I'm working on an R package for AT/Bluesky, and I'm trying out a sort of unconventional structure. Since the Bluesky methods are documented in very precise detail via the JSON lexicon defined here, I've copied these files into the /inst directory, so my package structure looks like this:
├── DESCRIPTION
├── LICENSE
├── LICENSE.md
├── NAMESPACE
├── R
│ ├── bluesky.R
│ ├── helpers.R
│ ├── lexicon.R
│ └── types.R
├── README.Rmd
├── README.md
├── blueRsky.Rproj
├── inst
│ └── lexicons
│ ├── app
│ │ └── bsky
│ │ ├── actor
│ │ │ ├── defs.json
│ │ │ ├── getProfile.json
│ │ │ ├── getProfiles.json
│ │ │ ├── getSuggestions.json
│ │ │ ├── profile.json
│ │ │ ├── searchActors.json
│ │ │ └── searchActorsTypeahead.json
│ │ ├── embed
│ │ │ ├── external.json
│ │ │ ├── images.json
│ │ │ ├── record.json
│ │ │ └── recordWithMedia.json
│ │ ├── feed
│ │ │ ├── defs.json
│ │ │ ├── getAuthorFeed.json
│ │ │ ├── getLikes.json
│ │ │ ├── getPostThread.json
│ │ │ ├── getPosts.json
│ │ │ ├── getRepostedBy.json
│ │ │ ├── getTimeline.json
│ │ │ ├── like.json
│ │ │ ├── post.json
│ │ │ └── repost.json
│ │ ├── graph
│ │ │ ├── follow.json
│ │ │ ├── getFollowers.json
│ │ │ ├── getFollows.json
│ │ │ ├── getMutes.json
│ │ │ ├── muteActor.json
│ │ │ └── unmuteActor.json
│ │ ├── notification
│ │ │ ├── getUnreadCount.json
│ │ │ ├── listNotifications.json
│ │ │ └── updateSeen.json
│ │ ├── richtext
│ │ │ └── facet.json
│ │ └── unspecced
│ │ └── getPopular.json
│ └── com
├── man
│ ├── bsky_get_profile.Rd
│ ├── bsky_get_session.Rd
│ ├── bsky_get_timeline.Rd
│ └── bsky_search_actors.Rd
└── tests
├── testthat
│ └── test-lexicon.R
└── testthat.R
I have functions that then look up the Lexicon schemas, for example
load_schema <- function(id) {
loc <- c("lexicons", strsplit(id, "\\.")[[1]]) |>
paste0(collapse = "/") |>
paste0(".json")
file_path <- system.file(loc, package = "blueRsky", mustWork = FALSE)
cat("file_path:", file_path, "\n")
if (!file.exists(file_path)) {
stop(paste("Schema", id, "not found"))
}
jsonlite::read_json(system.file(loc, package = "blueRsky", mustWork = TRUE))
}
This is very nice, because it allows me to abstract a lot of boilerplate for constructing the requests.
This works great locally, I can install the package, and all my tests pass, but it fails R CMD CHECK. I believe the reason is that R CMD CHECK is removing some of the files from the lexicons directory. The output says:
── R CMD build ──────────────────────────────────────────────────────────────────
✔ checking for file ‘/Users/colinfraser/projects/blueRsky/DESCRIPTION’ ...
─ preparing ‘blueRsky’: (1.1s)
✔ checking DESCRIPTION meta-information ...
─ checking for LF line-endings in source and make files and shell scripts (346ms)
─ checking for empty or unneeded directories
Removed empty directory ‘blueRsky/inst/lexicons/com’
─ building ‘blueRsky_0.0.0.9000.tar.gz’
Removed empty directory ‘blueRsky/inst/lexicons/com’.
That's not good, because in fact that directory is not empty, it just has nested folders in it. Then later on it fails because it can't load the package code.
Is there anything I can do to stop R CMD CHECK from deleting this file? Or could there be something else that's wrong here? I know that this is a bit of an unusual package structure, but it seems like it would work really well if I could get it to work.
R removed the directory inst/com/ because at the time of package building it was seen to be empty. R cannot know you plan to store files there later; it just sees an empty directory and aims to clean things up to make a neater package without any fluff.
The fix against this suggests itself: put a file into the directory, maybe inst/com/README.md, or, more minimally a zero-byte placeholder via
$ touch inst/com/.keep.this.directory
assuming your OS does have a touch binary.