Skip to content

cran-comments.md

Latest commit

 

History

History
64 lines (51 loc) · 3 KB

cran-comments.md

File metadata and controls

64 lines (51 loc) · 3 KB
Raw

Submission

This is a feature and bug-fix update (0.3.0), following the current CRAN version 0.2.0. It adds backward-compatible arguments to ggcorrplot() for a mixed per-triangle layout (lower.method/upper.method), cluster rectangles (hc.rect, hc.rect.col), built-in colorblind-safe palettes (palette), a one-token publication preset (preset), size-scaled squares (scale.square) and boxed cells (cell.grid, cell.grid.col), plus a new insig = "stars" value for a standalone significance map. Two vignettes are added.

Imports gains grDevices (a base R package, used for col2rgb()/rgb()).

Two changes are intentional and alter output for a small set of inputs; both are called out in NEWS.md:

  • The plot axis is now always drawn in matrix (row/column) order. This fixes hc.order = TRUE being silently ignored for matrices with numeric-looking variable names; as a consequence as.is no longer affects the plot and is kept only for backward compatibility.
  • show.diag = FALSE on a non-square matrix now removes only the genuine self-pairs rather than the positional diagonal.

Every other pre-existing argument combination produces byte-identical output to 0.2.0.

This follows 0.2.0 closer than usual because the two changes above correct cases where the plot was silently wrong rather than erroring: hc.order = TRUE was ignored for matrices with numeric-looking variable names, and show.diag = FALSE blanked the wrong cells of a non-square matrix. I am happy to hold the release if you would prefer the usual interval.

Test environments

  • local macOS, R 4.5.x
  • GitHub Actions: macOS-release, Windows-release, Ubuntu-{devel, release, oldrel-1}
  • win-builder release (R 4.x)
  • win-builder devel (R-devel)

R CMD check results

Local R CMD check --as-cran on the 0.3.0 tarball: 0 errors | 0 warnings | 0 notes. The same holds with _R_CHECK_DEPENDS_ONLY_=true. win-builder (release and R-devel) reported no errors and no warnings.

Some runs on the local machine add an "unable to verify current time" NOTE when the host cannot reach a time server; it is environmental and does not appear on CI.

Reverse dependencies

There are 22 reverse dependencies (15 Depends/Imports, 7 Suggests). The check was scoped to the behavior-changing surface, which is the two intentional changes listed above; everything else in this release is additive, behind new arguments whose defaults reproduce the 0.2.0 behavior.

I read the source of all 22 and extracted every ggcorrplot() / cor_pmat() call site:

  • No reverse dependency passes as.is, so the axis-order change reaches none of them.
  • Only two pass show.diag, both TRUE; the changed code path needs show.diag = FALSE on a non-square matrix, which no reverse dependency does.
  • All calls bind only corr positionally and name every other argument. The nine new arguments are appended at the end of the signature, no argument was removed or reordered, and the defaults of all 35 pre-existing arguments are unchanged.

No new problems were found.