Why are some mutable MNE objects hashable?

Several (mutable) MNE objects (e.g. Raw) are hashable, which violates one of the key assumptions that Python makes about hashable objects, namely that the hash value never changes during its lifetime. This can lead to bugs wherever hashable objects are expected not to change (so everywhere, such as using them as keys in dicts). @larsoner mentioned that this was done to support caching in joblib, but it would probably be a good idea to discuss other (Pythonic) solutions. Thoughts?

[larsoner]

Given that usually (or at least, often) user-defined classes are made to be mutable, it seems like a contradiction for Python to make user-defined classes by default be hashable, but say that hashable objects should not change.

So there is at least some inconsistency there, and at least our __hash__ functions are designed to make it so that if you do change something about the object (e.g., info or data), the __hash__ will also change (where it wouldn’t with the default id-based implementation).

[larsoner]

I’ll close assuming we can’t come up with a different way to use this. But in case you want to investigate @cbrnr here is a snippet (constructed quickly just from looking at the joblib docs), you could imagine the time.sleep could be any time-consuming process:

import time
import numpy as np
import joblib
import mne

data = np.random.RandomState(0).randn(1, 1000)
info = mne.create_info(1, 1000., 'eeg')
raw = mne.io.RawArray(data, info)
mem = joblib.Memory('.')


@mem.cache
def slow(raw):
    time.sleep(1.)


for _ in range(2):
    t0 = time.time()
    slow(raw)
    print(time.time() - t0)

Produces:

$ python -u cache.py 
________________________________________________________________________________
[Memory] Calling __main__--home-larsoner-Desktop-cache.slow...
slow(<RawArray | 1 x 1000 (1.0 s), ~15 kB, data loaded>)
_____________________________________________________________slow - 1.0s, 0.0min
1.008265733718872
0.001148223876953125
$ python -u cache.py 
0.003644227981567383
0.0012600421905517578