Release notes

0.5.0 (2026-09-15)

Version 0.5 adds transactional ggstyle.line(), ggstyle.points(), and ggstyle.ribbon() helpers for named tidy-data columns on caller-owned axes. Discrete color and linestyle plus continuous color mappings train across layers through one weakly axes-owned registry. Public immutable scale policies make ordering, missing values, limits, infinite values, palettes, and guide titles explicit while retaining ordinary Matplotlib artists and existing date-axis behavior.

ggstyle.guides() derives native legends and colorbars from the complete trained registry. Compatible color and linestyle entries merge, distinct mappings coexist, and managed guides refresh with later layers without replacing caller-owned guides. Artist, scale, guide, date, and axes mutations remain transactional across failed construction and refresh paths.

Line, point, ribbon, and guide results now implement ggstyle.RenderedResult. as_dict() exposes bounded JSON-compatible audit data and describe() renders the same data as deterministic strict JSON without axes or live artists. Concrete results continue to expose their native geometry-specific objects.

Release gates verify equivalent pandas and Polars semantics, bounded time and memory for representative multi-layer rendering, an executable semantic gallery, clean-wheel inspection and rendering, supported Python/dependency profiles, documentation, typing, and visual regressions.

0.4.0 (2026-09-15)

Version 0.4 adds a publication-finishing layer that adopts existing Matplotlib axes. ggstyle.finish() coordinates plot titles, layout-managed subtitles and captions, axis titles, numeric labellers, parameterized themes, and managed endpoint labels as one validated transaction. Dry runs return the same ggstyle.FinishPlan retained by successful results; plans and date-axis summaries now expose deterministic, strict-JSON inspection without live artists or callables.

Locale-independent percent, currency, grouped-number, and SI labellers install through ordinary FuncFormatter objects. Public qualitative, sequential, and diverging palettes provide immutable values, explicit missing and out-of-bounds policies, and colour-vision regression gates. Theme recipes add proportional base sizing, font-family selection, validated overrides, and a documented safe boundary for existing axes.

Line-series endpoint labels match source colours, resolve vertical collisions in display space, reserve a constrained-layout right margin, and use a whole-plot legend-or-raise fallback when an artist or geometry is unsupported. Repeated requests reuse managed annotations, and commit failures restore prior labels, legends, themes, formatters, and layout state.

ggstyle.save() adds explicit physical dimensions, deterministic SVG/PDF metadata, tight or standard bounds, transparency policy, atomic overwrite protection, and figure state restoration. The clean-wheel smoke test now covers finishing, direct labels, collapsed coordinates, and export.

The documentation includes an executable finishing and theme gallery, inspection and migration guidance, an explicit tool-selection boundary, a human pilot protocol, link checking, and GitHub Pages deployment. Stable minimum and newest dependency profiles, the pinned visual renderer, typing tests against the wheel, and the full Python/platform matrix remain release-blocking.

0.3.0 (2026-09-15)

Collapsed coordinates now use one registered matplotlib x-scale instead of rewriting Line2D data. Lines, scatter, fill_between, and native x-data annotations share the same invertible display transform while their calendar geometry and date-number limits remain unchanged. Empty registries still reject collapse; a one-observation registry now uses one calendar day per ordinal unit in both directions.

DateAxis.loc() now returns a native matplotlib date number in both modes. Code that passes its result to matplotlib artists or limits continues to work, while code that asserted observation ordinals should instead inspect the axis scale transform. This is a deliberate pre-1.0 migration to avoid double-transforming native matplotlib operations.

Observation provenance is now held by an owned, revisioned registry. Lines, scatter collections, and native fill_between polygons are discovered automatically; explicit dates remain sticky, and removed or mutated artists are reflected by ggstyle.DateAxis.refresh(). Synchronized handles share one live registry, so refreshing any member transactionally updates every member without changing their date-number view limits. Polygon discovery supports masks, NaNs, where regions, interpolated crossings, and multiple paths without changing vertices or path codes. Midpoint-stepped polygons require explicit data= because Matplotlib does not retain their complete source x sequence. fill_betweenx and non-data polygon transforms are rejected on an x-date handle.

ggstyle.DateAxis.dispose() disconnects callbacks and weak registry ownership while leaving Matplotlib artists in place. Failed timezone, formatter, caption, discovery, and shared-registry operations retain the previous valid state.

Public fluent methods now retain ggstyle.DateAxis in downstream type checking. Release CI covers Python 3.10 through 3.13, representative Linux/macOS/Windows jobs, reproducible minimum and newest-stable dependency profiles, the pinned pixel renderer, registry performance, and a headless plot imported from the built wheel. Offset parsing accepts both legacy and modern pandas frequency spellings across the supported pandas 2.x and 3.x families.

0.2.0

This release completes ggstyle’s built-in ggplot2-inspired theme set with bw, linedraw, light, dark, classic, void, and test. The existing minimal default and grey theme remain unchanged. Corresponding ggplot2 function names such as theme_bw and theme_classic are accepted as aliases.

Each theme ships as a standalone matplotlib stylesheet. See the Themes section for the full list and matplotlib-specific fidelity notes.

0.1.1

This maintenance release refactors date-axis behavior into focused policy modules while preserving the public API, and strengthens reproducible build and publishing checks.

0.1.0

The initial public release introduces the standalone date-axis handle, date extraction for common dataframe and array libraries, collapsed observation spacing, date-space annotations, and two opt-in matplotlib themes.

See Pitfalls for current lifecycle and timezone limitations.