The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.
The first feature-complete release of the core grammar. Everything is built on the same computed-geometry buffer, so static and interactive output can never disagree.
Nineteen further layers, closing most of the gap with the geoms an R
user expects to find: geom_rect(),
geom_polygon(), geom_raster(),
geom_abline(), geom_blank(),
geom_label(), geom_linerange(),
geom_crossbar(), geom_errorbarh(),
geom_rug(), geom_curve(),
geom_spoke(), geom_count(),
geom_freqpoly(), geom_function(),
geom_qq(), geom_qq_line(),
geom_dotplot() and geom_quantile().
Quantile regression is fitted by minimising the check loss directly, so no linear-programming dependency is needed. 2D contouring/binning and spatial layers remain unimplemented.
coord_flip() swapped the marks but not the axes, so
every flipped plot was mislabelled.geom_sankey(), geom_chord(),
geom_network(), geom_upset() and
geom_consort() ignored all of their styling arguments,
because those geoms build their marks inside the stat, which never saw
the layer’s parameters.NA, NaN,
Inf) reached the renderer and produced invalid SVG such as
cx="NA". They are now dropped with a warning reporting the
row count, matching what the histogram path already did. Interval
columns keep their NAs, which
geom_forecast_band() relies on.geom_smooth() with too few observations wrote
NaN coordinates; it now refuses with a message naming the
minimum for the chosen method.stats::density()
errors; they now name the geom and the offending group.stat_bin(bins = n) dropped empty bins, so
n was only honoured when every bin was occupied.geom_point(stat = stat_bin()) works.geom_spider_response(thresholds =),
geom_smooth(alpha =) and
theme(grid_color_minor =) had no effect; gradient endpoints
only applied when both were set.labels vector of the wrong length was recycled instead of
reported.to_json() emitted a bare Inf, which is not
valid JSON, and left C0 control characters unescaped.build_site(gallery = FALSE) rewrites the pages without
redrawing the figures, for quick iteration on prose. Figures an earlier
build wrote are kept rather than dropped.facet_wrap() and facet_grid(), with
scales = "fixed" (default), "free",
"free_x", and "free_y". The geometry buffer
now carries a list of panels; a non-faceted plot is the one-panel
case.coord_polar() — the engine behind radar charts,
circular bar charts, and pie/donut wedges — and
coord_flip() for horizontal layouts.animate() is implemented: it builds one geometry buffer
per level of the transition variable, with axes fixed across frames, and
the interactive target plays them with a scrubber and play/pause. (It
previously warned and did nothing.)interact() gains brush = TRUE (drag a
rectangle to zoom) and accepts a character vector of data columns for
tooltip.geom_*(), theme_*(),
scale_*(), coord_*(), facet_*(),
labs(), interact(), animate() and
plot_size() that takes at least one argument now also
accepts a plot as its first pipe argument:
p |> geom_point() |> theme_minimal() runs the same as
p + geom_point() + theme_minimal(). + remains
the grammar’s one true composition operator — reusable
Theme/Scale bundles built independently of any
plot, and Reduce(+, layers, p), both still
work exactly as before; the pipe form is an additive call-site
convenience, not a replacement. Non-standard evaluation is preserved, so
p |> facet_wrap(cyl) resolves cyl against
the data exactly as p + facet_wrap(cyl) always has. A
handful of genuinely zero-argument constructors
(theme_minimal(), theme_classic(),
theme_modern(), theme_dark(),
theme_void(), theme_ggnext(),
coord_flip()) have no parameter to route a piped plot
through and stay +-only.validate_plot() checks a plot for common
statistical-graphics mistakes — an empty plot, a continuous geom
(geom_point(), geom_line(), …) mapped to a
categorical y, a mapped aesthetic that is mostly missing, a
discrete color/fill scale with more levels than a legend can usefully
show, and a sqrt-transformed scale fed negative values —
and returns the findings.plot_check() prints that report and returns the plot
unchanged, so it sits inside a pipeline:
p |> plot_check() |> render().labs(), ggtitle(), xlab(),
ylab() for the title block (title, subtitle, caption, tag)
and axis titles.xlim(), ylim(), lims() for
quick limit setting.scale_x_continuous() /
scale_y_continuous() gain breaks,
labels (a vector or a function), expand, and
trans.scale_x_log10(),
scale_x_sqrt(), scale_x_reverse() and the
y equivalents.scale_x_discrete() / scale_y_discrete()
for explicit level ordering.scale_color_manual() and
scale_color_gradient() for palette control.plot_size() and ggnext(width =, height =)
for output dimensions.theme_ggnext(),
theme_minimal(), theme_classic(),
theme_modern(), theme_dark(),
theme_void().theme() now takes a base argument, so any
preset can be restyled, and exposes 35 settings covering colors,
typography, grid and axis visibility, legend position, and
palettes.plot_data() returns exactly the values a plot draws —
post-stat, post-position, post-facet, in data units, with internal
scratch columns and constant grouping columns removed.write_plot_data() writes that table to CSV.geom_radar(),
geom_ridgeline(), geom_sankey() /
geom_alluvial(), geom_treemap(),
geom_network(), geom_chord(),
geom_parallel(), geom_bump(),
geom_funnel(), geom_stream(),
geom_upset().geom_shap(),
geom_partial_dependence(),
geom_confusion_matrix(), geom_calibration(),
geom_lift_gain(), geom_residual(),
geom_learning_curve(), geom_silhouette(),
geom_embedding(), geom_decision_boundary(),
geom_forecast_band().geom_forest(), geom_swimmer(),
geom_spider_response(),
geom_waterfall_response(),
geom_bland_altman(), geom_cuminc(),
geom_spaghetti(), geom_ae_heatmap(),
geom_dose_response(), geom_shift(),
geom_consort().Layout algorithms are implemented from scratch: squarified treemaps (Bruls, Huizing & van Wijk), Fruchterman-Reingold force layout, Sankey node stacking with a single shared value scale, Aalen-Johansen cumulative incidence, and four-parameter log-logistic dose-response fitting.
circle, line,
rect, polygon, text) are now
supported by both render targets; a new geom needs no renderer
changes.build_site() generates the documentation site,
rendering every gallery figure with ggnext itself so the gallery cannot
drift from the code.ggnext_logo() draws the hex sticker with the package’s
own SVG writer.aes(color = "blue") is honored as a literal color
rather than being treated as a single-level category and assigned the
first palette entry.aes(color = class) when class is
not a column) now raises an error naming the problem and listing the
available columns, instead of failing later with “attempt to replicate
an object of type ‘builtin’”.print methods are re-registered in
.onLoad(); S7’s own method registration was dropping them,
so render() printed the full document source instead of a
summary.Initial de-risking spike: S7 class hierarchy,
geom_point(), one continuous linear scale, Cartesian
coordinates, stat_identity(), the hand-rolled SVG writer,
and the interactive canvas target.
These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.