charmonium-cache 0.6.0

Provides a decorator for caching a function and an equivalent command-line util.

Provides a decorator for caching a function and an equivalent command-line util.

It wraps an ordinary function. Whenever the function is called with the same arguments, the result is loaded from the cache instead of computed.

Quickstart

$ pip install charmonium.cache

>>> import charmonium.cache as ch_cache
>>> @ch_cache.decor(ch_cache.MemoryStore.create())
... def square(x):
...     print('computing')
...     return x**2
...
>>> square(4)
computing
16
>>> square(4) # square is not called again; answer is just looked up
16
>>> with square.disabled():
...     # disable caching; always recomptue
...     square(4)
...
computing
16

Customization

cache_decor is flexible because it supports multiple backends.

1. MemoryStore: backed in RAM for the duration of the program (see example above).

2. FileStore: backed in a file which is loaded on first call.

>>> import charmonium.cache as ch_cache
>>> @ch_cache.decor(ch_cache.FileStore.create("tmp/1"))
... def square(x):
...     return x**2
...
>>> # Now that we cache in a file, this is persistent between runs
>>> # So I must clear it here.
>>> square.clear()
>>> list(map(square, range(5)))
[0, 1, 4, 9, 16]
>>> import os
>>> os.listdir("tmp/1")
['__main__.square_cache.pickle']

3. DirectoryStore: backed in a directory. Results are stored as individual files in that directory, and they are loaded lazily. Use this for functions that return large objects.

>>> import charmonium.cache as ch_cache
>>> @ch_cache.decor(ch_cache.DirectoryStore.create("tmp/2"))
... def square(x):
...     return x**2
...
>>> # Now that we cache in a file, this is persistent between runs
>>> # So I must clear it here.
>>> square.clear()
>>> list(map(square, range(5)))
[0, 1, 4, 9, 16]
>>> import os
>>> sorted(os.listdir("tmp/2/__main__.square"))
['(0).pickle', '(1).pickle', '(2).pickle', '(3).pickle', '(4).pickle']

4. Custom stores: to create a custom store, just extend ObjectStore and implement a dict-like interface.

FileStore and DirectoryStore can both themselves be customized by:

• Providing a cache_path (conforming to the PathLike interface), e.g. one can transparently cache in AWS S3 with an S3Path object.

• Providing a serializer (conforming to the Serializer interface), e.g. pickle (default), cloudpickle, dill, or messagepack.

cache_decor also takes a “state function” which computes the value of some external state that this computation should depend on. Unlike the arguments (which the cache explicitly depends on), values computed with a different state are evicted out, so this is appropriate when you never expect to revisit a prior state (e.g. modtime of a file could be a state, as in make_file_state_fn).

With verbose=True, this will output to a logger.

>>> import charmonium.cache as ch_cache
>>> @ch_cache.decor(ch_cache.MemoryStore.create(), verbose=True)
... def square(x):
...     print('computing')
...     return x**2
...
>>> square(4) # doctest:+SKIP
2020-06-19 11:31:40,197 - __main__.square: miss with args: (4,), {}
computing
16
>>> square(4) # doctest:+SKIP
2020-06-19 11:31:40,197 - __main__.square: hit with args: (4,), {}
16

CLI

# cache a commandline function based on its args
$ cache --verbose -- compute_square 6
miss for square(["6"])
36

$ cache -- compute_square 6
hit for square(["6"])
36