[krlmlr]

Commit dcfe7b3 (eigen_centrality() stress test) moves to main in #2997.

Draft, do not merge: review vehicle for a fix collected for the series rebuild on main-fwd-build. Merging would move the main-green mirror, which the series loop owns.

This is the third of three stacked review vehicles. It is based on claude/f1-leiden-directed (#2991), which is itself based on claude/f2-lad-time-limit (#2990). It was rebased onto the rebased claude/f1-leiden-directed without conflicts, so its diff shows only the nits.

What it changes

The branch carries six commits, one per review nit.

• cluster_edge_betweenness() now raises the weighted-run warnings from R with cli::cli_warn() instead of from the glue with IGRAPH_WARNING(), which reported the glue file and line as their source. As in igraph 0.10, only the first warning is raised, and the typo "betwenness" is fixed.
• The six *_dijkstra_impl() aliases are removed, because upstream merged the weighted _dijkstra variants into the base functions and nothing in the package calls them. The generated-interface tests in test-aaa-auto.R that called four of them now call the base functions, and the one duplicating pseudo_diameter_impl() is gone.
• The stale igraph_rewire mode default is dropped from the Stimulus functions-R.yaml, since igraph_rewire() no longer has a mode parameter. Regenerating the interface leaves its output byte-identical.
• The redundant igraph_personalized_pagerank_vs DEPS override is dropped from the same file, since upstream's functions.yaml declares the same dependencies. Regenerating the interface leaves its output byte-identical.
• hits_scores(scale) carries the superseded badge in its documentation and says what FALSE still does, matching the lifecycle signal the function already raises.
• The eigen_centrality() stress test expects the near-zero warning on disconnected random graphs and muffles only that warning, instead of emitting hundreds of warnings on a passing test.

Proposed correction for series commit a7a553a

This is not a commit on this branch. The subsession that made these nits proposed the following corrected message for series commit a7a553a:

R-side fix: upstream renames the vertex_weights argument of igraph_community_leiden() to vertex_out_weights and adds vertex_in_weights, which must be NULL for undirected graphs. Its override keeps the R name vertex_weights and always passes NULL in-weights, so the arguments of community_leiden_impl() and cluster_leiden() are unchanged. Upstream also drops the error for directed graphs, so cluster_leiden() now runs on a directed graph instead of failing.

Note that PR 2991 (#2991, the cluster_leiden() fix) reverses that last sentence: with it, cluster_leiden() rejects directed graphs again. The rebuild must therefore either adjust the last sentence or fold that fix into this commit.

Tests

The package was built and installed from this branch tip, which includes both branches below it, into a scratch library, and the test files touched by the three branches were run against the installed package with NOT_CRAN=true. All have 0 failures and 0 errors.

• test-aaa-auto.R: 712 tests, 1313 expectations, 0 skips, 1 warning (from the hub_and_authority_scores_impl basic test).
• test-centrality.R: 44 tests, 2904 expectations, 0 skips.
• test-community.R: 37 tests, 355 expectations, 0 skips.
• test-topology.R: 45 tests, 626 expectations, 1 skip (an existing FIXME about errors in the callback being gobbled).

air format --check is clean on all changed R files.

Provenance

Made by a Claude subsession: https://claude.ai/code/session_01L5n6BLiYoPmk3NFiYDSKZZ

---

Generated by Claude Code