fix: An empty argument slot with a name is not an ambiguous argument

[krlmlr]

as_adjacency_matrix(make_ring(4), a = )

• CRAN 2.3.3: a 4×4 matrix
• main: Error: Argument amatches multiple arguments ofas_adjacency_matrix().

#2812 taught the migration blocks to handle empty slots by asking base::missing() inside .old_signature() — which is right, and is what fixed the four #2646 revdeps. But two guards run before that closure is called, and both read argument names without ever looking at the values:

• .arg_ambiguous, from names(substitute(...())), rejects an abbreviation that could mean two different arguments — 20+ functions.
• .arg_forbidden, from names(sys.call()), rejects a tag that is a prefix of both a head formal and a recoverable one.

An empty slot supplies nothing, so it cannot be ambiguous between two arguments and cannot conflict with a formal. Under the old signature it matched by position and left the formal missing.

The change

Both guards drop empty slots before looking at names. The empty slot is the empty symbol — a symbol whose name is the empty string — and Filter() keeps the test to one expression, which matters because the block is # fmt: skip and every line of it is already long:

.arg_ambiguous <- base::intersect(base::names(base::Filter(function(.x) !(base::is.symbol(.x) && !base::nzchar(base::as.character(.x))), base::as.list(base::substitute(...())))), base::c("a", "at", "att"))

228 generated blocks, no line-count growth.

Verified against CRAN

	2.3.3	main	this branch
as_adjacency_matrix(make_ring(4), a = )	4×4	error	4×4
as_adjacency_matrix(make_ring(4), sparse = ,)	4×4	4×4	4×4
as_adjacency_matrix(make_ring(4), )	4×4	4×4	4×4
as_adjacency_matrix(make_ring(4), a = 1)	error	error	error

The last row is the point: a value in the same slot is still rejected.

Tests

The empty-slot tests #2812 added only ever exercised migration_fixture, which carries neither guard — which is exactly why this survived. New tests cover both fixtures that do: migration_fixture_shadow for .arg_ambiguous, migration_fixture_prefix for .arg_forbidden, each with the empty slot accepted and a value rejected (snapshotted).

One more test pins #2646's two reproducers verbatim against real migrated functions rather than the fixture, plus the named-empty case on as_adjacency_matrix().

Full testthat::test_local() green apart from test-foreign.R:69, which downloads from github.com/igraph/graphsdb. Rscript tools/generate-migrations.R is idempotent on the committed tree, so the CI drift check passes.

🤖 Generated with Claude Code

https://claude.ai/code/session_01D1xpHRV7yVfgtJg4vp9P7z

---

Generated by Claude Code

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 8c47c15 is merged into main:

• ✔️as_adjacency_matrix: 735ms -> 737ms [-0.38%, +0.77%]
• ✔️as_biadjacency_matrix: 741ms -> 740ms [-0.58%, +0.45%]
• ✔️as_data_frame_both: 1.59ms -> 1.59ms [-1.12%, +1.57%]
• ✔️as_long_data_frame: 4ms -> 4ms [-0.93%, +0.85%]
• ✔️es_attr_filter: 2.68ms -> 2.68ms [-1.18%, +1.08%]
• ✔️graph_from_adjacency_matrix: 115ms -> 115ms [-1.42%, +0.6%]
• ✔️graph_from_data_frame: 3.68ms -> 3.69ms [-0.56%, +1.4%]
• ✔️vs_attr_filter: 1.51ms -> 1.51ms [-1.76%, +1.29%]
• ✔️vs_by_name: 971µs -> 965µs [-1.62%, +0.48%]
  Further explanation regarding interpretation and methodology can be found in the documentation.