Two helpers that don't do what their names say: disallow_overwrites (no-op) and cache_func_outputs (default cache fails, kwargs never cached)

[thorwhalen]

Found while verifying docstring claims for the WP6 documentation sweep (#100). Docs now state the current behaviour; the code itself was left alone.

dol.trans.disallow_overwrites(store, *, error_msg=None, disable_deletes=True) defines an inner __setitem__ but never assigns it to store, ignores error_msg and disable_deletes, and returns None. So disallow_overwrites(D) is None and D still overwrites. dol.mixins.OverWritesNotAllowedMixin does the intended job. Either finish it (attach the override, honour disable_deletes, return the class) or delete it.

dol.caching.cache_func_outputs(cache=HashableDict): with the default, get_cache(HashableDict) returns the class, so the first call raises TypeError: argument of type 'type' is not iterable. With an explicit cache, kwargs are wrapped in a fresh HashableDict whose hash is id(self), so calls with keyword arguments never hit the cache. Only positional-only calls with an explicit cache cache anything.

Also worth a look, not verified as a bug: dol.paths.add_prefix_filtering(d, relativize_prefix=True)['a/'] returned the whole dict for d = {'a/b': 1, 'a/c': 2, 'd': 3} in a quick run.

[thorwhalen]

Fixed in #104 (squash-merged) -- chose "finish" over "delete" for both helpers since they're now correct, useful, and not part of the public dol/__init__.py surface (so #94's cruft audit can still relocate them later if fleet adoption stays at zero). Reviewed by an independent sub-agent before merge, which caught two follow-up issues (the {k} error_msg placeholder didn't actually work; unhashable kwargs now raise TypeError instead of silently never caching) -- both fixed in the same PR. The third, unverified observation (add_prefix_filtering returning the whole dict for an out-of-scope-looking call) is the same family of bug as #82 (prefix relativization) -- tracked there, not here.

Closing.