semantic-fscacher 0.3.0

File system based caching

fscacher

Caching solution for functions that operate on the file system.

Installation

pip install <path to repository>

Usage

Simple usage:

from fscacher import Cache

cache = Cache(cache_path)
fn = cache.memoize(expensive_fn)

result = fn("arg1", "arg2")

When fn is called the first time, the function expensive_fn is evaluated and its return value is serialized and stored at cache_path. Subsequent calls to fn deserializes the stored result instead of re-evaluating expensive_fn.

Optional arguments to memoize include:

• key: Function with arguments (func, args, kwargs) and return value of type str, used to create the function call signature
• dump: Function with arguments (return_value, filename) and no return value for serializing the result of expensive_fn
• load: Function with arguments (filename, ) for de-serializing the binary data on disk as a return value
• digest: Function with arguments (stream, ) and return value of type str for digesting function call signature as well as the contents of serialized files
• protocol: Use predefined functions for key, dump, load and digest. A list of known protocol schemes are presented below. If both protocol and explicit functions are set, the explicit functions takes presedence.

If any of the arguments key, dump, load or digest are set to "default", the functions defined by the cache.defaults dict are used instead.

Default key function is constructed as follows:

1. Arguments and keyword arguments (both keys and values) are converted to string by the str function. Numpy arrays are converted to lists before conversion.
2. Any arguments which are too long (> 22 chars) or contain invalid characters (= \/:*?"<>|) are utf-8 encoded and converted to a 64-bit truncated sha256 hash.
3. Keyword arguments are joined as key-value pairs of the form k=v.
4. If short enough, the key is the function name followed by the space-separated argument list. If too long, the key is the function name followed by the full sha256 hash of the space-separated argument list.

Default dump and load functions are from the python pickle module.

Default digest function is sha256.

Implemented protocols include:

• filename/<suffix>: Return value is interpreted as the name of a temporary file, which should have the suffix <suffix>, including any leading dot. key is default key, except that <suffix> is appended. dump moves the file to the index location. load returns the file name as a string.
• filehash/<suffix>: Return value is interpreted as the name of a temporary file, which should have the suffix <suffix>, including any leading dot. key is default key with the suffix .hash appended. dump computes the hash of the temporary file and renames it to the hash value (unless already present). Thereafter, it copies the file (as a hardlink if possible) to the location specified by key, except that .hash is replaced by <suffix>. Finally, the hash is stored as a string (lowercase hex) at the location specified by key (the index location). load returns the contents of the file at the index location and interprets it as a pathlib path. In the end, this protocol works like filename/<suffix>, except that multiple function calls can be mapped to the same file if their return value has equal contents.