fix: heal main — unleak warnPartialMatchArgs on R 4.2 and name args at norm_coords()/make_star() call sites for the lifecycle-errors leg

[krlmlr]

Prepared with Claude Code (Claude Fable 5).

• By submitting this pull request, I assign the copyright of my contribution to The igraph development team.

---

main has been red since 17:36 UTC today,
after #2787 (guard-emitting migrations generator)
and the ellipsis-refactor train merged close together.
The diagnosis found four independent breakages;
the first was fixed by #2798 while this PR was in preparation
(it carries the identical regeneration this branch originally produced,
so that commit dropped out on rebase),
and this PR delivers the remaining three.

1. (landed via #2798) rcc-smoke migrations drift check

For the record, every main push from 17:36 to 19:04 UTC failed
"Smoke test: stock R" at Run ./.github/workflows/custom/after-install:

##[error]Generated ARG_HANDLE blocks are out of sync with the registry (tools/migrations/).
Run 'Rscript tools/generate-migrations.R' and commit the result.

because the refactor PRs had regenerated their ARG_HANDLE blocks
on branches cut before #2787's generator learned to emit
migrate_check_call_tags() guards.
#2798 committed the regenerated blocks;
on this branch Rscript tools/generate-migrations.R
is a verified no-op on R/.

2. rcc: ubuntu-26.04 (4.2) rewrites the migration-fixture snapshot

The runs whose smoke test still passed
(e.g. #2787's own merge run, 30289034966)
failed only in the R 4.2 leg,
at Run ./.github/workflows/update-snapshots,
which keeps force-pushing snapshot PR #2796 and then fails via false.
The recorded difference, in
tests/testthat/test-migration-fixture.R >
test_that("forbidden prefixes error only when legacy arguments engage recovery"):

     Code
       migration_fixture_prefix(c(2, 2), 0.5, di = 2)
     Condition
+      Warning in `migration_fixture_prefix()`:
+      partial argument match of 'di' to 'dimvector'
       Error in `migration_fixture_prefix()`:
       ! Argument `di` matches multiple formal arguments of `migration_fixture_prefix()`.

Root cause: an earlier test in the same file enables
warnPartialMatchArgs via rlang::local_options().
On R < 4.3, restoring that option to NULL (unset)
does not switch the warning off again —
the internal flag keeps its last set value (fixed in R 4.3) —
so it leaks into every later test in the file, on R 4.2 only.
Verified on stock R 4.2.3 (via rig):
set TRUE → restore to NULL still warns, while R 4.5.3 does not.
Beyond the snapshot, the leak would also fail the
expect_no_condition() test two blocks later on R 4.2,
so a skip alone would not heal the leg.

Fix (two layers):

• pin options(warnPartialMatchArgs = FALSE) before flipping it on,
  so the scoped restore lands on FALSE,
  which resets the flag on every R version;
• per maintainer decision, additionally skip the affected snapshot test
  on R 4.2 (skip_if(getRversion() < "4.3")),
  mirroring the existing version-gated skips in
  test-adjacency.R and test-community.R.
  (A sweep of every failed rcc run across all branches since 2026-07-18
  confirmed this is the only snapshot that ever diverges on the 4.2 leg;
  the local suite runs R 4.5.3, so the skip is exercised only in CI.)

3. The never-yet-run "with lifecycle errors" leg would fail next

#2782 added a matrix leg (ubuntu-26.04, R release,
IGRAPH_LIFECYCLE_ERRORS=true) that fails any test
in which an unasserted lifecycle deprecation fires.
It has not executed yet — the smoke gate broke first —
but once smoke passes it will,
and the refactors that merged after #2782 was cut
left exactly two unasserted call sites
(all 36 deprecation warnings of the default suite reduce to them):

• plot.igraph() itself called
  norm_coords(layout, -1, 1, -1, 1) positionally (R/plot.R) —
  32 warnings across the plotting tests;
• make_star(5, "undirected") with positional mode
  in test-stochastic_matrix.R and test-centralization.R — 5 warnings.

Fix: name the arguments at both sites
(norm_coords(layout, xmin = -1, xmax = 1, ymin = -1, ymax = 1),
make_star(5, mode = "undirected")).

The matrix plumbing itself is sound:
a faithful offline simulation of
.github/workflows/versions-matrix/action.R as of #2798
(stubbing only the svn fetch)
sources .github/versions-matrix.R cleanly
and appends the leg
{"os":"ubuntu-26.04","r":"4.6","env":"IGRAPH_LIFECYCLE_ERRORS=true","desc":"with lifecycle errors"}
to matrix JSON that jsonlite parses.

Validation (all on this branch, R 4.5.3)

• Rscript tools/generate-migrations.R is a no-op on R/
  (three consecutive runs; git status --porcelain R/ stays empty),
• air format R/ tools/ tests/ is a no-op,
• devtools::document() leaves man/ and NAMESPACE untouched,
• migration-focused tests
  (filter = "migration|generate-migrations|constant-defaults"):
  FAIL 0 | WARN 0 | SKIP 0 | PASS 83,
• full suite: FAIL 1 | WARN 36 | SKIP 8 | PASS 9181 before the
  call-site fixes,
• full suite with IGRAPH_LIFECYCLE_ERRORS=true after them:
  FAIL 1 | WARN 0 | SKIP 8 | PASS 9181 —
  the single failure is the sandbox-only test-foreign.R
graph_from_graphdb() download
  (no network in the sandbox; CI can reach it).

---

Generated by Claude Code

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 6fcaca3 is merged into main:

• ✔️as_adjacency_matrix: 792ms -> 793ms [-0.33%, +0.45%]
• ✔️as_biadjacency_matrix: 775ms -> 775ms [-1.06%, +1.19%]
• ✔️as_data_frame_both: 1.55ms -> 1.54ms [-1.4%, +0.31%]
• ✔️as_long_data_frame: 3.92ms -> 3.93ms [-0.61%, +0.85%]
• ✔️es_attr_filter: 2.73ms -> 2.73ms [-0.8%, +0.76%]
• ✔️graph_from_adjacency_matrix: 139ms -> 141ms [-0.57%, +2.87%]
• ✔️graph_from_data_frame: 3.57ms -> 3.55ms [-2.02%, +0.61%]
• ✔️vs_attr_filter: 1.57ms -> 1.56ms [-1.64%, +0.31%]
• ✔️vs_by_name: 1.01ms -> 1ms [-1.99%, +0.23%]
  Further explanation regarding interpretation and methodology can be found in the documentation.