Sharing the C++ code to other packages

[Nowosad]

Hi @HajkD,

I am currently working on various improvements and extensions to the https://github.com/Nowosad/supercells package. Details are probably not important, but one idea in that package is that a user can call one of a few built-in distance functions (written in C++) or any distance function from the philentropy package. However, in the second case, the computations are much slower -- the values from C++ need to be sent to R, then philentropy does its magic (in C++), and then they are sent back to the C++ supercells. This makes a large overhead.

One possible improvement would be for the philentropy package to share its C++ code for linking with other R packages (e.g., see https://epiverse-trace.github.io/posts/share-cpp/index.html). To make it happen, the line // [[Rcpp::interfaces(cpp)]] needs to be added in C++, which forces Rcpp to autogenerate header files usable in C++ by other R packages.

Actually, the philentropy package already provides a few of its utility functions -- see

philentropy/src/utils.h

Line 7 in 9e171b2

// [[Rcpp::interfaces(r, cpp)]]

and then https://github.com/drostlab/philentropy/blob/master/inst/include/philentropy_RcppExports.h.

What do you think about exporting philentropy C++ functions to other R packages?

CC: @brownag -- I am also interested in your opinion on this.

[brownag]

This seems like it would work and be very helpful for some use cases.

I considered making the C++ API from https://github.com/drostlab/philentropy/blob/master/src/distances.h public in this way in the big refactor, but didn't go quite that far. I didn't really want to expand the public API surface at the same time as making big changes, in case it needed to be reorganized

I believe all of the wrappers around the internal distance functions include sanity checks (e.g. for NA values) so they should be safe to call without an R wrapper

[Nowosad]

👍🏻

To show how it works I made a small changes to the philentropy package (see #47) and then connected it with the new toy package -- see https://github.com/Nowosad/ruler.