Refactor and extend plot.igraph(): renderer pipeline, and new features by schochastics · Pull Request #2711 · igraph/rigraph

schochastics · 2026-06-24T19:51:24Z

Summary

A substantial refactor and extension of the base-graphics plotting subsystem
(plot.igraph() and friends), staged as a long series of small, independently
reviewable commits. The work splits the old monolithic draw path into clear
stages (resolve aesthetics → geometry → draw list → render), lands several new
user-facing features on top of that foundation, fixes a real bug, and documents
the whole thing in a new article.

Everything new is opt-in / default-no-op: the vdiffr snapshot suite stays
the correctness gate, and default plot(g) output is unchanged unless a commit
is explicitly marked breaking.

Internal refactor (B/S/P/P2 commits)

Stage-1 edge aesthetic table (vctrs) with strict recycling; non-loop edges
routed through it.
Extracted pure geometry helpers (arrowhead shape, shaft endpoints, loop
angles, curved splines) and a vertex-label phase.
Strict recycling for plotting aesthetics — per-element args must be length
1 or vcount/ecount (igraph 3.0.0 behaviour; marked breaking).
add_shape() now validates the clip/plot signatures.
Removed dead defaults and duplicate tables; named magic constants.

New features

F1 — scales + legends: scale_color() / scale_colour() / scale_size()
map data to an aesthetic and record the mapping, so plot() draws a
categorical legend, a continuous colourbar, or a size legend. New legend
argument controls placement; guides live in their own resize-stable region.
F2 — label repel: vertex.label.repel = TRUE nudges overlapping labels
apart with leader lines.
F3 — edge styles: edge.style routing — straight / arc / elbow /
diagonal.
F4 — gradients + alpha: edge.gradient (source→target colour),
vertex.alpha / edge.alpha.
F5 — draw-list IR + SVG: base drawing routed through a renderer
abstraction; a record renderer materializes a backend-neutral draw list that a
new exported as_svg() consumes, emitting per-vertex/edge element IDs and
<title> tooltips (lightweight interactivity, no JS).
F6 — label readability: vertex.label.halo / edge.label.halo
(shadowtext-style outline) and label_top() to declutter dense graphs by
labelling only the top-N vertices by a metric.

Bug fix

Named per-vertex numeric aesthetics (e.g. vertex.size = scale_size(degree(g)),
since degree() is named) propagated names into i.edge_label_pos(), where
c(x = …, y = …) produced names like x.Alice instead of x/y and crashed
with "subscript out of bounds". Fixed at the source with regression tests.

Docs

New web-only pkgdown article "Plotting graphs"
(vignettes/articles/plotting.Rmd) walking through every plotting feature —
old and new — at least once, including an inline embedded SVG. Registered in
the navbar; the main igraph.Rmd plotting section trimmed to a teaser that
links to it.

Testing

Extended the plotting test suite up front; new tests for scales, repel, edge
styles, gradients, the renderer/SVG path, label halos, label_top(), and the
named-aesthetic regression.
Full plot suite green; the article knits cleanly end-to-end (strict
error = FALSE).

🤖 Generated with Claude Code

Add a behavior-locking safety net for the plotting subsystem before the planned refactor of parameter resolution and edge subsetting. - New test-plot-params.R: unit tests for i.parse.plot.params() precedence, recycling, and NA handling, plus the getter helpers (i.get.arrow.mode, i.get.labels, i.get.main/xlab, igraph.check.shapes, curve_multiple, i.rescale.vertex). - test-plot.shapes.R: clip-math tests for non-circle shapes, including the per-endpoint vertex.size vector selection. - test-plot.R: integration snapshots for vector edge params across loops and non-loops, auto-curved multi-edges, NA-attribute warning, multi-group mark.groups, label.dist/degree, add=TRUE overlay, and palette indexing. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

P1.4: drop the redundant second i.default.values[["edge"/"plot"]] assignments in plot.common.R (exact duplicates). P1.5: introduce VERTEX_SIZE_SCALE (1/200 size->user-coord factor) and ARROW_WIDTH_FACTOR (1.2/4 arrowhead scaling) in plot.R, applied across plot.igraph() and igraph.Arrows(). Behaviour-preserving; rglplot() occurrences left untouched (no snapshot coverage here). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add R/plot-aes.R with i.aes_table()/i.edge_aes_table(), which package resolved per-edge plotting parameters into a vctrs data frame that can be sliced by edge index. Use it in plot.igraph() to replace the 11x repeated `if (length(x) > 1) x[loops.e]` block for loop edges with a single table build + vec_slice(). Behaviour-preserving: recycling keeps the historical rep(length.out=) semantics (downstream mapply/igraph.Arrows recycle anyway), so the strict-recycling change flagged in the plan is intentionally NOT adopted here (it needs a revdep check). The non-loop block's entangled per-arrow-code curved handling is left for the P2.4 consolidation step. Also document the resolution precedence on i.parse.plot.params() and add unit tests for the new table helpers. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Pull the arrowhead-outline computation out of the igraph.Arrows() drawing loop into i.arrowhead_shape(cin, w, delta), a pure function returning the head outline in polar form. Adds a device-free unit test for it (previously the geometry was only exercised indirectly via the standard-arrow* snapshots). Scoped to the arrowhead geometry; the shaft-endpoint and curved-path math remain inline (more entangled with drawing/mutation) and can follow in the P2.4 consolidation. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Move the four nested functions (point.on.cubic.bezier, compute.bezier, plot.bezier, loop) from inside plot.igraph()'s body to file-level internal functions, renamed with an i. prefix. i.plot.bezier in particular avoids being mistaken for an S3 plot() method for class "bezier". This removes ~155 lines from the plot.igraph() body and makes the helpers independently reusable. The functions capture no enclosing state; loop's self-referential arg defaults (always supplied by the mapply call site) become harmless literals. Also drops a dead duplicate `ec <- edge.color` extraction left over from the Stage 1 change. Behaviour-preserving (loop-graph / multi-loops snapshots unchanged). The larger phase-helper extraction (plot_vertices/plot_labels) is deferred: threading the 40+ locals is higher risk than snapshot-only verification covers. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Two behaviour-preserving render-path cleanups: - Extract i.hide_zero_frame() for the "frame.width <= 0 hides the border" rule that was copy-pasted into the circle/square/rectangle plot functions. - Extract i.init_plot_canvas() for the empty-canvas plot() setup, so plot.igraph() reads as setup -> edges -> vertices -> labels. Snapshots unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

BREAKING CHANGE (igraph 3.0.0): a per-element plotting aesthetic must now be length 1 or exactly vcount()/ecount(). Previously a wrong-length vector (e.g. 3 colors for 5 vertices) was silently recycled, masking user mistakes; it is now a clear error via i.check_aes_lengths(). Applied to the unambiguous per-element vertex/edge aesthetics. Intentionally excluded: arrow.mode (its "a:" form reads a vertex attribute, so it can be vcount-long), and label.adj / pie / raster, which have non-per-element length semantics. Adds unit tests for i.check_aes_lengths plus plot()-level error/valid-length tests and error snapshots. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Build the edge aesthetic table once (now including arrow.width) before both the loop and non-loop edge blocks, and slice it per block with vec_slice(). This replaces the remaining per-parameter `if (length(x) > 1) x[idx]` subsetting and fixes two latent bugs in the mixed-arrow-mode path: - `curved` was sliced to non-loop edges and then re-expanded with `rep(curved, length.out = ecount)[nonloops.e]`, double-indexing and scrambling curvature values. - `arrow.size` / `arrow.width` were not subset by the per-arrow-code `valid` mask, so igraph.Arrows() received the leading elements rather than the matching ones. Both now index consistently by `valid`. BREAKING: plots that combined mixed arrow modes with per-edge curved/arrow.size/arrow.width render differently (now correctly) — the vector-edge-params-loops snapshot is updated accordingly, and a new mixed-modes-curved regression snapshot covers the path. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Pull two self-contained phases out of plot.igraph()'s body: - i.loop_angles(graph, layout, loops.v): the "flower-petal" distribution of self-loops into the largest angular gap at each vertex, returning aligned angle/narrowing vectors. Drops two dead locals (loop_table, loop_idx). Now unit-testable. - i.draw_vertex_labels(...): the vertex-label placement + drawing block, with xpd scoped to the helper. Behaviour-preserving (loop / label snapshots unchanged); adds a unit test for i.loop_angles. The loop/non-loop edge-drawing blocks remain inline for now. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Extract the remaining per-edge geometry from the igraph.Arrows() drawing loop into pure helpers: - i.arrow_shaft_endpoints(): shaft segment endpoints, pulled back at the arrowed end(s) per `code`. - i.edge_label_pos(): straight-edge label anchor (2/3 along the edge). - i.curved_spline(): the X-spline control points + curve for a curved edge. igraph.Arrows() now computes geometry via these helpers and only draws. Adds device-free unit tests for the two pure helpers; arrow snapshots unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

add_shape() now checks that the clip and plot functions accept the arguments igraph calls them with (clip(coords, el, params=, end=) and plot(coords, v=, params=)), via i.check_shape_fun(). Functions taking `...` are exempt. A malformed custom shape now fails at registration with a clear message instead of cryptically at plot time. Scoped to signature validation; the larger metadata-driven shape registry (per-shape parameter table + plot-time "unknown param" validation) is deferred — it would change what shapes() returns and risks false positives. Adds tests for the new validation, with cleanup so test shapes don't leak into the global registry. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

i.plot.default carried `frame = FALSE`, but plot.igraph() reads `frame.plot` (which falls back to `axes` when unset), so the entry was never used. Remove the misleading dead config and document the actual behaviour. No behaviour change (rescale-coords snapshot with axes = TRUE is unchanged). The file splits in B7 (plot.common.R / layout.R) are deferred: pure code-motion with high merge-conflict cost and no behaviour value, best done once the restructuring churn has settled. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

@export

Add scale_color()/scale_colour() and scale_size(): pass them to vertex/edge colour or size arguments to map a data column to the aesthetic AND draw a matching guide automatically, e.g. plot(g, vertex.color = scale_color(V(g)$group)) plot(g, vertex.size = scale_size(degree(g))) - scale_color(): discrete data -> categorical_pal() + a legend; numeric data -> a sequential colour ramp + a colorbar. - scale_size(): numeric data -> a size range (optional transform) + a size legend. - A new `legend` argument to plot.igraph() controls placement ("topright" default, any corner keyword) or suppression (FALSE). Implemented natively on base graphics (R/plot-scales.R): scales are resolved to plain aesthetic vectors before i.parse.plot.params() (whose recycling strips attributes), with their guides collected and drawn after the vertices/labels. Wrong-length scale data is caught by the existing strict-recycling check. Adds unit tests (test-plot-scales.R) and vdiffr snapshots covering discrete, continuous, size, combined, edge-colour, repositioned, and suppressed guides. NAMESPACE exports are added via @export (regenerated in CI). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Legends/colorbars previously sat in a plot corner and overlapped the graph. Now they are drawn in reserved outer-margin space on one side, never over the graph: - New `legend` semantics: TRUE/"right" (default), "left", "top", "bottom" (corner keywords map to the nearest side); FALSE suppresses. - "top"/"bottom" arrange legend entries horizontally and draw a horizontal colorbar; "left"/"right" stack vertically with a vertical colorbar. - plot.igraph() reserves par("mar") on the chosen side before drawing (scaled to label width for left/right) and restores it on exit; guides are drawn with xpd = NA into that margin. - Multiple guides stack (down for left/right, across for top/bottom). Snapshots for the scale cases are regenerated for the new placement; adds horizontal bottom-legend and top-colorbar snapshots. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The previous corner/margin placement positioned guides in data coordinates, which (with asp = 1) drift as the device is resized and could be off-centre or disappear. Switch to the standard two-region approach: split the device with par("fig") into a plot region and a guide region (device-relative NDC, so it is stable across resizes), draw the graph in the plot region, then draw the guides into the guide region via par(new = TRUE). Guides are centred within their region (two-pass measure then layout): stacked vertically for left/right, laid out in a row for top/bottom, with horizontal colorbars for top/bottom and vertical for left/right. Replaces i.draw_guides/i.draw_one_guide/i.draw_colorbar/i.legend_reserve_mar with i.legend_fig + i.draw_guides_region + i.guide_draw + i.colorbar. Legends are only drawn when add = FALSE (a guide region can't be carved out of an existing plot). Scale snapshots regenerated for the new placement. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add an opt-in ggrepel/Gephi-style label-adjust: with vertex.label.repel = TRUE, overlapping vertex labels are iteratively nudged apart and a thin leader line connects each moved label to its original anchor. - i.repel_labels(): deterministic force layout — labels repel along the axis of smaller box overlap and are sprung back toward their anchor. Pure (takes label box half-sizes), so it is unit-testable without a device. - i.draw_vertex_labels() gains a `repel` argument: it measures each non-empty label's box (strwidth/strheight), runs the repel, draws leader lines for labels that moved, then draws the text at the adjusted positions. - New `label.repel` vertex default (FALSE) resolved in plot.igraph(); documented in the plotting parameters. Default rendering is unchanged. Adds unit tests for i.repel_labels and a vdiffr snapshot of repelled clustered labels. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add a per-edge `edge.style` aesthetic selecting the routing of non-loop edges: "auto" (default; straight, or arc when edge.curved != 0 — unchanged behaviour), "straight", "arc" (curved; strength from edge.curved or a default), "elbow" (two-corner orthogonal connector), and "diagonal" (smooth S-curve with axis-aligned ends). - New pure geometry helpers i.elbow_path() and i.diagonal_path() (the latter reuses the existing Bezier helpers); device-free and unit-tested. - igraph.Arrows() gains a `style` arg and two new draw branches; the existing straight/arc branches are kept byte-identical so no existing snapshot changes. Arrowheads for elbow/diagonal align with the path's end segment. - Wired through i.edge.default, the edge aesthetic table, and the non-loop dispatch; resolved and validated in plot.igraph() (unknown style -> error). Documented in the plotting parameters; ignored for loops and rglplot(). Adds unit tests for the path helpers and vdiffr snapshots for elbow, diagonal, mixed per-edge styles, and forced arcs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add two opt-in edge/vertex aesthetics, both no-ops by default: - edge.gradient: when TRUE, an edge is drawn as a colour gradient from its source vertex's colour to its target vertex's colour (a direction cue), with the arrowhead taking the target colour. Implemented via i.draw_gradient_path() (arc-length resample + per-segment colorRamp); igraph.Arrows() gains `gradient`/`col.to` args and a gradient branch per edge style. Source/target colours come from the base vertex fill (resolved lazily, only when a gradient is used). - vertex.alpha / edge.alpha: per-element opacity in [0, 1], folded into the fill colours via i.apply_alpha() (multiplies existing alpha; strict no-op when all 1, so existing plots/snapshots are unchanged). edge.alpha also applies to gradient endpoints; vertex.alpha is injected into vertex.color before the shape draw. Frame colour, pie slices and labels are unaffected. Wired through i.vertex.default/i.edge.default, the edge aesthetic table, and the non-loop dispatch; documented in the plotting parameters. Adds unit tests for i.apply_alpha and vdiffr snapshots for gradients and translucency. Gradients are ignored for self-loops and by rglplot(). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Introduce a Stage-3 rendering indirection: all of plot.igraph's 2D drawing now emits primitives through i.r_*() dispatchers that forward to a "current" renderer (R/plot-render.R), instead of calling base graphics directly. The default base renderer simply calls the matching base-graphics function, so on-screen output is unchanged (the full plotting snapshot suite passes byte-identical). Migrated every 2D draw site to the dispatchers: the canvas, edges/arrowheads (igraph.Arrows), loop Beziers, the gradient path, mark-group xsplines, vertex and edge labels, and the vertex shape $plot functions (circle/square/rectangle symbols, pie polygons, sphere/raster images). The public shape (coords, v, params) contract is unchanged -- shapes call the i.r_* dispatchers internally. This is the foundation for a record/SVG renderer (phase 2) consuming the same primitive stream. No user-facing change. Adds unit tests for the indirection. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add a recording renderer that materializes the Stage-3 draw list (primitives with hex-canonicalised colours and their vertex/edge group), and a small SVG writer that consumes it. New exported as_svg(graph, file, width, height, tooltips, ...) renders a graph to standalone SVG reusing plot.igraph's geometry (via an offscreen measurement device), emitting per-vertex <g id="vertex-N"> with <title> tooltips and per-edge <g id="edge-N"> groups -- lightweight hover/click interactivity with no JavaScript. Coverage: vertices (circle/square/rectangle), every edge style (straight/arc/elbow/diagonal/gradient), arrowheads, labels, mark groups and pie slices render; sphere/raster shapes draw as a placeholder box (v1). NA/empty labels are skipped to match base output. Per-edge grouping is threaded through igraph.Arrows via a new `ids` arg; the base renderer ignores grouping, so base plotting output is unchanged (snapshots byte-identical). Adds tests for the draw list and the SVG (well-formedness, per-element ids, tooltips, file output). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add two opt-in, default-no-op label features to plot.igraph(): - Label halo: a shadowtext-style legibility outline behind vertex and edge label text, via new i.r_text_halo() and `vertex.label.halo` / `edge.label.halo` (colour, default NA) + `*.halo.width` params. With NA the draw path is byte-identical to before, so existing snapshots are unchanged. - label_top(): an exported helper returning a label vector with NA outside the top-N by a metric, composing with plot.igraph()'s NA-omission to declutter dense graphs (mirrors the scale_*() helper style). Halo params flow through i.edge_aes_table, the default registry, and the length checks; routed through all three 2D label sites (vertex, non-loop edge, loop edge). Adds label_top unit tests, halo vdiffr snapshots, and an as_svg halo smoke test. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add a web-only pkgdown article (vignettes/articles/plotting.Rmd) that walks through every plotting feature at least once: layouts, vertex shapes/colour/ size/alpha, all label controls plus repel and halos, edge width/lty/arrows, curved fans and the arc/elbow/diagonal edge styles, edge alpha and gradients, palettes, scales & legends (categorical legend + colourbar + size legend), label_top() decluttering, mark.groups highlighting, annotations, custom shapes, igraph_options precedence, and as_svg() with an inline embedded SVG. Register it in the _pkgdown.yml Articles navbar, and trim the now-redundant plotting reference tables in igraph.Rmd down to a teaser that links to the new article. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

A named per-vertex numeric aesthetic — e.g. vertex.size = scale_size(degree(g)), since degree() returns a named vector — propagated its names through edge clipping into i.edge_label_pos(). There, c(x = <named>, y = <named>) produced names like "x.Alice" instead of "x"/"y", so the downstream lab[["x"]] / lab[["y"]] lookups failed with "subscript out of bounds". unname() the components in i.edge_label_pos() so the result is always named x/y regardless of the inputs. Add regression tests covering a named scale, a named raw size vector, and i.edge_label_pos() directly. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Two examples produced empty/partial figures: - Self-loops: make_graph(~ A -+ A, ...) silently drops self-loops, so only the A->B edge rendered. Build the edge list explicitly with make_graph(c("A","A","A","B","B","B"), directed = TRUE) and add a fixed layout/margin so both loops are clearly visible. - Custom triangle shape: with a uniform shape, plot.igraph() calls the shape plot function once for all vertices (v = NULL), so a scalar vertex.size mismatched the coordinate matrix. Recycle the size with rep(..., length.out = nrow(coords)), as the built-in shapes do. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

- Use a denser make_full_graph(25) with larger labels for the vertex.label.repel example so the de-overlapping is clearly visible. - Add an example of several self-loops on a single vertex, which igraph arranges in a flower-petal pattern. - Drop the trailing sessionInfo() chunk. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Remove the development-scaffolding tags from comments and test labels (feature F1-F6, B-numbers, "phase N", and the pipeline "Stage N" labels), keeping the descriptive prose. These referred to the planning notes and add no value in the shipped code. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

- as_svg(): set/reset i.render_state$vertex_titles in the function's own frame instead of inside the i.with_renderer() block, where it relied on a subtle on.exit(add = TRUE) interaction across the forced-promise boundary. Clearer and more robust; behaviour unchanged. - base renderer init_canvas(): use if/else instead of ifelse() for the scalar frame.plot fallback. - reflow an aesthetic-table comment. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Run: https://github.com/igraph/rigraph/actions/runs/28126122425

github-actions · 2026-06-24T20:19:07Z

This is how benchmark results would change (along with a 95% confidence interval in relative change) if b069ab2 is merged into main:

✔️as_adjacency_matrix: 734ms -> 736ms [-0.38%, +0.92%]
🚀as_biadjacency_matrix: 743ms -> 738ms [-1.25%, -0.08%]
✔️as_data_frame_both: 1.45ms -> 1.45ms [-0.79%, +1.16%]
✔️as_long_data_frame: 3.88ms -> 3.85ms [-2.65%, +1.27%]
✔️es_attr_filter: 2.6ms -> 2.62ms [-0.1%, +1.48%]
✔️graph_from_adjacency_matrix: 115ms -> 116ms [-1%, +2.01%]
❗🐌graph_from_data_frame: 3.33ms -> 3.39ms [+0.21%, +3.59%]
✔️vs_attr_filter: 1.5ms -> 1.53ms [-1.21%, +4.62%]
✔️vs_by_name: 957µs -> 960µs [-1.62%, +2.41%]
Further explanation regarding interpretation and methodology can be found in the documentation.

Elbow and diagonal edges attached to vertices at the centre-to-centre boundary point (the shape clip functions clip along the straight line between centres, regardless of style). Since these styles route along an axis, their first/last segment then started off the vertex's centre axis — very visible in tree layouts, where a parent's stubs left from down-left/down-right of its bottom and children were entered left/right of their top. For elbow/diagonal edges, re-clip the endpoints along the dominant axis (chosen from the vertex centres) so they attach at the top/bottom- or left/right-centre of each vertex, and pass that axis into the path builders so routing matches the attachment. Straight, arc and self-loop edges are unchanged. - i.elbow_path()/i.diagonal_path(): new `vertical` arg (NULL = infer as before). - igraph.Arrows(): new `axis` arg, recycled and forwarded to the path builders. - plot.igraph(): re-clip elbow/diagonal endpoints via new i.axis_clip_endpoints() (reuses the shape clip functions with axis-aligned synthetic segments) and pass the per-edge axis to igraph.Arrows. - Tests: explicit-axis unit tests; an integration test asserting centre-axis attachment via the record renderer; re-blessed the elbow/diagonal/mixed snapshots (arc/straight snapshots unchanged). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Run: https://github.com/igraph/rigraph/actions/runs/28155463550

github-actions · 2026-06-25T08:11:06Z

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 70e36ff is merged into main:

✔️as_adjacency_matrix: 833ms -> 824ms [-3.04%, +0.68%]
✔️as_biadjacency_matrix: 836ms -> 840ms [-1.24%, +2.23%]
✔️as_data_frame_both: 1.58ms -> 1.57ms [-3.02%, +2.73%]
✔️as_long_data_frame: 3.98ms -> 4.06ms [-1.69%, +5.32%]
✔️es_attr_filter: 2.83ms -> 2.82ms [-2.12%, +1.41%]
✔️graph_from_adjacency_matrix: 155ms -> 154ms [-2.17%, +1.24%]
✔️graph_from_data_frame: 3.43ms -> 3.45ms [-0.51%, +1.59%]
✔️vs_attr_filter: 1.65ms -> 1.65ms [-3.43%, +4.32%]
✔️vs_by_name: 1.22ms -> 1.09ms [-28.39%, +5.97%]
Further explanation regarding interpretation and methodology can be found in the documentation.

github-actions · 2026-06-25T19:33:42Z

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 8cbc440 is merged into main:

✔️as_adjacency_matrix: 773ms -> 782ms [-0.78%, +3.06%]
✔️as_biadjacency_matrix: 788ms -> 789ms [-1.02%, +1.17%]
✔️as_data_frame_both: 1.55ms -> 1.61ms [-0.64%, +8.25%]
✔️as_long_data_frame: 4.06ms -> 4.03ms [-3.74%, +2.21%]
✔️es_attr_filter: 2.95ms -> 2.88ms [-9.3%, +4.23%]
✔️graph_from_adjacency_matrix: 130ms -> 128ms [-3.62%, +0.49%]
✔️graph_from_data_frame: 3.52ms -> 3.49ms [-4.6%, +2.75%]
✔️vs_attr_filter: 1.69ms -> 1.71ms [-3.81%, +6.57%]
✔️vs_by_name: 1.03ms -> 1.02ms [-6.27%, +3.53%]
Further explanation regarding interpretation and methodology can be found in the documentation.

schochastics and others added 25 commits June 23, 2026 21:31

schochastics added the hackathon-topic 📆 label Jun 24, 2026

schochastics and others added 2 commits June 24, 2026 21:57

schochastics removed the hackathon-topic 📆 label Jun 24, 2026

schochastics marked this pull request as draft June 24, 2026 20:05

schochastics changed the title ~~Refactor and extend plot.igraph(): renderer pipeline, new features, and a plotting article~~ Refactor and extend plot.igraph(): renderer pipeline, and new features Jun 24, 2026

chore: Auto-update from GitHub Actions

ab0ab38

Run: https://github.com/igraph/rigraph/actions/runs/28126122425

schochastics and others added 2 commits June 25, 2026 09:55

chore: Auto-update from GitHub Actions

3728a88

Run: https://github.com/igraph/rigraph/actions/runs/28155463550

Merge branch 'main' into refactor-plotting

8cbc440

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Refactor and extend plot.igraph(): renderer pipeline, and new features#2711

Refactor and extend plot.igraph(): renderer pipeline, and new features#2711
schochastics wants to merge 31 commits into
mainfrom
refactor-plotting

schochastics commented Jun 24, 2026

Uh oh!

github-actions Bot commented Jun 24, 2026

Uh oh!

github-actions Bot commented Jun 25, 2026

Uh oh!

github-actions Bot commented Jun 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Uh oh!

Conversation

schochastics commented Jun 24, 2026

Summary

Internal refactor (B/S/P/P2 commits)

New features

Bug fix

Docs

Testing

Uh oh!

github-actions Bot commented Jun 24, 2026

Uh oh!

github-actions Bot commented Jun 25, 2026

Uh oh!

github-actions Bot commented Jun 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant