You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Provided that the only thing that changes in the two hypothetical de-serialization environment you're thinking about, the only reason (I'm seeing) that would cause loading files created using an old version of cloudpickle using a newer one would be a change in behavior/signature of objects reconstructors used by cloudpickle between the two versions. This situation has happened in the past - we've tackled the most recent cases by following some form of deprecation cycle of old pickle files, see #368: so in that case, an 'old' pickle file relying on removed/deprecated function would remain compatible for loading using cloudpickle for a few more minor releases, although this deprecation is best-effort only. It would be worth adding a paragraph about this in the docs, that currently only discuss the case of larger environment changes (different Python versions, a case we don't have control over since cloudpickle relies on python internals, that can change over time).
I'm unable to find information about forwards/backwards compatibility expectations between versions of cloudpickle.
For example if I dump data encoded in version 1.4 of cloud pickle should version 1.5 of cloudpickle be able to decode it?
From my reading it seems like there should be expectation, is that right? If so ideally that could be marked in the readme.
Relate issue:
#126
The text was updated successfully, but these errors were encountered: