Prompt before installing python dependencies

[JosiahParry]

I wanted to give ragnar a shot. Below is my experience. I was first of all surprised to see that it installed python, downloaded ~100mb of dependencies, and didn't tell me where it did so.

As a request I would ask that ragnar:

1. asks before installing python and dependencies,
2. that it report where these are being installed,
3. l'd ask that ragnar expose a function to identify the python environment location so that one can remove it if wanted

> pak::pak("ragnar")
                                      
→ Will install 1 package.
→ Will download 1 package with unknown size.
+ ragnar   0.2.1 🔧 ⬇
ℹ Getting 1 pkg with unknown size
✔ Got ragnar 0.2.1 (aarch64-apple-darwin20) (3.14 MB)
✔ Downloaded 1 package (3.14 MB) in 637ms
✔ Installed ragnar 0.2.1  (55ms)
✔ 1 pkg + 45 deps: kept 37, added 1, dld 1 (3.14 MB) [1.9s]
> ragnar::read_as_markdown(doc_urls[1])
Downloading cpython-3.12.12-macos-aarch64-none (download) (16.2MiB)
 Downloaded cpython-3.12.12-macos-aarch64-none (download)
Downloading jedi (1.5MiB)
Downloading numpy (4.9MiB)
Downloading sympy (6.0MiB)
Downloading cryptography (6.9MiB)
Downloading magika (12.7MiB)
Downloading pdfminer-six (5.4MiB)
Downloading pygments (1.2MiB)
Downloading youtube-transcript-api (2.1MiB)
Downloading pandas (10.2MiB)
Downloading debugpy (2.4MiB)
Downloading pillow (4.4MiB)
Downloading lxml (8.3MiB)
Downloading onnxruntime (16.4MiB)
Downloading pyzmq (1.2MiB)
Downloading speechrecognition (31.3MiB)
 Downloaded pyzmq
 Downloaded pygments
 Downloaded jedi
 Downloaded youtube-transcript-api
 Downloaded debugpy
 Downloaded pillow
 Downloaded numpy
 Downloaded pdfminer-six
 Downloaded sympy
 Downloaded cryptography
 Downloaded lxml
 Downloaded pandas
 Downloaded magika
 Downloaded onnxruntime
 Downloaded speechrecognition
Installed 79 packages in 348ms

[JosiahParry]

somewhat related issue: #142

[t-kalinowski]

Thanks for trying ragnar and taking the time to write up the issue!

Totally fair feedback — sorry the first-run experience was surprising.

ragnar’s Python usage is an internal implementation detail: only read_as_markdown() and ragnar_find_links() currently use Python, via reticulate::py_require(). On first use of those functions, reticulate may auto-download a standalone CPython build plus Python wheels. This doesn’t touch your system Python; everything goes into reticulate’s managed cache, which is designed to be self-cleaning over time.

If you want to see or remove what was installed:

# Cache location
tools::R_user_dir("reticulate", "cache")

# Exact Python binary reticulate is using (after first run)
reticulate::py_config()$python

# Remove cached Python + packages
unlink(tools::R_user_dir("reticulate", "cache"), recursive = TRUE)

On your requests:

• A confirmation prompt would need to live in reticulate itself. Adding one there would introduce noticeable friction, and it’s not something I’m planning to pursue as a default. We’ve put a lot of effort into making py_require() “just work,” in part to undo reticulate’s earlier reputation for being hard to set up, and a mandatory prompt would cut against that.

• I agree we should be clearer about the first-run downloads and where they land; I’ll add a short note to the docs/README pointing out that only read_as_markdown() / ragnar_find_links() trigger this, along with the cache location and cleanup commands.

• I’m hesitant to add a ragnar helper that “exposes the env path,” since that would make an internal detail feel like supported API. The reticulate helpers above are the recommended way to inspect or clean things up.

(Separately, we may be able to move reticulate to Suggests so users who only use the pure‑R parts don’t even need it installed, and gate the reticulate-backed functions with a clear message when they’re called.)

[JosiahParry]

Thank you! Yes, it worked very well which was great!

For your first 2 points, I think adding something like the following to the .onLoad() or the read_as_markdown() / ragnar_find_links() wouldn't be a terrible experience! In fact, maybe even a nice one, imo!

prompt_reticulate <- function() {
  if (!reticulate::py_available()) {
    cli::cli_inform(
      c(
        "i" = "{.pkg ragnar} requires a python installtion managed by {.pkg renv}.",
        "!" = "Continue to install python via {.fn reticulate::py_require}?"
      )
    )
    accept <- menu(c("yes", "no"))

    if (!identical(accept, 1L)) {
      cli::cli_abort("Please install python via {.code reticulate::py_available(initialize = TRUE)}")
    }
  }
}

For the last point, if the docs could provide reticulate related functions to discover the environment / cache would be great! I can't find it with either reticulate::conda_list() or reticulate::virtualenv_list(). I suspect it's related to the new (awesome) uv features and related to the UV_PYTHON_INSTALL_DIR if my sleuthing is correct!

> reticulate:::uv_python_list
function (uv = uv_binary(), python_preference = Sys.getenv("UV_PYTHON_PREFERENCE", 
    "only-managed")) 
{
    if (isTRUE(attr(uv, "reticulate-managed", TRUE))) 
        withr::local_envvar(c(UV_CACHE_DIR = reticulate_cache_dir("uv", 
            "cache"), UV_PYTHON_INSTALL_DIR = reticulate_cache_dir("uv", 
            "python")))