Map NULL to empty file to consume zero memory instead of block size

[lorenzwalthert]

I came across your package and thought it would be a good match to implement caching with styler in r-lib/styler#538. A feature of the problem at hand is that there is no need to cache a specific value - the only thing we want to remember is if a computation has already been executed (in my case if the text to style has already been styled and hence does not need restyling). I could R.cache::saveCache(object = NULL, key = text, ...) or similar, but unfortunately, this will create a file with size > 0 bytes, and on my Mac, any file that is not totally empty takes 4KB of space beacuse that's the block size. In my usecase, it is likely that people will accumulate hundreds of cached values, which makes R.cache::saveCache(object = NULL, ...) a solution with too much overhead.

For this reason, I propose to map NULL to an empty file, For this reasons, I propose to introduce an argument shallow (defaulting to FALSE) in R.cache::saveCache() (or to create another function for shallow hashing if you prefer). When set, a file with base::file.create() would be created that is completely empty to accommodate the case in which keys should be cached, but values are not needed. The advantage is that this will take 0 bytes on disk (ignoring some metadata like file name, permissions ect. stored somewhere else on the file system). We could use the current implementation of R.cache::findCache() without an error, other functions that expect meta data would probably need to be adapted but I believe we could easily identify if a cached value is NULL, just check if the file size is 0. I belive it is generic enough to have other applications than r-lib/styler#538.

You can find a basic implementation without tests and documentation in my fork.

[krlmlr]

Would it be easier to map NULL to empty files?

[lorenzwalthert]

Good idea, that's much more elegant I think.

[krlmlr]

@HenrikBengtsson: Do you think this is feasible? Would you consider a PR?

[HenrikBengtsson]

Thanks for the bump - this one slipped. Yes, I'm open to PRs. If just like to think thought that your needs are and what the options forward are.

[HenrikBengtsson]

I've just exported generateCache() in the develop branch;

remotes::install_github("HenrikBengtsson/R.cache@develop")

This will allow you to generate cache pathnames without creating files, so you can to do things like:

> pathname <- generateCache(key = list("foo"), dirs = c("styler", "this", "that"), suffix = ".styled")
> pathname
[1] "/home/hb/.cache/R/R.cache/styler/this/that/cfb1611e527451f9a3334e6f60768f5b.styled"
> already_styled <- utils::file_test("-f", pathname)
> already_styled
[1] FALSE
> res <- file.create(pathname)
> res
[1] TRUE
> utils::file_test("-f", pathname)
[1] TRUE
> file.size(pathname)
[1] 0

Would this be sufficient?

[lorenzwalthert]

It seems yes: r-lib/styler@36606a2

Thanks.

[HenrikBengtsson]

Good to hear. FYI, I'll soon start preparing the next CRAN submission of R.cache (https://github.com/HenrikBengtsson/R.cache/milestone/4?closed=1)