Consider delaying data download

[nflverse]

Consider delaying data download

[Enchufa2]

If these downloads are not small, an improvement would be to cache them for the current session under R's temporary directory. In this way, it would be easier to enable an option so that the user can specify a permanent cache directory.

Alternatively, tools::R_user_dir returns an appropriate directory in every platform to cache data permanently. This is essentially the same functionality as the rappdirs package provides, which was ported to R.

[guga31bb]

Yes, I realize this is annoying, sorry. The reason for this is that CRAN has a very strict package size limit that the models included in nfl4th exceed, so the package has to load these models externally and there's no way to make it work without internet connectivity.

An option for not using the package itself is just getting the pre-computed data here.

[Enchufa2]

Yes, I know the CRAN policy. My suggestion would be to provide a function that the user needs to call in order to download that data and place it under the user dir, instead of doing that on package load. But yet again, that's a suggestion. :)

The reason for this request is that I have an RPM repository for Fedora with all (most) CRAN packages and this one cannot be built due to this issue. For now, I had to remove nfl4th and nflverse (which depends on nfl4th) for now.

If you decide to address this, please let me know to add them back.

[mrcaseb]

Instead of loading the files on load we could add something like this to the package and call it from inside the other functions

init_nfl4th <- function() {
  pkg_env <- ls("package:nfl4th")
  
  if (!".games_nfl4th" %in% pkg_env){
    .games_nfl4th <- get_games_file()
    assign(".games_nfl4th", .games_nfl4th, envir = parent.env(environment()))
  }
  
  if (!"fd_model" %in% pkg_env){
    fd_model <- load_fd_model()
    assign("fd_model", fd_model, envir = parent.env(environment()))
  }
}