Load all packages in tar_option_set()?

[malcolmbarrett]

Occasionally, it's useful to load all the packages that I'm using in tar_option_set() in my interactive session. Is there a good procedure for this? e.g. if I have tar_option_set(packages = c("ggplot2", "gt")), then load ggplot2 and gt in my local interactive session

[gorkang]

Not sure what is the best way to do it, but my approach has been:

1. Run targets::tar:renv() once: This creates _targets_packages() with all the packages in tar_option_set(packages)
2. Using shrtcts, I have a shortcut defined (CTRL+P) that points to .shrtcts.R where I have the following:

message("Loading _targets_packages.R and R/helper_functions.R")

if (file.exists("./_targets_packages.R")) suppressPackageStartupMessages(source("./_targets_packages.R"))
if (file.exists("./_targets.R")) source("./_targets.R")
if (file.exists("./R/helper_functions.R")) source("./R/helper_functions.R")

[wlandau]

Yeah, that sounds like a great workaround, @gorkang. Alternatively, Target Markdown accomplishes some of this. In interactive/notebook/prototyping mode, if you run a {targets ..., tar_globals = TRUE} chunk that contains the tar_option_set() call, then any subsequent {targets ..., tar_globals = FALSE} chunks with actual targets will load the packages into your session automatically for interactive prototyping.

[malcolmbarrett]

Thank you both for your responses. Unfortunately, I don't think either of these solutions is what I'm looking for. I'm looking for more of a spiritual sibling to tar_load(everything()) but to load whatever packages (or perhaps whatever prelude is in _targets.R) are needed. It really doesn't need to be efficient computationally the way targets normally operates with the run environment.

My perspective on this comes from beginning to teach targets in my R courses and workshops. For beginners, this is all too much overhead and too different from the mental model they have about how packages work in R. IMO, a simpler solution is really needed to help onboard people to this great tool, and I don't think that solution's design needs to be too complex. Happy to transfer this to a feature request and discuss potential solutions.

[wlandau]

This would require sourcing _targets.R in the interactive R session and then loading the packages. That would load the user's global objects and functions into memory, which is probably desirable from a pedagogical perspective.

tar_load_globals <- function(
  envir = parent.frame(),
  script = targets::tar_config_get("script")
) {
  force(envir)
  source(script, local = envir)
  lapply(
    X = tar_option_get("packages"),
    FUN = library,
    character.only = TRUE,
    lib.loc = tar_option_get("library")
  )
  invisible()
}

This should be safe since targets defaults to a new reproducible R process for all its serious work. Happy to support this function.

[malcolmbarrett]

That is perfect! Exactly what I had in mind, and I think it is good that it sources _targets.R since there may be other options or files sourced there.

[wlandau]

Now implemented in dev.

[malcolmbarrett]

Thanks to @wlandau, the best answer to this question is now to use tar_load_globals(): 460e25e