Skip to content

cran-comments.md

Latest commit

 

History

History
76 lines (62 loc) · 3.54 KB

cran-comments.md

File metadata and controls

76 lines (62 loc) · 3.54 KB
Raw

CRAN comments

Resubmission

CRAN's incoming pretest (2026-08-21) flagged 1 NOTE on Windows and 2 on Debian:

  • Possibly misspelled word "conda" in DESCRIPTION -- correct as spelled, see below. No change made.
  • Debian only: tests NOTE, "Running R code in 'testthat.R' had CPU time 3.4 times elapsed time." Root cause: tests/testthat/test-setup.R calls carbon_tracker_ready() unguarded (deliberately -- it verifies graceful behavior with no Python configured, which is exactly CRAN's situation), and that call forces reticulate's interpreter-discovery machinery to run, which can spin up background threads during the search on a machine with no Python at all. Not this package's own test code doing anything improper. Fixed by adding testthat::skip_on_cran() to that one test -- it still runs everywhere else (local development, CI on Windows/Mac/Linux via GitHub Actions), just not on CRAN's check machines specifically, since eliminating a policy-adjacent NOTE deterministically is preferable to relying on a reviewer accepting an explanation for something a one-line skip resolves cleanly.

Submission

This is a new package. The previous submission attempt was blocked by CRAN's automated incoming pretest before reaching human review; see Resubmission above for what changed.

CodeCarbonR wraps the Python codecarbon package (via reticulate) to measure the energy consumption and estimated carbon emissions of R code. There is no R-native equivalent of codecarbon, which is why this package depends on a Python backend rather than reimplementing emissions tracking natively.

Notes on Python/conda dependency

  • setup_carbon_tracker() is the only function that installs software. It is opt-in (never called automatically), requires explicit user confirmation via utils::menu(), and installs codecarbon into its own dedicated conda environment (r-codecarbon) rather than touching the user's system Python or any existing R/Python installation.
  • setup_carbon_tracker() refuses to run in a non-interactive session (stop()s immediately if !interactive()), so it never runs during R CMD check, and this refusal is itself covered by a test (tests/testthat/test-setup.R).
  • No example, test, or vignette installs software, writes outside tempdir(), or requires network access during R CMD check. Examples and tests that would otherwise touch Python/codecarbon are wrapped in \donttest{}/\dontrun{} or guarded with testthat::skip_on_cran() plus a runtime carbon_tracker_ready() check, so they no-op cleanly on CRAN's check machines (which will not have codecarbon installed).
  • vignettes/quickstart.Rmd uses eval = FALSE on every chunk and shows realistic example output as literal text, so building it never touches Python or the network.

R CMD check results

0 errors | 0 warnings | 0 notes with R CMD check --as-cran on this machine (Windows; see RELEASING.md for details of this package's release process).

check_win_devel() (R-devel, Windows) additionally reported 2 NOTEs:

  • Possibly misspelled word "conda" in DESCRIPTION -- this is correct as spelled; conda is the name of the package manager CodeCarbonR's Python backend is installed into, not a typo.
  • The carbon_tracker_ready() example exceeded the 10s check threshold on that machine, because it has no Python configured and reticulate's interpreter discovery is slow in that case. Fixed by wrapping the example in \donttest{}.

Downstream dependencies

This is a new package; there are no reverse dependencies.