test: Move the last file-local test fixtures into helpers, with self-describing names

[krlmlr]

Closes #2793.

Prepared with Claude Code.

Fixtures out of test files

Seven fixture functions still sat at the top level of test files, and three fixtures in helper-test-functions.R were plain objects rather than functions. Both shapes make a test_that() block depend on code that ran before it: the block cannot be run on its own, and — for the shared objects — one block can mutate what the next one sees.

The functions move to helper-test-functions.R, each under a section naming the test file that uses it, and the objects become functions returning a fresh value per call:

• make_weighted_clique_graph() from test-cliques.R
• tarjan_yannakakis_graph() from test-decomposition.R
• load_migration_generator() and write_migration_registry() from test-generate-migrations.R
• make_weighted_graphlet_graph() from test-glet.R
• hrg_graph_and_fit() from test-hrg.R
• simple_graph_with_edge_attrs() from test-operators.R
• unique_vertex_seq_cases(), karate_oldstyle_graph() and karate_oldstyle_vertex_names(), previously top-level objects in the helper file

Each test file keeps a one-line comment pointing at the helper file, so the fixture is still discoverable from where it is used. The one snapshot that printed a fixture (V(karate_oldstyle) in _snaps/other.md) is updated to the call form.

Scanning all 73 files under tests/testthat/ confirms no test file defines a top-level function or object any more. What remains at the top level is only setup code in the files meant for it — setup.R, setup-lifecycle.R, helper-migrations.R — plus the file-level skip_on_cran() in test-aaa-auto.R, which is neither a function nor an object and whose semantics depend on running at file scope.

Names that read out of context

These helpers all share one environment, so a name has to be unambiguous far from its definition. Several were not — unique_tests() named neither what was unique nor that it held cases rather than tests, and rn(), sortgl() and threshold.net() said almost nothing. Renamed, with no change in behaviour:

• unique_tests() → unique_vertex_seq_cases()
• rn() → name_rows_by_edge_endpoints()
• order_by_two_first_columns() → sort_edgelist_rows() (it sorts rows, it does not return an ordering)
• simple_graph_with_attrs() → simple_graph_with_edge_attrs()
• order_by_magnitude() / sort_by_magnitude() → order_eigenvalues_by_magnitude() / sort_eigenvalues_by_magnitude()
• make_graphlet_graph() → make_weighted_graphlet_graph()
• sortgl() → sort_graphlets_by_size()
• threshold.net() → max_cliques_above_weight()
• graphlets.old() / graphlets.project.old() → graphlet_basis_reference() / graphlet_proj_reference()
• karate_oldstyle() → karate_oldstyle_graph(), karate_oldstyle_names() → karate_oldstyle_vertex_names()
• local_generator() → load_migration_generator(), write_registry() → write_migration_registry()
• hrg_fixture() → hrg_graph_and_fit()

load_migration_generator() also loses its env argument, which was never read and only carried the local_ prefix's implication of withr-style scoping. Helpers whose names were already unambiguous keep them; the four that were only terse gain a comment saying what they return.

Verification

Tests for the touched files pass locally. The one failure in test-other.R (can create graphs when igraph is not attached) is unrelated: it spawns a callr subprocess that needs igraph installed, which this sandbox does not have.

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

🤖 Generated with Claude Code

https://claude.ai/code/session_01GkfBDoTdJ8qVNAqRCvBpbn