Skip to content

cran-comments.md

Latest commit

 

History

History
78 lines (64 loc) · 3.51 KB

cran-comments.md

File metadata and controls

78 lines (64 loc) · 3.51 KB
Raw

Resubmission

This is a resubmission of 0.2.0. The four points raised in review are addressed:

  • Software names are single-quoted in the Description ('Python', 'Stable Diffusion', 'Hugging Face', with a URL for the last).
  • The doubled spaces are gone. They came from trailing whitespace on every continuation line, which DCF folding turned into two spaces.
  • \value is now present on all 50 exported .Rd files that were missing it, describing the class and meaning of the result. (Topic pages that document no function, and therefore have no \usage, do not have one.)
  • 14 of the 23 \dontrun{} examples were unwrapped and now run during check; they were also rewritten to be self-contained rather than referring to objects that were never defined. The 10 that remain are listed below with the reason for each.

Test environments

  • Ubuntu 24.04 (local), R 4.6.x: R CMD check --as-cran
  • Windows 10, R 4.6.0 and R-devel: R CMD check --as-cran, with torch's lantern backend installed (so the full suite runs)
  • Ubuntu, with lantern deliberately absent, to exercise the path win-builder and CRAN take: examples and tests skip gracefully rather than erroring
  • win-builder, R-devel

R CMD check results

0 errors | 0 warnings | 1 note

  • New submission (checking CRAN incoming feasibility).

The local Linux check additionally shows a NOTE from the torch package's own load hook in check subprocesses ("could not find function packageName"); it does not reproduce on any Windows check or win-builder and is attributable to the local torch build, not this package.

Notes for the reviewers

  • Ten examples remain under \dontrun{}. Each needs multi-GB model weights on disk, so none can execute on a check machine:

    • download_model(), download_component() - network downloads of multi-GB weights.
    • load_pipeline(), load_model_component() - require those downloaded weights.
    • txt2img(), txt2img_sd21(), txt2img_sdxl() - full image generation; needs the weights and minutes of compute.
    • ltx23_open_checkpoint() - needs a ~20 GB LTX-2.3 checkpoint the user downloads under Lightricks' own license.
    • resident_load() - loads a full pipeline onto a GPU.
    • load_decoder_weights() in the vae_decoder_native() page - needs a checkpoint file. The rest of that example runs.

    \donttest{} would not help for these: CRAN runs \donttest{} examples, and each would either attempt a large download or fail on a missing file.

  • save_video()'s MP4 example is deliberately shown as a comment rather than run. Both encoder backends hand off to an ffmpeg process that inherits the session's stdin, which R CMD check --as-cran uses to feed the example script to R, so a live call consumes part of the script and later examples parse short. The frame-writing path is exercised live instead.

  • Model weights are never bundled and never downloaded without consent: every download function prompts interactively with the size stated, and non-interactive sessions require options(diffuseR.consent = TRUE). Downloaded source weights use hfhub's user cache; locally derived quantized artifacts are stored under tools::R_user_dir("diffuseR", "data").

  • torch is in Imports; all tests, examples, and vignette code degrade gracefully when torch's backend (lantern) is not installed, as on win-builder. Note that torch::cuda_is_available() raises an error rather than returning FALSE in that state, so the package probes it through tryCatch() everywhere it is reachable without a GPU.