A cache function to cache arbitrary R code

[jimhester]

This is a proof of concept, caching arbitrary R code.

This uses the existing memoise machinery to compute a hash based on the input code / expression and any additional variables (specified as formulas in ...), so does not take a cache filename.

You can use it with the in-memory cache, but you need to specify the cache up front, otherwise a new cache is created on every invocation.

mem <- cache_memory()
set.seed(42)
cache({runif(2)}, cache = mem)
#> [1] 0.9148060 0.9370754
cache({runif(2)}, cache = mem)
#> [1] 0.9148060 0.9370754
cache({runif(1)}, cache = mem)
#> [1] 0.2861395
cache({runif(1)}, cache = mem)
#> [1] 0.2861395

Improvements would be to use rlang to construct the function and quote the code / extra arguments. We would also need to change how the ... arguments are handled in regular memoise to use rlang::dots_quos() and rlang::tidy_eval().

[jimhester]

@hadley this is not exactly the API you were prototyping as it does not take a filename parameter, instead the filenames (if you are using the filesystem backend) are a hash of the function body.

[hadley]

Is the eval necessary here? Won't list(...) force all the promises?

[hadley]

Oh because they're formulas

[hadley]

Why not just do cache$set(key, f()) ?

[jimhester]

Good question!