing.

[ines]

@sam-qordoba That’s interesting – I just double-checked again and the only difference between the two files are the unicode declarations (that should only be necessary for Python 2, so I didn’t bother adding them to my initial example). So I guess there’s something else going on here.