feat: Add attribute combination support to graph operators by schochastics · Pull Request #2676 · igraph/rigraph

schochastics · 2026-05-28T11:37:31Z

…tion, compose, disjoint_union)

schochastics · 2026-05-28T11:59:26Z

devtools::document() doesnt work locally for me

Run: https://github.com/igraph/rigraph/actions/runs/27070207213

github-actions · 2026-06-10T07:05:06Z

This is how benchmark results would change (along with a 95% confidence interval in relative change) if c26ab02 is merged into main:

✔️as_adjacency_matrix: 777ms -> 779ms [-0.27%, +0.78%]
✔️as_biadjacency_matrix: 777ms -> 779ms [-0.47%, +1.17%]
✔️as_data_frame_both: 1.61ms -> 1.59ms [-2.88%, +1.5%]
✔️as_long_data_frame: 4ms -> 4ms [-1.29%, +1.34%]
✔️es_attr_filter: 2.8ms -> 2.82ms [-1.01%, +2.48%]
✔️graph_from_adjacency_matrix: 147ms -> 148ms [-0.73%, +1.43%]
✔️graph_from_data_frame: 3.56ms -> 3.56ms [-1.83%, +1.32%]
✔️vs_attr_filter: 1.65ms -> 1.63ms [-3.3%, +0.25%]
✔️vs_by_name: 1.06ms -> 1.06ms [-2.31%, +2.11%]
Further explanation regarding interpretation and methodology can be found in the documentation.

…lper Address review feedback on #2676: - Rename the new graph.attr.comb/vertex.attr.comb/edge.attr.comb arguments of union()/intersection()/compose()/disjoint_union() to snake_case (these args are new in this PR, so no deprecation is needed). - Extract rename_attr_if_needed() back out of combine.attrs() for the "rename" strategy, preserving the historical overwrite-on-clash semantics under chains of %du%. - Add a clarifying comment on the backward-compatible "rename" default. - Move the make_named_pair() test helper into helper.R. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…tted names Follow-up to review feedback on #2676 ("clean up edge.attr.comb"): - simplify(), as_undirected() and contract() gain snake_case edge_attr_comb / vertex_attr_comb arguments via the argument-migration infrastructure (tools/migrations.R). The old dotted names still work and emit a single lifecycle::deprecate_soft() warning. - Rename the matching igraph option keys edge.attr.comb -> edge_attr_comb and vertex.attr.comb -> vertex_attr_comb, with back-compatible aliasing in igraph_opt()/igraph_options(); the dotted keys still read and set, soft- deprecated. Update the stimulus codegen default (types-RR.yaml) and the regenerated *_impl defaults to match. Deprecated wrappers (as.undirected(), contract.vertices()) are left frozen; their snapshot now records the resulting layered deprecation warnings. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Run: https://github.com/igraph/rigraph/actions/runs/27270008480

github-actions · 2026-06-10T10:43:04Z

This is how benchmark results would change (along with a 95% confidence interval in relative change) if b58da44 is merged into main:

✔️as_adjacency_matrix: 731ms -> 728ms [-1.46%, +0.89%]
✔️as_biadjacency_matrix: 735ms -> 740ms [-0.52%, +1.76%]
✔️as_data_frame_both: 1.51ms -> 1.5ms [-2.15%, +1.45%]
✔️as_long_data_frame: 4.06ms -> 4.05ms [-3.39%, +2.93%]
✔️es_attr_filter: 2.9ms -> 2.93ms [-3.59%, +5.69%]
✔️graph_from_adjacency_matrix: 118ms -> 119ms [-0.35%, +2.21%]
✔️graph_from_data_frame: 3.54ms -> 3.58ms [-1.51%, +4.05%]
✔️vs_attr_filter: 1.66ms -> 1.66ms [-2.81%, +3.13%]
✔️vs_by_name: 1.05ms -> 1.09ms [-4.13%, +11.57%]
Further explanation regarding interpretation and methodology can be found in the documentation.

…e combination in operators The `graph_attr_comb` parameter in `union()`, `intersection()`, `disjoint_union()`, and `compose()` now defaults to `igraph_opt("graph_attr_comb")` (which is `"rename"` by default) instead of a hard-coded `"rename"`. This allows users to globally configure graph attribute combination behavior via `igraph_options()`.

Run: https://github.com/igraph/rigraph/actions/runs/27640941656

github-actions · 2026-06-16T19:16:16Z

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 6d94668 is merged into main:

🚀as_adjacency_matrix: 753ms -> 714ms [-5.86%, -4.62%]
🚀as_biadjacency_matrix: 760ms -> 719ms [-6.08%, -4.74%]
✔️as_data_frame_both: 1.53ms -> 1.55ms [-1.38%, +3.28%]
✔️as_long_data_frame: 3.91ms -> 3.9ms [-1.4%, +1.38%]
✔️es_attr_filter: 2.72ms -> 2.76ms [-1.17%, +4.01%]
🚀graph_from_adjacency_matrix: 119ms -> 112ms [-6.8%, -4.45%]
❗🐌graph_from_data_frame: 3.41ms -> 3.46ms [+0.56%, +2.73%]
✔️vs_attr_filter: 1.59ms -> 1.61ms [-4.11%, +6.69%]
✔️vs_by_name: 1.02ms -> 1.02ms [-2.37%, +1.64%]
Further explanation regarding interpretation and methodology can be found in the documentation.

schochastics · 2026-06-16T19:30:44Z

@maelle it is now *_combine ;-)

Run: https://github.com/igraph/rigraph/actions/runs/27642731642

github-actions · 2026-06-16T19:44:03Z

This is how benchmark results would change (along with a 95% confidence interval in relative change) if a85a78c is merged into main:

✔️as_adjacency_matrix: 680ms -> 679ms [-0.8%, +0.74%]
❗🐌as_biadjacency_matrix: 678ms -> 683ms [+0.03%, +1.19%]
✔️as_data_frame_both: 1.49ms -> 1.48ms [-2.85%, +1.13%]
✔️as_long_data_frame: 3.94ms -> 3.92ms [-2.14%, +1.01%]
✔️es_attr_filter: 2.7ms -> 2.72ms [-1.14%, +2.69%]
✔️graph_from_adjacency_matrix: 110ms -> 110ms [-1.23%, +1.6%]
✔️graph_from_data_frame: 3.32ms -> 3.4ms [-1.65%, +6.38%]
✔️vs_attr_filter: 1.5ms -> 1.5ms [-2.3%, +2.33%]
✔️vs_by_name: 942µs -> 946µs [-1.47%, +2.21%]
Further explanation regarding interpretation and methodology can be found in the documentation.

maelle · 2026-06-23T07:54:41Z

Yay 🎉

maelle

Thank you!!

maelle · 2026-06-23T07:58:57Z

-      if (is.na(x)) {
+      idx <- pmatch(tolower(x), known_names)
+      if (is.na(idx)) {
+        if (!allow_rename && identical(tolower(x), "rename")) {


It'd be more logical to me to invert the two

if (identical(tolower(x), "rename") && !allow_rename) {

why identical btw?

you seem to hold a grudge against identical() 😛
Just very defensive type-strict scalar comparison. Probably == would work here but its igraph, so better be save 😉

maelle · 2026-06-23T07:59:41Z

+      if (is.na(idx)) {
+        if (!allow_rename && identical(tolower(x), "rename")) {
+          cli::cli_abort(
+            "{.val rename} is only supported by graph operators ({.fn union}, {.fn intersection}, {.fn compose}, {.fn disjoint_union}), not by this function."


Rephrase to

Can't use rename with function X (possible to add its name)

Hint: rename is only supported by blablabla

This gets surprisingly complicated so I would opt out of that and keep it as is

Why not

Can't use rename with this function (not naming it)
Hint: rename is only supported by blablabla

…n igraph.i.attribute.combination

…erals

… before checking allow_rename flag

Run: https://github.com/igraph/rigraph/actions/runs/28152398180

github-actions · 2026-06-25T07:07:42Z

This is how benchmark results would change (along with a 95% confidence interval in relative change) if fbf2bcc is merged into main:

✔️as_adjacency_matrix: 798ms -> 797ms [-1.22%, +1.08%]
✔️as_biadjacency_matrix: 781ms -> 778ms [-1.47%, +0.6%]
✔️as_data_frame_both: 1.62ms -> 1.64ms [-2.16%, +4.51%]
✔️as_long_data_frame: 4.25ms -> 4.27ms [-3.59%, +4.67%]
✔️es_attr_filter: 2.88ms -> 2.93ms [-2.8%, +6.51%]
❗🐌graph_from_adjacency_matrix: 126ms -> 128ms [+0.61%, +3.22%]
✔️graph_from_data_frame: 3.55ms -> 3.62ms [-0.43%, +4.31%]
✔️vs_attr_filter: 1.67ms -> 1.64ms [-4.2%, +0.92%]
✔️vs_by_name: 1.13ms -> 1.14ms [-2.13%, +4.11%]
Further explanation regarding interpretation and methodology can be found in the documentation.

krlmlr

I'd like to review of the "par" code in a separate PR. Let's keep the original option names dotted in this PR (revert as a single commit).

krlmlr · 2026-06-25T09:29:14Z

Suggested change

if (...length() == 0) {

github-actions · 2026-06-25T10:02:34Z

This is how benchmark results would change (along with a 95% confidence interval in relative change) if a8987c4 is merged into main:

✔️as_adjacency_matrix: 731ms -> 730ms [-0.69%, +0.44%]
✔️as_biadjacency_matrix: 737ms -> 732ms [-1.99%, +0.56%]
✔️as_data_frame_both: 1.46ms -> 1.48ms [-0.66%, +2.8%]
✔️as_long_data_frame: 3.91ms -> 3.9ms [-1.35%, +1.06%]
✔️es_attr_filter: 2.66ms -> 2.65ms [-1.57%, +0.47%]
✔️graph_from_adjacency_matrix: 117ms -> 117ms [-2%, +1.62%]
❗🐌graph_from_data_frame: 3.39ms -> 3.45ms [+0.12%, +3.49%]
🚀vs_attr_filter: 1.55ms -> 1.5ms [-6.79%, -0.14%]
✔️vs_by_name: 978µs -> 977µs [-2.9%, +2.62%]
Further explanation regarding interpretation and methodology can be found in the documentation.

…eparate PR Per review feedback (PR #2676), the snake_case renaming of the igraph options and of the *.attr.comb function arguments (with soft-deprecation of the dotted names) is "par" code that will be reviewed in its own PR. The full renaming work is preserved on branch `par-attr-comb-options`. This reverts the renaming here, keeping the original dotted option names: - R/par.R: restore dotted `vertex.attr.comb` / `edge.attr.comb` options and drop the alias/soft-deprecation machinery. Keep the feature's new `graph.attr.comb` option (dotted), defaulting to "rename". - simplify() / as_undirected() / contract(): drop the argument-migration blocks; keep the dotted `edge.attr.comb` / `vertex.attr.comb` arguments (community.R, conversion.R, simple.R, aaa-*.R, migrations.R, types-RR.yaml, and the associated tests/snapshots revert to main). - union() / intersection() / compose() / disjoint_union(): the new attribute-combination arguments use dotted names (graph.attr.comb / vertex.attr.comb / edge.attr.comb). Also addresses review comment r3473306646: use `...length() == 0` instead of `nargs() == 0` in igraph_i_options(). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… names Split out from #2676 per review feedback: the renaming of the igraph options and the *.attr.comb function arguments (with soft-deprecation of the dotted names) is carried here so it can be reviewed on its own. - igraph options vertex.attr.comb / edge.attr.comb (and the new graph.attr.comb) are renamed to snake_case *_combine, with back-compatible aliasing in igraph_opt() / igraph_options(); the dotted keys still read and set, soft-deprecated (R/par.R). - simplify() / as_undirected() / contract() gain snake_case edge_attr_combine / vertex_attr_combine arguments via the argument-migration infrastructure (tools/migrations.R); dotted names keep working and soft-deprecate. - union() / intersection() / compose() / disjoint_union() use the snake_case *_combine argument names. - Updates stimulus codegen defaults (types-RR.yaml), regenerated *_impl defaults, and the affected tests/snapshots. Retains the igraph_i_options() ...length() fix from the base branch. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Run: https://github.com/igraph/rigraph/actions/runs/28971001024

github-actions · 2026-07-08T20:01:54Z

This is how benchmark results would change (along with a 95% confidence interval in relative change) if a190d59 is merged into main:

🚀as_adjacency_matrix: 700ms -> 679ms [-4.45%, -1.43%]
✔️as_biadjacency_matrix: 689ms -> 684ms [-2.25%, +0.73%]
✔️as_data_frame_both: 1.4ms -> 1.37ms [-4.25%, +0.69%]
✔️as_long_data_frame: 3.61ms -> 3.63ms [-1.17%, +2.5%]
✔️es_attr_filter: 2.67ms -> 2.72ms [-1.79%, +5.32%]
✔️graph_from_adjacency_matrix: 134ms -> 135ms [-2.08%, +3.89%]
✔️graph_from_data_frame: 3.07ms -> 3.04ms [-2.94%, +1.14%]
✔️vs_attr_filter: 1.45ms -> 1.43ms [-4.21%, +1.19%]
✔️vs_by_name: 928µs -> 926µs [-2.27%, +1.86%]
Further explanation regarding interpretation and methodology can be found in the documentation.

… names Split out from #2676 per review feedback: the renaming of the igraph options and the *.attr.comb function arguments (with soft-deprecation of the dotted names) is carried here so it can be reviewed on its own. - igraph options vertex.attr.comb / edge.attr.comb (and the new graph.attr.comb) are renamed to snake_case *_combine, with back-compatible aliasing in igraph_opt() / igraph_options(); the dotted keys still read and set, soft-deprecated (R/par.R). - simplify() / as_undirected() / contract() gain snake_case edge_attr_combine / vertex_attr_combine arguments via the argument-migration infrastructure (tools/migrations.R); dotted names keep working and soft-deprecate. - union() / intersection() / compose() / disjoint_union() use the snake_case *_combine argument names. - Updates stimulus codegen defaults (types-RR.yaml), regenerated *_impl defaults, and the affected tests/snapshots. Retains the igraph_i_options() ...length() fix from the base branch. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add attribute combination support to graph operators (union, intersec…

c75e174

…tion, compose, disjoint_union)

schochastics added the hackathon-topic 📆 label May 28, 2026

krlmlr and others added 2 commits May 28, 2026 20:00

Merge branch 'main' into feat-attrib_comb

41c4b0c

make document work locally

5f7521a