fix(cran): v3.1.1 — clear gcc-UBSAN additional issue (varPro test skip + vignette precompute)

[ehrlinger]

Why

CRAN flagged a gcc-UBSAN additional issue on ggRandomForests 3.1.0 (Ripley, 2026-06-12; correct before 2026-06-29):

entry.c:184:55: runtime error: pointer index expression with base
0x000000000001 overflowed to 0xfffffffffffffff9
    #0 rfsrcGrow .../randomForestSRC/src/entry.c:184

Root cause is upstream, in randomForestSRC (not us): rfsrcGrow does RF_yWeight = REAL(yWeight); RF_yWeight-- unconditionally. When yvar.wt arrives length-0 (the unsupervised family path every varPro rule-grow uses), REAL() returns R's 0x1 empty-vector sentinel and the -- forms an out-of-bounds pointer — UB that UBSAN flags (the pointer is never dereferenced). ggRandomForests is pure R (NeedsCompilation: no); it only surfaces the path by exercising varPro. Upstream maintainers are aware and staging a fix; root cause + suggested patch in dev/randomForestSRC-ubsan-report.md.

What this PR does

1. Tests: the varPro test fixtures now skip_on_cran(), so they don't run on CRAN's check machines (incl. gcc-UBSAN). The memtest that flagged 3.1.0 ran only the test suite, so this resolves the reported issue.
2. Vignette: the varpro vignette grew ~10 varPro forests live (varpro/uvarpro/isopro/ivarpro/beta.varpro). All are now precomputed into vignettes/varpro_precomputed.rds and loaded with a live fallback, so R CMD check performs no live varPro grow anywhere. Unused embedded forests ($rf, $isoforest, redundant ivarpro attrs) are stripped before saving — validated that every gg_* wrapper call returns output identical to the un-stripped object — keeping the file at 414 KB and the tarball at 4.13 MB (< CRAN's 5 MB).
3. Docs: NEWS.md + cran-comments.md updated to describe both changes.

Verification

• R CMD check --as-cran with manual: 0 errors, 0 warnings, 1 NOTE (days-since-update, expected and pre-explained in cran-comments).
• examples / --run-donttest / tests / vignette rebuild: all OK.
• Overall check time 2:43 (< 10-min budget); tarball 4.13 MB; urlchecker clean; 0 reverse deps; DESCRIPTION ↔ NEWS both 3.1.1.

Residual (intentional)

The .Rd varPro examples remain \donttest{} + requireNamespace()-guarded (the standard idiom). They run only under --run-donttest, not on normal CRAN checks, and were not the memtest trigger.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.04%. Comparing base (0802143) to head (7b792c0).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #119   +/-   ##
=======================================
  Coverage   87.04%   87.04%           
=======================================
  Files          44       44           
  Lines        3938     3938           
=======================================
  Hits         3428     3428           
  Misses        510      510           

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR prepares ggRandomForests v3.1.1 for CRAN by avoiding an upstream randomForestSRC gcc-UBSAN “additional issue” that is surfaced when ggRandomForests exercises varPro forest growth during checks.

Changes:

• Skip varPro-growing tests on CRAN via testthat::skip_on_cran().
• Precompute all varPro vignette fits into vignettes/varpro_precomputed.rds and load them with live fallbacks to avoid any live varPro grow during R CMD check.
• Update release metadata and CRAN-facing documentation (DESCRIPTION/NEWS/cran-comments) and add an upstream UBSAN report note under dev/.

Reviewed changes

Copilot reviewed 9 out of 10 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
vignettes/varpro.qmd	Loads precomputed varPro fits (with fallback) to avoid live forest growth during vignette rebuilds.
vignettes/precompute_varpro.R	Script to generate and trim the precomputed vignette objects saved to varpro_precomputed.rds.
tests/testthat/test_gg_varpro.R	Skips varPro-based gg_varpro tests on CRAN.
tests/testthat/test_gg_udependent.R	Skips varPro-based gg_udependent tests on CRAN.
tests/testthat/helper-varpro-fixtures.R	Skips varPro fixture generation on CRAN to avoid triggering upstream UBSAN.
NEWS.md	Bumps version to 3.1.1 and documents the CRAN workaround approach.
dev/randomForestSRC-ubsan-report.md	New developer note describing the upstream UBSAN issue and suggested patch.
DESCRIPTION	Version/date bump for the 3.1.1 release.
cran-comments.md	CRAN submission notes updated to explain the UBSAN workaround and test/vignette changes.

Comments suppressed due to low confidence (1)

tests/testthat/helper-varpro-fixtures.R:12

• Adding skip_on_cran() here will also skip these fixtures under CI R CMD check unless NOT_CRAN=true is set. With the current workflow env, gg_beta_varpro-related tests that rely on these helpers will no longer run in CI either.

.varpro_mtcars <- function() {
  testthat::skip_on_cran()
  if (is.null(.varpro_cache$v)) {
    if (!requireNamespace("varPro", quietly = TRUE)) {
      testthat::skip("varPro not installed")
    }

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ehrlinger]

Resolved the Copilot review threads — all addressed in commits after the review (which ran against b088769e):

• test_gg_varpro.R:9 / test_gg_udependent.R:9 (skip_on_cran also skips in CI) → 7b792c03: set NOT_CRAN=true in all check/coverage workflows, so the varPro tests now run in CI. Verified green on all platforms (Windows/macOS/Linux).
• NEWS.md / cran-comments.md ("still run on CI" inaccurate) → 7b792c03 (+8a249161): with NOT_CRAN=true the claim is true again; wording updated to "run in CI and locally; skipped only on CRAN's check machines".
• precompute_varpro.R strip comments (forests "never" used — survival C-path gg_partial_varpro() / gg_isopro(newdata=) do read $rf) → 8a249161: scoped the comments. Not a bug — the cached path never invokes those on a stripped object, and the live fallback grows full objects.

Also fixed the lint failure (d7c65af8): braced the vignette else fallbacks + multi-line .strip_* helpers. All 15 checks green.