Changelog#

Release versions follow Semantic Versioning i.e., <major>.<minor>.<patch>.

Backward incompatible breaking changes will only be introduced in <major> versions, with advance notice provided in the 🔥 Deprecations section of a prior <minor> release.


v0.6.0.dev1362 (2026-11-20)#

💣 Breaking Changes#

  • #1468: Adopted PEP 570 and PEP 3102 to enforce clarity on position-only and keyword-only parameters in public API functions and methods. (@bjlittle)

  • #1751: Renamed the following geovista download command line interface (CLI) flags --image to --images, and --raster to --rasters. (@bjlittle)

  • #1984: Renamed the utility function sanitize_data to geovista.common.sanitize_vtk(). (@bjlittle)

  • #2176: Renamed the convenience dataclass SampleStructuredXYZ to geovista.pantry.data.SamplePointsXYZ. (@bjlittle)

  • #2259: Replaced the geovista.theme module with geovista.themes. Previously the global plotting theme was updated as a side-effect by importing geovista.theme. Instead geovista.themes adopts the new extensible pyvista.plotting.themes.Theme plugin registry framework for third-party packages available from >=0.48.0. (@bjlittle)

  • #2561: Dropped support for python 3.12, as per SPEC 0, which retires a minor version three years after its release. The minimum supported version of python is now 3.13. (@claude)

  • #2591: Planar zlevel offsets change size. A mesh spanning the globe drops to 0.64 of its old offset in eqc, 0.75 in robin and 0.71 in moll, while a smaller mesh rises, the more so the smaller it is: about twice as high for one over the North Atlantic, 23 times for one 10° across and 229 times for one 1° across. Layers keep their order, but anyone who tuned zscale for a projection will see a different depth. To restore it, multiply zscale by the old offset over the new one, which raises it for a global mesh and lowers it for a regional one. (@claude)

  • #2711: Removed geovista.qt.GeoMultiPlotter, without a deprecation period. It mixed the geovista plotter API into pyvistaqt’s MultiPlotter, which is a grid of plotters rather than a plotter itself, so adding a mesh to it always raised an AttributeError. pyvistaqt has also deprecated MultiPlotter, for removal in its 0.14 release. For subplots, give GeoBackgroundPlotter a shape instead, e.g., GeoBackgroundPlotter(shape=(1, 2)), and select each subplot with subplot. (@claude)

  • #2766: radius= is removed from every function that took it, with no alias. Each takes globe=, a geovista.globe.Globe, in its place, and the methods of geovista.geoplotter.GeoPlotter take the globe of their scene. A mesh records its globe, and no longer writes gvRadius. A mesh saved by 0.5 after a z-level lifted it is drawn at the scene’s radius without its lift: add it with its z-level, or fix the file once with resize(mesh, globe=Globe(), zlevel=...). (@claude)

✨ New Features#

🚀 Enhancements#

🐛 Bug Fixes#

  • #1184: Fixed geovista.bridge.Transform.from_points() to create a mesh from scalar spatial points. (@bjlittle)

  • #1202: Configured bespoke User-Agent within requests header of pooch.Pooch.fetch() to avoid flaky HTTP 403 Forbidden error responses when downloading assets from the geovista-data repository. (@bjlittle)

  • #1301: Fixed geovista.common.from_cartesian() to handle polar points with no associated cell. (@bjlittle)

  • #1436: Fixed geovista.cache tests for Windows platform. (@beroda)

  • #1466: Ensured to close the pooch.Pooch cache registry after configuration. (@bjlittle)

  • #1956: Fixed gallery carousel generation to support mixed-case URL targets for sphinx-design card carousels. (@tkknight)

  • #1977: Fixed geovista.transform such that transform_mesh(), transform_point(), and transform_points() correctly handle a projected CRS source transformation. Additionally, geovista.geoplotter.GeoPlotterBase.add_mesh() correctly renders a projected CRS geometry within a plotter scene. (@bjlittle)

  • #2018: Fixed issue where mesh_args were not always being passed to the GeoPlotter when adding graticule meridians/parallels. (@ukmo-ccbunney)

  • #2042: Removed the geovista.theme opinionated scalar bar configuration for fixed position and size. Rather, default to adaptive positioning and scale behaviour offered by pyvista, leaving the user to customise. Otherwise a geovista themed scalar bar can easily render incorrectly. (@bjlittle)

  • #2050: Added defensive guard within geovista.geoplotter.GeoPlotterBase.add_mesh() prior to accessing the is_empty property, which does not exist on all pyvista instances. (@bjlittle)

  • #2416: Fixed the extraction.panel_manifold example, which now requests a border width of 2 pixels. pyvista 0.49 draws a single seam shared between adjacent subplots rather than a border around each one, which would otherwise halve the width of the seam rendered between the panels. (@larsoner)

  • #2427: Fixed the ICON Mesh and ICON Mesh (Projected) gallery examples, which use the cet_CET_L17 colormap but never imported colorcet, and so relied on some other module having already registered the cet_* colormaps with matplotlib. Running either example on its own raised a ValueError. (@claude)

  • #2514: Fixed geovista.crs.get_central_meridian() and geovista.crs.set_central_meridian(), which recognised the central meridian only as EPSG parameter 8802. A projection method defines its longitudinal origin using whichever parameter suits its geometry, so eight cartopy.crs projections silently lost their shift and were seam-cut at longitude 0, tearing the mesh across the projection. Closes #2512. (@claude)

  • #2522: Fixed the detection of a mesh that has already been flattened, which previously relied on pyproj.crs.CRS.is_projected. A flat geographic CRS such as +proj=latlong, which is what cartopy >=0.26 uses for ccrs.PlateCarree(), reports is_projected as False, so such a mesh was processed as if it were still a sphere. geovista.transform.transform_mesh() and geovista.geoplotter.GeoPlotterBase.add_mesh() then sliced the mesh at a seam and spun its longitude/latitude values with rotate_z, corrupting a shifted reprojection and desynchronising geovista.geodesic.BBox.enclosed(). add_mesh with a zlevel radially rescaled it, and geovista.geoplotter.GeoPlotterBase.view_poi() placed the camera edge-on to the plane. (@claude)

  • #2578: Fixed two races between processes sharing an empty cache. A process could read a compressed sample asset while another was still decompressing it, which raised read errors or, in one case, rendered the wrong data without any error. Two processes could also both try to create the same cache directory, and the second failed with a FileExistsError. (@claude)

  • #2580: transform_mesh now honours a per-point zlevel given as a list, tuple or array, as its ArrayLike annotation and docstring promise. On anything but a point cloud a list or tuple failed with TypeError, and an array of more than one element failed everywhere with ValueError. A zlevel that does not broadcast to the points now raises ValueError before any point is written. transform_point is now annotated and documented to accept a single valued list or array for x, y and z, as it always has at runtime. geovista.transform also leaves the typing ratchet, so mypy checks it in full. (@claude)

  • #2587: transform_mesh no longer flattens a mesh it sends to WGS84 with a non-zero zlevel, or a point cloud it brings back to WGS84 from a planar CRS. A point cloud now keeps its z-levels from one transform to the next, carried in the new geovista.common.GV_POINT_ZLEVEL point array while it is in a planar CRS, so a cloud moved between planar CRSs no longer has its z scaled twice. A point cloud also returns to WGS84 on the radius it carries, as it already did with its zscale, and a radius or zscale given to a transform becomes the cloud’s own. Closes #2581. (@claude)

  • #2591: In a planar CRS, transform_mesh now lifts a mesh by zlevel * zscale times the radius of the Earth in the units of that CRS, the proportion a level already has on the globe. Every mesh and point cloud in a projected scene shares that length, so meshes of different extents at one level sit at one height, a layer at a higher level draws above one at a lower level, and a feature taken from a point cloud sits at the depths it has in the whole cloud. Before, the length was a quarter of each mesh’s own extent. Closes #2588. (@claude)

  • #2597: slice_lines now carries a line mesh’s point and cell data across the antimeridian, so a line crossing the seam keeps its data in a projected plot. Each point it adds takes the values interpolated at the crossing, a duplicated vertex keeps its own, and both halves of a split segment keep its cell data. Every array keeps its role, such as the active scalars or pedigree ids, and any point or cell added gets a global id of its own. transform_mesh can now carry a per-point zlevel along such lines too. Closes #2583. (@claude)

  • #2599: Transform.from_points now accepts lists for xs and ys when vectors share the points’ latitude-longitude CRS and it isn’t WGS84, as its ArrayLike annotation promises. It raised AttributeError. Transform.from_unstructured drops points from their faces when xs and ys carry the same mask, in any CRS, whether the connectivity comes from the shape of the points or is given as a tuple. Before, it did so only for WGS84 points with no connectivity given. A face left with fewer than three points is dropped, with a warning, so cell data must fit the faces that remain. geovista.bridge also leaves the typing ratchet, so mypy checks it in full. (@claude)

  • #2612: vectors_to_cartesian now accepts lists for its longitudes, latitudes and vector components, as its ArrayLike annotation promises. It raised AttributeError. cast_UnstructuredGrid_to_PolyData raises the TypeError it intends for anything other than an unstructured grid, where it raised AttributeError. nan_mask returns an array when given a list, and fills any masked arrays in the list with NaN. triangulated returns a bool rather than a NumPy bool. wrap returns an array for a 0D array, where it returned a NumPy scalar and raised TypeError for a value close to the end of the period. geovista.common also leaves the typing ratchet, so mypy checks it in full. (@claude)

  • #2613: Fixed geovista.pantry.data.name_reykjanes(), which looked up the sample’s release location with an online geocoder each time it loaded. The lookup needed the network, and returned the nearby town of Grindavík, 4.8 km from the release. The location is now read from the sample itself. Closes #2608. (@claude)

  • #2615: Fixed the warning from geovista.bridge.Transform.from_unstructured() when masked connectivity leaves a face with fewer than three vertices. It said the dropped face had no vertices, even when it had one or two, and now says it has fewer than three. Closes #2606. (@claude)

  • #2616: Fixed the dtype argument of geovista.common.wrap(). Its type annotation now accepts any data-type that NumPy understands, such as np.float32 or "f4", as its documentation always said. An integer dtype is now honoured: the longitudes are wrapped before they are converted, where they were truncated first and returned as float64. Closes #2614. (@claude)

  • #2619: geovista.geodesic.BBox treats crs=None as WGS84, as documented, where it raised CRSError, and npoints and npoints_by_idx treat npts=None as GEODESIC_NPTS, where they raised TypeError. Graticule labels accept a GraticuleGrid whose points or mask are lists, where they raised TypeError or ValueError. GeoPlotter.add_mesh, add_points, add_base_layer and add_coastlines are annotated to return None for an empty mesh, which they already did. geovista.geodesic and geovista.geoplotter also leave the typing ratchet, so mypy checks them in full. (@claude)

  • #2620: Fixed geovista.bridge.Transform.from_unstructured() raising ProjError for a masked point outside the domain of a bounded projection, such as an off-disk fill value in +proj=ortho. The points that it drops from their faces, when xs and ys carry the same mask, are no longer transformed. They stay in the mesh at the position of the first point that isn’t masked, so their fill values no longer stretch its bounds. Closes #2603. (@claude)

  • #2626: Fixed geovista.bridge.Transform.from_unstructured() transforming a masked point that no face uses, when the connectivity is given as an array. Such a point raised ProjError outside the domain of a bounded projection, such as +proj=ortho, and stretched the bounds of the mesh inside it. It is now left out of the transform when xs and ys carry the same mask, and stays in the mesh at the position of the first point that is transformed. Closes #2625. (@claude)

  • #2628: Fixed geovista.bridge.Transform.from_unstructured() and geovista.bridge.Transform.from_2d() building a face from the value under a mask. When xs and ys carry the same mask, a face of a connectivity array or of the grid that uses a masked point is now dropped, with a warning. Before, it took the fill value as a vertex, or raised ProjError outside the domain of a bounded projection. When any face is dropped, data with a value for each face keeps only the values of the faces that remain. Closes #2621. (@claude)

  • #2632: Fixed geovista.bridge.Transform.from_1d() ignoring a mask on its bounds. A masked bound was used as its fill value, or became a NaN point in the mesh, with no warning. It is a gap in a contiguous axis, so from_1d now raises a ValueError that names the axis and counts the masked bounds. Closes #2627. (@claude)

  • #2668: Fixed geovista.geoplotter.GeoPlotterBase.add_mesh() raising on a zlevel array when the mesh needs no transform, as a point cloud on a plotter with the same CRS does. geovista.core.resize() now takes a zlevel at each point of a cloud, and raises a ValueError for an array given with any other mesh, which needs a single radius on the sphere. It also no longer truncates a scalar zlevel to a whole number, which geovista.transform.transform_mesh() never did. Closes #2618. (@claude)

  • #2673: Fixed geovista.crs.to_wkt() failing for a CRS whose OGC WKT holds non-ASCII text, which pyvista refuses. That is 3372 of the 7468 EPSG CRSs, including Web Mercator (EPSG:3857), the British National Grid (EPSG:27700) and the UTM zones, so neither geovista.transform.transform_mesh() nor geovista.geoplotter.GeoPlotter could target them. The WKT is now stored escaped, and a mesh serialized before still reads. Closes #2595. (@claude)

  • #2676: geovista.core.combine() raises TypeError for a single mesh that isn’t a PolyData, where it returned the mesh unchecked, and geovista.search.find_nearest_cell() treats z=None as zero, as documented, where it raised TypeError on a projected mesh. geovista.crs.projected() returns a Python bool. geovista.core, geovista.crs, geovista.filters, geovista.gridlines and geovista.search also leave the typing ratchet, so mypy checks them in full. (@claude)

  • #2680: geovista.themes.GeoVistaTheme and geovista.themes.GeoVistaDocumentTheme compare with == and convert with to_dict() and save(), where each raised AttributeError. Like pyvista’s themes, they now refuse an attribute they don’t have, so a misspelt one raises AttributeError instead of being set silently. geovista.pantry and its modules, geovista.raster, geovista.report and geovista.themes also leave the typing ratchet, so mypy checks the whole library in full, and geovista.themes loses its five type: ignore comments. (@claude)

  • #2682: Fixed the graticule labels of geovista.geoplotter.GeoPlotter on a planar CRS, which ignored the zscale the graticule was given and so sat at a different height from their gridlines. Closes #2594. (@claude)

  • #2683: Fixed geovista.bridge.Transform.from_points() failing for 2D points with vectors, whatever the CRS of the vectors, and for 3D points with or without them, although it documents both. The points of a cloud are now taken as one sequence, and a zlevel array still broadcasts against their shape. Closes #2604 and #2605. (@claude)

  • #2684: Fixed geovista.core.combine() returning a single mesh as it was, without the checks, clean or data it applies to two or more. A single mesh is now combined into a new mesh like any other. A mesh with vertices, such as a point cloud, or with strips, is now refused as one with lines is, where before its points or strips were dropped without a word. Each array combined now keeps the role it plays in the first mesh, such as its active vectors, normals or texture coordinates. Closes #2679. (@claude)

  • #2685: Fixed geovista.core.combine() dropping a point array that shares its name with a cell array, and raising for one that shares it with a field array, which also cost geovista.core.slice_cells() the point array. Fixed geovista.core.slice_cells() returning the vtkOriginalPointIds and vtkOriginalCellIds arrays that VTK adds, which geovista.transform.transform_mesh() then passed on for a global mesh, and raising a KeyError for a mesh whose active scalars were one of them. Closes #2675. (@claude)

  • #2686: Fixed geovista.geoplotter.GeoPlotterBase.add_mesh() refusing a zlevel at each point of a surface that crosses the seam of a CRS it transforms the surface to. The levels are now carried through the seam, as geovista.transform.transform_mesh() carries them through its own. Closes #2669. (@claude)

  • #2688: Fixed geovista.common.from_cartesian() reading a surface at the mean radius of its points, which gave the wrong latitudes for one whose points sit at uneven radii, such as a surface geovista.transform.transform_mesh() lifts by a zlevel at each point onto the sphere. Each point of such a surface is now read at its own radius. Closes #2670. (@claude)

  • #2689: Fixed geovista.crs.get_central_meridian() giving the central meridian of a projected CRS east of its own prime meridian rather than of Greenwich, which made geovista.transform.transform_mesh() cut a mesh at the wrong seam for the 59 EPSG CRSs on a prime meridian such as Ferro or Paris. geovista.crs.set_central_meridian() takes the prime meridian off again, so the two still round-trip. Closes #2518. (@claude)

  • #2715: Fixed the points that geovista.core.slice_cells() adds where a meridian crosses the interior of cells, which were left inside the sphere, along the chords of those cells. A textured surface added to a GeoPlotter with no transform and no zlevel was cut this way, leaving a seam below its surface. Each added point now takes the radius interpolated from its neighbours. (@claude)

  • #2718: Fixed a mesh projected to a CRS on another datum, such as the British National Grid, taking a different datum operation at different points, which left a quiet step of 129 m where a datum grid ended, and taking each datum shift 2° from where it belonged, which left the rest about 15 m out. geovista.transform.transform_points() now takes one operation for every point, the most accurate that covers them, and warns when none does, and geovista.transform.transform_mesh() and geovista.geoplotter.GeoPlotterBase.add_mesh() take each shift where its point is. Closes #2714. (@claude)

  • #2777: geovista.filters.remesh(), and so every cut along a meridian through the cells of a mesh, now interpolates the point data of each point it adds over the triangle that point lies in. vtkIntersectionPolyDataFilter interpolated some of them from the wrong points. (@claude)

  • #2844: Fixed geovista.common.to_cartesian() giving a real point for a masked longitude or latitude, such as [1, 1, 1] when both are masked. A masked coordinate is now read as NaN, so the point it gives is missing too. (@claude)

  • #2889: Fixed geovista.search.KDTree.query() matching a masked point-of-interest to the cell under its mask, and a NaN one to an index past the end of the tree. A point whose longitude or latitude is masked or NaN now has no neighbour: its distance is inf and its index is the number of points in the tree, as for a point beyond distance_upper_bound. geovista.search.find_nearest_cell() and geovista.geoplotter.GeoPlotterBase.view_poi() refuse such a point, naming the coordinate, and all three refuse an infinite coordinate. This lands row 1 of positions spec §9. (@claude)

  • #2893: Fixed geovista.geodesic.line() building a line almost all of NaN points for a masked or NaN position, or raising a TypeError that named no argument, and geovista.geodesic.BBox and geovista.geodesic.wedge() building a box around a masked or NaN corner. All three now refuse a position whose longitude or latitude is masked or NaN, naming the argument and counting the missing positions, and refuse an infinite coordinate. When an infinite coordinate or height sits in an array of more than one dimension, the refusal now gives its index in each dimension, such as (1, 1). This lands row 2 of positions spec §9. (@claude)

  • #2899: Fixed geovista.bridge.Transform.from_points() drawing a point at the value under a mask, or a NaN point, for a position whose x or y value is masked or NaN, and raising a TypeError for a None. It now drops each such point before it transforms the rest, along with its entries in data, vectors, the heights and a zlevel array, and warns once, counting the points dropped. A cloud with no points, whether every point was missing or none was given, now keeps its vectors, where it raised a ValueError, and keeps its arrays when cleaned. geovista.geoplotter.GeoPlotterBase.add_points() cuts its scalars by the same points. geovista.geodesic.BBox now builds from corners given as a grid, taking them in order, where its mesh failed on an assertion. This lands row 3 of positions spec §9. (@claude)

  • #2905: Fixed geovista.geodesic.wedge() building a bounding-box for two equal longitudes, though the message of its width check asked for a width in the open interval (0, 180). It now refuses a wedge of no width with that message. This lands row 6 of positions spec §9. (@claude)

  • #2908: Fixed geovista.bridge.Transform building faces at the value under a mask, or with a NaN vertex, for a point whose x or y value was NaN or masked alone, or that was masked in both while the masks differed elsewhere. geovista.bridge.Transform.from_unstructured(), geovista.bridge.Transform.from_2d() and the constructor now drop the faces that use such a point, with a warning, as they already did for a point masked alike, and the constructor now cuts data given for each face to the faces that remain. Data cut this way stays on the faces. It used to move to the points whenever as many faces remained as there were points. The constructor also cleans each mesh after attaching its data, as geovista.bridge.Transform.from_unstructured() does, so it takes data for the points and faces given, and cleaning no longer moves the value of one face onto another. geovista.bridge.Transform.from_1d() refuses a NaN bound as it refuses a masked one, and geovista.bridge.Transform.to_structured_grid() refuses a masked or NaN position. Both refuse an infinite one too. This lands row 4 of positions spec §9. (@claude)

  • #2912: Fixed geovista.transform.transform_points() transforming the value under a mask, or raising beyond the domain of the target CRS, for a position whose x or y value was masked, and geovista.common.vectors_to_cartesian() computing the vector at that value. Both now give NaN for a missing position, in every component, as do geovista.common.to_cartesian(), whose z came from the latitude alone, geovista.transform.transform_point(), geovista.geodesic.npoints() and geovista.geodesic.npoints_by_idx(). Each refuses an infinite coordinate, naming it. transform_point() and npoints() read None as missing, where they raised a TypeError, and a masked end of npoints no longer warns. geovista.bridge.Transform.from_points() refuses a zlevel array given with vectors before it drops any point, so it names the shape that was given. This lands row 5 of positions spec §9. (@claude)

  • #2940: Fixed geovista.core.slice_mesh(), geovista.core.slice_cells() and geovista.transform.transform_mesh() writing the gvOriginalCellIds and gvOriginalPointIds arrays onto the mesh they were given. geovista.core.MeridianSlice now traces each cell and point on a shallow copy of that mesh, and attaches each id to the cells or the points by name, so the mesh given is left as it was. What each function returns is unchanged. (@claude)

🔗 Dependencies#

  • #1000: Skip the tests.bridge.test_from_tiff tests if the rasterio optional package dependency is not installed. (@bjlittle)

  • #1001: Skip individual geovista.examples image tests if their optional package dependencies are not installed. (@bjlittle)

  • #1013: Introduced minimum pin sphinx-tags >=0.4 for multi-line tags support. (@bjlittle)

  • #1019: Introduced minimum pin sphinx-autoapi >=3.0.0 and purge use of obsolete autoapi_type configuration option. (@bjlittle)

  • #1020: Added the sphinx-changelog package dependency for rendering a changelog from news fragments using towncrier and generating a release changelog. (@bjlittle)

  • #1039: Introduced temporary maximum pin towncrier <24.7.0 due to breaking change for sphinx-changelog. (@bjlittle)

  • #1040: Bumped mambaforge from 22.9 to 23.11 for readthedocs and introduced minimum pin lazy_loader >=0.4. (@bjlittle)

  • #1051: Introduced minimum pin sphinx-autoapi >=3.2.0. (@bjlittle)

  • #1067: Unpinned towncrier and introduced minimum pin sphinx-changelog >=1.6.0. (@bjlittle)

  • #1118: Adopted Scientific Python SPEC-0, a common time-based policy for dropping dependencies. This replaces the NEP-29 deprecation policy. (@bjlittle)

  • #1157: Introduced temporary minimum pin h3-py ==3.7.7 to avoid breaking major release 4.x behaviour. (@bjlittle)

  • #1178: Dropped the optional pyarrow dependency as pandas >2.2.0 no longer issues a DeprecationWarning. (@bjlittle)

  • #1188: Introduced minimum pin rasterio >=1.4.2 to support gdal >=3.10.0. (@bjlittle)

  • #1242: Enabled sphinx.ext.duration to assist with identifying documentation building bottlenecks. (@bjlittle)

  • #1243: Adopted numpy >=1.25 minimum pin to comply with SPEC-0 package deprecation policy. (@bjlittle)

  • #1255: Avoid using pydata-sphinx-theme 0.16.1, see #1247. (@bjlittle)

  • #1276: Introduced temporary maximum pin pydata-sphinx-theme <0.16.0, see #1247. (@bjlittle)

  • #1372: Introduced minimum pin sphinx-gallery >=0.17 for README.rst to GALLERY_HEADER.rst migration. (@bjlittle)

  • #1428: Introduced minimum pin setuptools >=77.0.3 to support PEP 639 changes to license and license-files fields in the pyproject.toml. (@bjlittle)

  • #1494: Introduced minimum pin pyvista >=0.45.0 and dropped the pytest plugin pytest-xvfb for headless display support. (@bjlittle)

  • #1496: Extended distribution support to cover py313. (@bjlittle)

  • #1515: Introduced minimum pin ruff >=0.12.0. (@bjlittle)

  • #1561: Introduced pin trame-vtk ==2.8.15 to avoid subplot interactive scene behaviour regression. (@bjlittle)

  • #1572: Added pytest-pretty plugin for pretty printing of the pytest summary. (@bjlittle)

  • #1730: Bumped pixi dependencies. (@bjlittle)

  • #1760: Added vtk-xref sphinx extension for linking directly to VTK’s documentation using the :vtk: reference role. (@bjlittle)

  • #1763: Refreshed the geovista.report.Report package dependencies. (@bjlittle)

  • #1808: Added sphinxcontrib-mermaid sphinx extension for rendering embedded mermaid diagrams. (@bjlittle)

  • #1943: Added sphinxcontrib-bibtex sphinx extension for BibTeX style citations and references. (@bjlittle)

  • #2064: Bumped the minimum pin pyvista >=0.47.0. (@bjlittle)

  • #2072: Refreshed the geovista.report.Report to reflect the package migration from nest-asyncio to nest-asyncio2 for pyvista >=0.47.0. (@bjlittle)

  • #2128: Migrated from the obsolete libgl1-mesa-glx package to the modern equivalent libgl1 and libglx-mesa0. (@bjlittle)

  • #2183: Added the geocoding package geopy as a core dependency. (@bjlittle)

  • #2199: Added sphinx-sitemap sphinx extension to generate sitemaps.org compliant documentation sitemaps. (@bjlittle)

  • #2202: Added sphinx-iconify sphinx extension to use iconify-icon web component. (@bjlittle)

  • #2240: Added the check-manifest package dependency to the devs pixi 🧚 feature. (@bjlittle)

  • #2254: Added the jupyterlab package dependency to the devs pixi 🧚 feature, and added the wurlitzer package dependency to the docs pixi 🧚 feature. (@bjlittle)

  • #2259: Bumped the minimum pin pyvista >=0.48.0. (@bjlittle)

  • #2266: Adopted trame-pyvista package as a managed replacement for trame, trame-client, trame-server, trame-vtk and trame-vuetify. (@bjlittle)

  • #2511: Temporarily pin cartopy <0.26.

    cartopy 0.26 redefines cartopy.crs.PlateCarree from +proj=eqc to +proj=latlong, which flips crs.is_projected from True to False. geovista branches on crs.is_projected to decide whether a scene is rendered as a flat plane or a 3D sphere, so a GeoPlotter(crs=ccrs.PlateCarree()) scene placed its base layer and coastlines at the same z-level as the data mesh, hiding the data behind the base layer. (@claude)

  • #2520: Introduced a pyproj >=3.8 minimum pin, which guarantees PROJ >=9.8 from both conda-forge and the PyPI wheels. PROJ 9.8 changed +proj=eqc to the ellipsoidal EPSG:1028 method, which previously required a directory of baseline image variants for each affected test, along with bespoke fixture machinery to adjudicate between them. All 11 variant directories are now collapsed to a single baseline, and the verify_image_cache fixture override is retired, which also restores the default error threshold to the tests that carried it. (@claude)

  • #2522: Lifted the temporary cartopy <0.26 pin and introduced a cartopy >=0.26 minimum pin. cartopy 0.26 redefines ccrs.PlateCarree() from +proj=eqc to +proj=latlong, which flips pyproj.crs.CRS.is_projected and would otherwise have rendered a PlateCarree scene as a 3D sphere. See #2511. (@claude)

  • #2525: Bumped the pyvista max-pin to <0.50, which admits pyvista 0.49. The >=0.48.0 minimum pin is unchanged, as no 0.49 API is adopted here.

    pyvista 0.49 renames the output directory of its plot_directive sphinx extension from plot_directive to pyvista_plot_directive, so the doc_images_dir pytest-pyvista option is repointed accordingly. Without this the documentation image tests abort during pytest configuration, as the directory they scrape rendered scenes from no longer exists. (@claude)

  • #2528: Unpinned pydata-sphinx-theme, which #1247 had held at 0.15.4, and bumped sphinx-book-theme to >=1.4.0,<1.5. The two must move together, as every sphinx-book-theme release pins an exact pydata-sphinx-theme version: 1.4.0 requires pydata-sphinx-theme 0.20.0. (@claude)

  • #2533: Added playwright to the test dependencies, required by the new documentation theme chrome browser tests. It is sourced from PyPI, as the conda-forge package of the same name provides the node command line interface but no Python bindings. (@claude)

  • #2544: Migrated from vtk-xref to sphinx-vtk-xref and bumped it from 0.1.2 to 0.4.0. It is the same sphinx extension under a renamed PyPI distribution, so the conf.py extensions entry is renamed vtk_xref to sphinx_vtk_xref to match. The old name stopped at 0.1.2, so neither pixi nor dependabot could see the three releases that followed under the new one.

    0.1.2 warned whenever it could not reach vtk.org to validate a :vtk: class reference, cached that class as invalid, then warned again for every remaining reference to it, so one network blip was enough to fail the documentation build under --fail-on-warning. 0.4.0 reports an unreachable check as transient and uses the class URL unvalidated. Closes #2541. (@claude)

  • #2561: Added support for python 3.14, which is now built and tested alongside python 3.13. (@claude)

  • #2599: Added typeshed’s types-rasterio stubs to the devs environment, so mypy checks geovista.bridge against rasterio’s own interface. (@claude)

  • #2631: Added the types-shapely and types-click-default-group stubs from typeshed, and pandas-stubs from pandas, to the devs environment, so mypy can check geovista’s use of those libraries against their real interfaces. Its use of shapely and click-default-group is checked now; pandas follows when geovista.pantry.data leaves the typing ratchet. geovista.cache, geovista.cli, geovista.config, geovista.geometry and geovista.qt leave it in this change. (@claude)

  • #2729: The pixi ceilings of sphinx-gallery and trame-pyvista now match the pip ones, which dependabot had raised, so the lock moves to sphinx-gallery 0.22.1 and trame-pyvista 0.2.3. sphinx-gallery 0.22 ignores reference_url, which conf.py drops, and is the new floor on both sides. (@claude)

  • #2761: Bumped the minimum pin pyvista >=0.49.0, the first release that keeps the gv namespace of a mesh copied with copy.copy apart from the original’s. (@claude)

  • #2783: The earthquakes sample reads its parquet with pyarrow in place of fastparquet, whose pandas engine pandas 3.1 deprecates. The exam extra installs pyarrow >=13.0.0 instead of fastparquet. (@claude)

  • #2985: Every pip requirement now declares a floor, equal to its pixi floor, so pip install geovista no longer accepts a release older than the ones the tests floors check runs against. Before this, 58 of the 71 requirement lines had no floor. The new Dependency Floors check, tests/test_dependency_floors.py, keeps the two sides equal (floors spec §4.4). (@claude)

💽 Assets#

  • #1185: Updated the geovista.cache to reference geovista-data =2024.10.2, which includes the corrected NASA Blue Marble texture. (@bjlittle)

  • #1303: Updated the geovista.cache to reference geovista-data =2025.1.4, which includes new vector data assets. (@bjlittle)

  • #1765: Updated the geovista.cache to reference geovista-data =2025.10.2, which includes the documentation test image assets. (@bjlittle)

  • #1886: Updated the geovista.cache to reference geovista-data =2025.12.0, which includes the rehosted FESOM time-series decimated to the first time-step. (@bjlittle)

  • #1908: Updated the geovista.cache to reference geovista-data =2025.12.1, which includes a new BBox image unit test asset. (@bjlittle)

  • #1977: Updated the geovista.cache to reference geovista-data =2026.01.1, which includes new geovista.transform.transform_mesh() and geovista.geoplotter.GeoPlotterBase.add_mesh() image unit test assets. (@bjlittle)

  • #1982: Updated the geovista.cache to reference geovista-data =2026.01.2, which includes new geovista.geodesic.BBox.enclosed() image unit test assets. (@bjlittle)

  • #2033: Updated the geovista.cache to reference geovista-data =2026.02.0, which includes the new Plotter Extraction image unit test asset. (@bjlittle)

  • #2035: Updated the geovista.cache to reference geovista-data =2026.02.1, which includes updated point-cloud and DYNAMICO image unit test assets. (@bjlittle)

  • #2064: Updated the geovista.cache to reference 2026.02.5, which includes updated documentation image test filenames as generated by the pyvista-plot directive for pyvista >=0.47.0. (@bjlittle)

  • #2425: Updated the geovista.cache to reference geovista-data =2026.08.1, which replaces the single +proj=eqc baseline image test assets with a directory of baseline image variants, one for PROJ <9.8 and one for PROJ >=9.8. PROJ 9.8 adopted the ellipsoidal Equidistant Cylindrical method (EPSG:1028) for +proj=eqc, which sub-pixel shifts the render of every eqc image test. (@claude)

📚 Documentation#

  • #1006: Added sphinx Internationalization support. (@tkoyama010)

  • #1009: Enabled rendering a version badge in the primary sidebar, and a latest to stable version banner announcement for documentation on readthedocs. (@bjlittle)

  • #1010, #1012: Updated the version announcement banner for readthedocs. (@bjlittle)

  • #1013: Refactored structure of geovista.examples sphinx-tags. Closes #823. (@bjlittle)

  • #1014: Moved the tags link from the primary sidebar to a sphinx-design card in the Reference section. (@bjlittle)

  • #1016: Added changelog and whatsnew sphinx-design cards to the Reference section. (@bjlittle)

  • #1022: Fixed unset author in page footer of documentation. (@bjlittle)

  • #1023: Renamed the runtime generated examples gallery carousel from gallery-carousel.rst to gallery_carousel.rst in order to comply with implicit documentation file naming convention. (@bjlittle)

  • #1025: Added initial towncrier infrastructure to support the changelog. (@bjlittle)

  • #1026: Retro generated changelog entries for all relevant pull-request post 0.5.0 release. (@bjlittle)

  • #1053: Adopted sphinx-autoapi 3.2.1 templates. (@bjlittle)

  • #1054: Updated the Region Manifold Extraction tutorial to include the approximate article reading time and also shorten links within margin notes for readability. (@bjlittle)

  • #1055: Cross-referenced module headings to improve primary sidebar readability for the API Reference section. (@bjlittle)

  • #1056: Added the initial What’s New infrastructure. (@bjlittle)

  • #1057: Updated the What’s New to reference the associated release Discussion. (@bjlittle)

  • #1058: Clarified pending unreleased What’s New details as To Be Done (TBD). (@bjlittle)

  • #1070: Updated the towncrier changelog template. (@bjlittle)

  • #1072: Added Changelog contributor guidelines. (@bjlittle)

  • #1073: Rebranded the CLI icon from to . (@bjlittle)

  • #1075: Introduced CONTRIBUTING.md. (@bjlittle)

  • #1077: Updated README.md with introduction for Star History section. (@bjlittle)

  • #1080: Updated the Changelog contributor guidelines. (@bjlittle)

  • #1085: Added buttons to the reference About section. (@bjlittle)

  • #1096: Documented the changelog quality assurance performed by the ci-changelog Action. (@bjlittle)

  • #1099: Adjusted the Region Manifold Extraction tutorial to render better on reduced footprint mobile platforms. (@bjlittle)

  • #1109: Improved README.md accessibility. (@bjlittle)

  • #1152: Used explicit strings in enum.StrEnum instances (instead of auto()), thus making the documentation clearer. (@trexfeathers)

  • #1183: Adopted sphinx-autoapi 3.3.3 templates. (@bjlittle)

  • #1226: Resolved unnecessary linkcheck URL redirects. (@bjlittle)

  • #1269: Rebranded the changelog Font Awesome icon. (@bjlittle)

  • #1280: Added a convenience link in the GeoVista 101 sidebar to the changelog. (@bjlittle)

  • #1284: Added a gallery example for region manifold extraction using geovista.geodesic.wedge(). Closes #1285. (@ESadek-MO)

  • #1294: Updated the docstring of the region manifold extraction example. (@bjlittle)

  • #1305: Added a gallery example using pyvista.PolyDataFilters.contour_banded(). (@HGWright)

  • #1306: Added the new sample sphinx-tags category. (@bjlittle)

  • #1307: Renamed the gallery examples Region Extraction section to Extraction to maintain the primary sidebar menu in alphabetical order. (@bjlittle)

  • #1309: Added references to the new cast, contour and triangulate sphinx-tags for the filter category. (@bjlittle)

  • #1312: Added trame admonition for open Issue regarding contour_banded() interactive scene rendering. See Contour. (@bjlittle)

  • #1328: Removed use of the shadow option from all Examples titles and scalar bars, as shadowed text does not render correctly with trame in the Interactive Scene on readthedocs. (@bjlittle)

  • #1339: Simplified several Examples by using geovista.pantry.meshes instead of loading geovista.pantry.data to construct sample meshes. (@bjlittle)

  • #1362: Enabled tooltips for the Tutorials and README.md gallery carousel, and disabled sphinx-tippy tooltips for the gallery Examples and Tags. (@bjlittle)

  • #1364: Added a cubed-sphere panel region extraction gallery example using geovista.geodesic.panel(). (@bjlittle)

  • #1376: Added the Domain gallery section, including the Clouds and Clouds (Projected) examples. (@bjlittle)

  • #1379: Added the new resolution sphinx-tags category. (@bjlittle)

  • #1385: Reduced the Clouds and Clouds (Projected) examples by rendering only a single cloud mesh in order to reduce documentation build time. (@bjlittle)

  • #1388: Updated gallery Examples to use an outlined and filled scalar bar to aid readability. The scalar bar is also repositioned and resized by geovista.theme. (@bjlittle)

  • #1391: Enabled sphinx-build optimisation to distribute its workload automatically across the number of available CPUs. (@bjlittle)

  • #1395: Enabled sphinx-gallery to distribute its workload automatically across the number of available CPUs. Disable by setting the GEOVISTA_SPHX_GLR_SERIAL environment variable. (@bjlittle)

  • #1424: Converted pyvista documentation URLs to use snake_case post pyvista v0.45.0 release. Reference pyvista #7169. Closes #1423. (@bjlittle)

  • #1518: Added support for optional pyvista plot directives of inline documentation to streamline building execution time. Includes the new make html-inline command. Closes #1150. (@bjlittle)

  • #1567: Updated README.md and Installation section to include pixi 🧚 instructions. Closes #1476. (@bjlittle)

  • #1579: Added the Pixi Workflow section to document the pixi 🧚 tasks available to clean, build and render the documentation. (@bjlittle)

  • #1583: Added introductions to each gallery example section. (@bjlittle)

  • #1606: Registered the console prompt ($) with sphinx-copybutton to automatically excluded it when copying console code blocks. (@bjlittle)

  • #1616: Added reference to pixi 🧚 install instructions. (@bjlittle)

  • #1628: Added a VHS tape recording to Installation showing a developer install with pixi 🧚. (@bjlittle)

  • #1631: Fixed VTKBook broken URL. Closes #1626. (@bjlittle)

  • #1633: Added a VHS tape recording to the README.md showing how to create a new pixi 🧚 project, install, activate then import geovista. (@bjlittle)

  • #1653: Rebranded the geovista title using La Machine Company TrueType font. (@bjlittle)

  • #1660: Updated the Installation section to include uv instructions. (@bjlittle)

  • #1669: Updated the intersphinx mapping for cartopy, which is now hosted on readthedocs. (@bjlittle)

  • #1676: Refreshed README.md plotting example images. (@bjlittle)

  • #1681: Refreshed Tutorials card image for Region Manifold Extraction. (@bjlittle)

  • #1684: Refreshed Key Bindings GIFs and static images. (@bjlittle)

  • #1762: Applied the :vkt: role from the sphinx extension vtk-xref to cross-reference items within the VTK documentation. (@bjlittle)

  • #1764: Adopted the pyvista document_build plotting theme style for our documentation. (@bjlittle)

  • #1818: Used the pyvista.CameraPosition to explicitly clarify the camera position vectors in the GeoTIFF DEM gallery example. (@bjlittle)

  • #1821: Enabled zooming and panning of mermaid diagrams in the Packaging guidelines. (@bjlittle)

  • #1860: Disabled zooming and panning of mermaid diagrams in the Packaging guidelines. (@bjlittle)

  • #1908: Enabled sphinx-autoapi documentation of special-members (aka dunder methods). (@bjlittle)

  • #1912: Added the icon in the primary sidebar as a navigation convenience to the pyvista documentation. (@bjlittle)

  • #1931, #1932: Added the geovista badge geovista-badge as a shields.io endpoint asset. (@bjlittle)

  • #1936: Added the Community reference section. (@bjlittle)

  • #1941: Reformatted the community Third-Party Packages as a table for readability. (@bjlittle)

  • #1945: Added the community Publications section with a bibliography to reference scientific publications using geovista. (@bjlittle)

  • #1960: Applied colour Branding to the Font Awesome icons. (@bjlittle)

  • #1962: Completed first and second level title Font Awesome icon coverage. (@bjlittle)

  • #2033: Added the Plotter Extraction gallery example. (@bjlittle)

  • #2035: Updated the scalar-bar formatting for README.md and gallery point-cloud and DYNAMICO examples. (@bjlittle)

  • #2082: Added rebranded logo to README.md. (@bjlittle)

  • #2114: Added rebranded Logo section to the contributor guidelines. (@bjlittle)

  • #2117: Added Badge section to the contributor guidelines. (@bjlittle)

  • #2164: Added the new version sphinx-tags category. (@bjlittle)

  • #2178: Added the experimental 🧪 tag for highlighting experimental features and content. (@bjlittle)

  • #2201: Create a sitemap.xml only for the stable version of the documentation on readthedocs to improve Search Engine Optimization and discoverability. (@bjlittle)

  • #2251: Added Installation section tab synchronization. (@bjlittle)

  • #2255: Used the wurlitzer IPython extension to remove third-party diagnostic output from the rendered myst-nb tutorial. (@bjlittle)

  • #2276: Adopted sphinx-llms-txt to generate Large Language Model (LLM) friendly Markdown summaries of the documentation. (@bjlittle)

  • #2307: Added the custom sphinx readingtime directive to automatically estimate the audience reading time of a page and render a branded banner. (@bjlittle)

  • #2312: Added the Sphinx Directives section to the contributor guidelines. (@bjlittle)

  • #2313: Added support for the optional words-per-minute argument to the custom sphinx readingtime directive. (@bjlittle)

  • #2528: Fixed the gallery carousel, where each card painted an empty body over its own background image, blanking the thumbnail. Since pydata-sphinx-theme 0.16 a card body is rendered even when it has no content, so the carousel cards, which carry a background image and nothing else, now suppress theirs. Closes #1247. (@claude)

    Fixed the primary and secondary sidebar toggles, which did nothing when clicked. Both sphinx-book-theme and pydata-sphinx-theme bind their handlers to the first .primary-toggle and .secondary-toggle button on the page, but sphinx-book-theme adds a second of each in the article header and hides the first, leaving the only buttons a reader can see inert. On a narrow viewport, such as a phone, that stranded every link in both sidebars, as they are off-canvas there with no other way to reveal them. (@claude)

    Fixed the primary sidebar dialog blocking the page when a narrow viewport was widened while it was open. It inherits the sidebar’s classes, so it also inherited the collapsed state that sphinx-book-theme applies on a wide viewport, which left it open but invisible. An open dialog holds the top layer, so nothing else on the page could be clicked. (@claude)

  • #2533: Added developer Browser Tests guidelines, covering the prerequisites, markers, fixtures and helpers of the new tests/docs suite, along with the previously undocumented ci-tests-docs.yml workflow and the tests-docs-browser pixi tasks. Also added a Theme section, explaining the sidebar_toggle.js shim that the suite exists to guard. (@claude)

  • #2538: Made the documentation build resilient to an upstream intersphinx outage. A copy of each inventory is now vendored and offered as a fallback, so that a documentation site which cannot be reached no longer fails our build with a warning for the inventory plus a cascade of nitpick misses for every cross-reference it would have resolved. A new ci-inventories.yml Action refreshes the inventories monthly. Closes #2517. (@claude)

  • #2549: Vendored the sphinx-tippy tooltip runtime. The extension leaves tippy_js pointing at @popperjs/core@2 and tippy.js@6 on unpkg. Both float on a major version, every reader of every page fetched them from a third party, and no tooltip appeared at all when that host was unreachable. Pinned copies of the two UMD bundles now sit under docs/src/_static/js/, where sphinx serves them with its own cache-busting query. Both are MIT licensed, so each is accompanied by its upstream notice in a sibling <bundle>.LICENSE.txt, which is published along with the documentation.

    sphinx-tippy also moved from a PyPI dependency to a conda-forge one, and the conf.py import guard that treated it as optional went with it. That guard let a missing extension strip every tooltip out of the build and still leave it green. (@claude)

  • #2550: Corrected which links are given a sphinx-tippy tooltip. Three of the skip patterns in docs/src/conf.py were matching nothing at all, so the sphinx-gallery thumbnails raised a tooltip that collided with the one sphinx-gallery renders itself, every sphinx-tags badge raised one that merely repeated its own text, and the sphinx-design card on the Tutorials index raised one carrying the heading and opening paragraph of the page behind it.

    The gallery and tag patterns now cover both of the shapes an href takes in a built page, bare and directory-prefixed, and sd-stretched-link has been restored to tippy_skip_anchor_classes, which replaces the defaults of the extension rather than extending them. (@claude)

  • #2551: Corrected the readingtime estimates, which were counting the markup of a page rather than the page. The words are now counted on the parsed document, so directive names, option lines, link targets, comments and substitution definitions no longer inflate the figure, and a body produced by another directive is included in it. Counting the source ran between 1.3 and 2.0 times the rendered word count, so every calculated estimate moves: Testing drops from 37 minutes to 22, and Installation from 6 to 3.

    The default rate is now 150 words-per-minute rather than 120, pitched below the 175 to 300 band that Brysbaert (2019) reports for adults reading English non-fiction, because these pages alternate prose with code and output that a reader works through line by line. The paper is cited on Documentation, and the bibliography on Publications is filtered so that a work the documentation cites for its own reasons is not listed there as a publication using geovista.

    The banner is built from docutils nodes rather than a raw HTML block, so a text, man or latex build carries the estimate too, and its stylesheet now asks the theme for colours the theme defines. It had been asking for --article-info-bg and --article-info-fg, which neither theme defines, and a custom property that resolves to nothing fails silently, so the banner published a transparent box. An argument the directive does not understand fails the build rather than being ignored. (@claude)

  • #2553: Adopted a stated policy for which pages carry a readingtime banner, and gated it. Every page a reader reads start to finish opens with one, placed at column zero after the title and before the first section heading, and carries exactly one. A page that is navigated rather than read carries none and is named in EXEMPT in tests/docs/test_readingtime_coverage.py, with the reason beside it. The exempt pages fall into four groups: the site root and the section landing pages, the reference pages whose body a directive generates, the lookup tables such as Glossary and Key Bindings, and Changelog, whose body is an include of a page that carries the banner already.

    The gate derives the pages it governs from docs/src instead of listing them, so a page is covered from the day it lands. It reads the sources as text and needs neither a browser, a build nor sphinx, so unlike the rest of tests/docs it never skips. Adding a page now means adding the banner or adding the reason.

    The docs/src/developer/changelog.md banner drops its literal 6, which the word count arrives at unaided and which would otherwise need revising by hand whenever the page changed. The 30 on Region Manifold Extraction stays, being an estimate the word count cannot make: its reader stops to run every snippet. (@claude)

  • #2554: Replaced the retired YouTube badge on the README.md with a static one. shields.io has withdrawn its whole YouTube family: subscribers, channel views, video views, likes and comments each render the words retired badge now, so there was no live-count endpoint left to move to. The replacement matches the Bluesky badge beside it, and carries the @geovista_devs handle in place of a subscriber count. It has no upstream API behind it, so nothing can retire it in turn. (@claude)

  • #2555: Simplified how the documentation cross-references the top-level pyvista package. Its intersphinx inventory carried no entry for the top-level module itself, so each reference had to be written as a :doc: link to the index page with the display text supplied by hand, and a matching nitpick_ignore_regex entry in docs/src/conf.py silenced the :mod: role that could not resolve. pyvista#5663 has since added that entry, so the four docstring references in src/geovista/themes.py and src/geovista/bridge.py and the one on Environment Variables are now ordinary :mod: roles, and the ignore entry has gone with them. (@claude)

  • #2559: Dropped the sphinx-iconify extension and vendored the four icons it drew. The extension put a loader script from code.iconify.design on 171 of the 173 built pages, and the <iconify-icon> element that script defines then resolved each icon against api.iconify.design as the page was read. Only the first of those two hops is visible in the HTML, so vendoring the script by itself would have left the icons still coming off the network.

    Those icons are now four SVG files under docs/src/_static/icons/, reached through substitutions in common.txt and sized to the surrounding text by a gv-icon rule in style.css. They render as they did before. Three are MIT licensed and the fourth is CC BY 4.0, so each is accompanied by its upstream notice in a sibling <icon>.svg.LICENSE.txt, which is published along with the documentation.

    No page in the build now loads a script or a stylesheet from a third party. (@claude)

  • #2662: The image tests section of the developer guide now explains that CI renders with Mesa’s llvmpipe software rasterizer, and how to check for and select it locally, so that new baseline images match CI and a local comparison isn’t thrown by a hardware GPU. (@claude)

  • #2663: In dark mode, the icon on each card of the landing page and of the developer, reference and next steps pages now sits on the colour of its card instead of a darker strip. The cards had looked uniform only because pydata-sphinx-theme 0.15.4 asked for a colour it never defined, which 0.20.0 replaced with a real one. (@claude)

  • #2712: Added a title style to the documentation guide: Chicago Manual of Style headline style, with prepositions of four letters or more capitalized, as tephpy titles its pages. The headings of the design specifications now follow it. (@claude)

  • #2730: The repository checks specification no longer keeps its own list of the checks. Its table in checks spec §1 is now the eight the design started from, and the repository checks page, which the gate keeps complete, is the inventory, so a new check needs no change to the specification. (@claude)

  • #2731: The published specifications design now names, in docs spec §3.2, the citation namespace table of the specifications index as the list of prefixes in use, rather than listing the two it began with, so a new specification adds its prefix there. (@claude)

  • #2734: Added the citation links design, a child of the published specifications design. It links each citation of a specification section at build time, from one plain-text form, and adds a guard that fails a renumbered section’s untouched citations, with a nightly audit behind it. (@claude)

  • #2738: Citations of a design specification section are now links wherever the documentation renders them in prose, in a page, a specification or the API reference. A build-time transform (citation spec §4.2), ported from tephpy, links each one from the plain text it is written as, and tests/docs/test_rendered_citations.py checks the built site for any citation left unlinked, or linked to an anchor that is not there. The ceiling on citations outside the specifications, which held this work back, retires. (@claude)

  • #2745: Every citation of a design specification section is now written in plain text, which the documentation build links, rather than as a hand-written role whose display text and target could disagree (citation spec §4.4). The 267 roles became the citations they displayed, the built links unchanged, and the conventions test now refuses a role that names a section. (@claude)

  • #2778: Grouped the specifications by family, as docs spec §3.7 sets out. Each banner names its lineage and its children, the index lists each child under its parent, and every specification is titled for its topic. (@claude)

  • #2827: The title style now covers table titles and column headers, and the column headers of the design specifications follow it. (@claude)

  • #2902: Each test file a design specification’s Verification table names as evidence now links to the first line in it that cites the decision it verifies, so a reader can follow a decision to the test that holds it. The links point into the commit the documentation was built from, or the tag of a release, so every version of the documentation shows its own tests (ledger spec §2 decision 18, row 9 of ledger spec §8). (@claude)

  • #2917: The specifications index now shows each specification’s status, worked out from its roadmap: 📋 not started, 🚧 how many of its rows have landed, or ✅ complete, with any 💡 candidates counted beside it. A reader can see which specifications have work left without opening each one, and nobody keeps the column up to date by hand. This lands row 7 of docs spec §4. (@claude)

  • #2980: The conventions test now checks the figures of the design specifications. Each figure sits in its specification’s own folder, is a PNG of 150 KB or less, and is shown with alt text and a caption, and a figure file that nothing shows fails the test, as does a folder left behind by a renamed specification. This lands row 8 of docs spec §4. (@claude)

  • #2990: A design specification now writes the order between its roadmap rows in the row’s Scope cell, as · 🔗 after row 1, in one of four relations, and the conventions test holds the form and refuses an ordering written in prose. Every ordering the specifications held in prose has moved into its cell. This lands row 1 of order spec §9. (@claude)

  • #2994: The conventions test now reads the orderings of every roadmap together, as one graph, and refuses a cycle, a 📋 not started row waiting on a 💡 candidate, and a status the order contradicts, such as a row in progress after one that has not landed. Its check of rows landing together becomes one case of that last rule. For each row in progress, the verification report lists any row it lands with that still owes open items, and names those items, which its pull request must answer. When the test finds an ordering in prose, its message also suggests rewording the sentence. This lands row 2 of order spec §9 and row 8 of items spec §8. (@claude)

  • #3001: Each design specification whose roadmap rows are ordered now shows them as a diagram under its roadmap table, with a key beneath it. Each row is coloured by its status, the rows ready to start are green, and each box links to its row. The specifications index gains an Open Work section that draws every roadmap’s work still to do. The build draws them all, from the same reading of the orderings that the conventions test checks, so nothing is written for them. This lands row 2 of diagram spec §9. (@claude)

  • #987: Updated the Region Manifold Extraction tutorial to correct typos and add a cross-reference to pyvista.DataSetFilters.select_enclosed_points(). (@bjlittle)

  • #994: Adopted the standard of naming the plotter as p throughout documentation, docs-strings and examples. (@bjlittle)

💼 Internal#

  • #1036: Adjusted the number of contributors rendered per line to better suit reduced footprint on mobile platforms. (@bjlittle)

  • #1040: Adopt nodefaults to explicitly ignore the conda defaults channel when building the conda lock files. (@bjlittle)

  • #1050: Fixed typos and extended the list of files to skip with codespell. (@bjlittle)

  • #1075: Applied fix to avoid breaking change behaviour to code coverage upload, as introduced by the actions/upload-artifact v4.4.0 Action. (@bjlittle)

  • #1076: Introduced SECURITY.md. (@bjlittle)

  • #1081: Added missing concurrency to Actions. (@bjlittle)

  • #1082: Added ci-changelog Action to perform quality assurance on pull-request changelog news fragment/s. Closes #1031. (@bjlittle)

  • #1087: Reduced the all-contributors image size for better rendering on mobile platforms. (@bjlittle)

  • #1090: Report to the console by default for the ci-changelog Action. (@bjlittle)

  • #1091: Use SHA checksums for 3rd party Action versions. (@bjlittle)

  • #1094: Add the skip-changelog label to dependabot and ci-locks Action generated pull-requests. (@bjlittle)

  • #1120: Removed technical debt by migrating builtin open() usage to pathlib.Path.open() since enforcing SPEC-0 minimum python>=3.11 support. (@bjlittle)

  • #1121: Updated dependabot configuration to include composite actions. (@bjlittle)

  • #1133: Merged back 0.5.x release feature branch into main for 0.5.2. (@bjlittle)

  • #1136: Changed the frequency of the ci-locks Action cron to be every 2nd day to assist with earlier detection of breaking regressions in package dependencies, see #1131. (@bjlittle)

  • #1137: Removed Action duplication by migrating common job level defaults definitions to the workflow level. (@bjlittle)

  • #1138: Added ci-locks-test Action to assist with early detection of breaking regressions in package dependencies by running nightly tests against the latest resolved conda-lock environment. (@bjlittle)

  • #1139: Added lock file artifact upload for ci-locks-test Action to allow contributor reproducibility of any regression behaviour. (@bjlittle)

  • #1141: Configured setup-miniconda Action to avoid using the defaults channel. (@bjlittle)

  • #1144: Refactored Preference enums to use enum.StrEnum and enum.auto as py311 is now the minimum supported distribution, see #1118. (@bjlittle)

  • #1155: Reverted use of the jwalton/gh-find-current-pr Action within the ci-changelog Action. (@bjlittle)

  • #1158: Added ci-tests-pypi Action to assist with early detection of breaking regressions in package dependencies by running nightly tests against the latest resolved PyPI environment. (@bjlittle)

  • #1174: Merged back 0.5.x release feature branch into main for 0.5.3. Closes #1156. (@bjlittle)

  • #1219: Avoid using a pyvista private attribute when gathering active scalars information. (@user27182)

  • #1274: Removed a __all__ duplicate entry in geovista.pantry.data. (@HGWright)

  • #1318: Added a py.typed file in compliance with PEP 561. (@trexfeathers)

  • #1398: Migrated from sphinx-linkcheck builder to cron activated lychee Action. The link check workflow is no longer performed as part of a pull-request. (@bjlittle)

  • #1409: Added missing optional packages h3-py and rasterio to the geovista.report. (@bjlittle)

  • #1411: Added https://earthexplorer.usgs.gov/ to the list of URLs ignored by the lychee Action. Closes #1410. (@bjlittle)

  • #1438: Enabled caching and updated the list of URLs ignored by the lychee Action. Closes #1435. (@bjlittle)

  • #1516: Dropped use of deprecated ActiveArrayInfoTuple in the geovista.core. (@bjlitte)

  • #1547: Enforced an open-pull-request-limit for automated dependabot generated pull-requests. (@bjlittle)

  • #1564: Included trame-common package to the scooby geovista.report.Report. (@bjlittle)

  • #1597: Added nodefaults channel to lock2yaml template. (@bjlittle)

  • #1750: Dropped support for py311 as recommended by SPEC-0. (@bjlittle)

  • #1754: Configured the infrastructure to support unit test and documentation test image caching using pytest-pyvista. (@bjlittle)

  • #1777: Added ci-tests-docs Action to perform pytest-pyvista documentation image testing of pyvista-plot directive static scenes. (@bjlittle)

  • #1780: Added artifact upload of failed documentation test images for the ci-tests-docs Action. (@bjlittle)

  • #1850: Adopted explicit default permissions for Actions. (@bjlittle)

  • #1901: Endeavour to provision a community readable pooch.Pooch cache. (@bjlittle)

  • #1911: Updated ci-tests-lock and ci-tests-pypi Actions to capture and upload failed image unit tests as artifacts. (@bjlittle)

  • #1929: Added README.md light/dark theme support for our Star History Chart. (@bjlittle)

  • #1958: Applied consistent formatting to our labels. (@bjlittle)

  • #1997: Relaxed the regex checking for the existence of an author in new changelog fragments. (@ukmo-ccbunney)

  • #2110: Tidied the CLI entry-point to always route through geovista.__main__ and support direct script invocation. (@bjlittle)

  • #2179: Enabled generation of a generic conda environment YAML file that tracks the latest supported distribution of python. (@bjlittle)

  • #2247: Dropped the use of pyvista.start_xvfb when building the documentation. (@bjlittle)

  • #2425: Added the verify_image_cache fixture override in tests/plotting/conftest.py to collapse the pytest-pyvista error threshold onto the warning threshold for image tests which have a directory of baseline image variants in the cache. Without this a render which only exceeds the warning threshold is never compared against the remaining variants. (@claude)

  • #2434: Replaced the use of the private pyvista.PolyData._offset_array property in geovista.common.triangulated() and geovista.core.combine() with the new private geovista.common._face_offsets helper, which defers to the underlying vtkCellArray of the mesh instead. pyvista removed the property in pyvista#8873 when adopting fixed-size storage for regular cell arrays, which broke geovista in the pyvista downstream integration tests. (@claude)

  • #2463: The ci-locks workflow decided whether pixi update had changed anything by measuring the length of the rendered summary, which reads No diff — eight bytes — when nothing has moved. Against a threshold of ten that sat two bytes from passing an empty update through into a weekly pull-request with nothing in it. It now asks pixi update --json directly, reading the answer in an assignment so that errexit covers a missing interpreter rather than letting it read as “nothing moved”. (@claude)

  • #2467: Fixed the first-contribution greeting, which could not reach a contributor on a fork. The workflow triggered on pull_request, and GitHub withholds secrets from such a run when it is raised from a fork – which is where a first contribution comes from – so the bot token was empty exactly there. It now triggers on pull_request_target, which runs in this repository’s context; that is safe here because the job checks nothing out and runs no code from the pull-request. (@claude)

  • #2521: Documented the intersphinx CI failure mode in AGENTS.md. A 503 from any one of the 12 remote inventories fails the doctest and test images jobs, behind a cascade of unrelated nitpick warnings that reads like a regression in the pull request under test. See #2517. (@claude)

  • #2523: Added mypy to the devs environment, pinned to match the mirrors-mypy revision used by the pre-commit hook, and documented the geovista superset environment along with the local pre-commit workflow in AGENTS.md. Note that pre-commit run mypy remains the authoritative type check, as the hook runs in an isolated environment without the third-party libraries. (@claude)

  • #2529: Replaced the tests UniformGrid compatibility fallback with a direct from pyvista import ImageData post pyvista >=0.48 floor raise. (@HarshRajSinghania)

  • #2533: Added tests/docs, a playwright suite that drives a headless browser over a built documentation site to cover the theme chrome that no Python test can reach: the sidebar toggles, the dialogs they open, and the gallery carousel. The regressions fixed in #2528 were all rendered defects, so none of them had a test. Every prerequisite skips rather than fails, so a plain pytest run is unaffected, whilst the new --browser-strict option requires them instead. Without it, a regression that stops the carousel being generated would skip its tests and leave CI green over the lost coverage. Closes #2531. (@claude)

  • #2690: Tidied two annotations left by typing change 5 ( #2619): the keywords of three GeoPlotterBase methods, and the return type of geovista.geodesic.npoints() and geovista.geodesic.npoints_by_idx(), which promised one longitude and one latitude. Closes #2623. (@claude)

  • #2691: Tidied the minors left by the final review of typing change 3 ( #2599), in geovista.bridge.Transform and its tests. A vectors given to geovista.bridge.Transform.from_points() as a generator now raises the ValueError that states the rule, rather than a bare TypeError. Closes #2607. (@claude)

  • #2721: Added the projection-coverage repository check, row 1 of crs spec §10, the globe-first CRS roadmap. It projects a global mesh through geovista.transform.transform_mesh() to each of PROJ’s 162 projection methods and cartopy’s 38 projection classes, records which take it cleanly, and fails when an outcome changes without that record. (@claude)

  • #2729: tests/test_dependency_ceilings.py checks that every pixi ceiling in pyproject.toml admits at least what its counterpart in requirements/pypi-*.txt admits. dependabot raises only the pip ceilings, so a pixi ceiling left behind kept the lock on the old release, and CI never tested what a pip user could install. The check fails on the dependabot pull request itself, which is where the pixi bump now goes. (@claude)

  • #2735: The citation grammar that tests/test_spec_conventions.py reads the repository with now lives in docs/src/_ext/geovista_citations.py, so the documentation build and the renumber guard of the citation links design can share the one definition. Its own tests moved with it, to tests/docs/test_geovista_citations.py. Nothing it checks has changed. (@claude)

  • #2737: The citation grammar in docs/src/_ext/geovista_citations.py is now loaded once however it is reached, by the conventions test, its own tests or an import by name, instead of as a separate copy each time, whose objects never compared equal to the others’. (@claude)

  • #2740: A pull request that renumbers a specification’s sections now has to update every citation of the sections it moves or removes. A new check, the renumber guard of citation spec §4.5 in .github/scripts/check_renumbering.py, pairs each section with its counterpart on the base branch, even when it was renamed, and fails on each untouched line still citing an old number, released changelog entries included. pixi run -e test check-spec-renumbering fix rewrites the citations of moved sections. Each night it also audits every citation against how its specification has changed since the line was written, and numbered items and roadmap rows are held to being appended. (@claude)

  • #2744: The check of the rendered citations reports a broken link under a root reached by a symlink rather than failing on it, names where each unlinked citation sat, and explains a citation nested in two links. The citation transform leaves a >>> example alone, and its warning of a citation in a heading names the rule behind it. Citation links keep their tooltips, which show the section cited. (@claude)

  • #2751: The renumber guard of citation spec §4.5 reads a specification mid-edit without failing, ignores local diff settings, counts files not yet added as changed, follows a specification renamed with git mv, and reads what each commit held once. (@claude)

  • #2759: Added the readme-examples repository check. It runs every Python example in README.md as a test of its own, off screen, and fails when one raises or warns, so the README, which is also the project’s description on PyPI, keeps working as the public API changes. (@claude)

  • #2773: Replaced the private pyvista.core.filters._get_output with public API in cast_UnstructuredGrid_to_PolyData() and remesh(), as PyVista main now requires its output_type argument. (@claude)

  • #2788: The decision ledger’s checks read a table cell holding an escaped pipe whole, read a decision’s title the same way in the conventions test and the renumber guard, and report a decision with no title. (@claude)

  • #2789: The specification status check now asks GitHub about each decision ledger, as ledger spec §3.7 sets out. A decision’s pull request must have merged, a verified or retired row is dated by its pull request’s merge, and a sign-off holds only when the reviewer it names added the signed-off label to the pull request it cites. ci-spec-status.yml re-runs when a label is added or removed, and reports on every pull request. (@claude)

  • #2791: The specification status check now sees a specification moved out of the collection, and reads a sign-off’s reviewer (ledger spec §3.7) with the conventions test’s own pattern. (@claude)

  • #2795: The verification report passes over a plan that is not UTF-8, writes a status as GitHub Markdown, with #N and @login, and lists a line once however often it names a decision. (@claude)

  • #2798: Verified zlevel spec §2 decision 4 with a test that fails if a transform reuses a point cloud’s planar z instead of recomputing it, so the decision is no longer signed off. (@claude)

  • #2821: Settled the follow-ups to row 1 of items spec §8, the open items work. The renumber guard now passes over a changed text file that isn’t UTF-8 instead of failing, and reads the corpus for citations only once a specification has changed. The status checker and the citation grammar share one way to check a merge and read a section, and the build and the grammar are held to the same item and row anchors by new tests. (@claude)

  • #2823: Fixed the renumber guard’s reading of changed files whose names aren’t plain ASCII. A non-ASCII path is no longer dropped from the diff, so its changed lines are judged like any other file’s, and a file name that isn’t UTF-8 is read by the name Python gives it rather than ending the check. (@claude)

  • #2824: Fixed three edge cases in the renumber guard. It reports a stale citation in a file whose name isn’t UTF-8 instead of crashing, reads the changed lines of a file whose name git quotes, and no longer takes an added line that looks like a diff header for one. (@claude)

  • #2843: Added the pixi-task-tables repository check. It holds the “Pixi Tasks” tables of the developer pages to the tasks in pyproject.toml, and fails when a task has no row on the packaging page or on the page for its feature, or when a row names a task that no longer exists. (@claude)

  • #2903: Added tests that geovista.bridge.Transform.from_points() and geovista.geoplotter.GeoPlotterBase.add_points() refuse coordinates of different sizes as before, with no point dropped, a path that row 3 of positions spec §9 left untested. (@claude)

  • #2907: Made local builds of the documentation incremental again. conf.py turned notebook execution off after Sphinx had compared the configuration with the last build’s, so every build saw a changed setting and re-read all 166 pages. It now turns execution off as the configuration is read, and a rebuild with nothing changed re-reads 47 pages in about a quarter of the time. (@claude)

  • #2916: Stopped ci-spec-status cancelling its own runs when a pull request opens with labels, which left its status references check cancelled until the last run reported. The run in progress now finishes, and only a run still waiting for it is replaced. (@claude)

  • #2921: The last step of ci-linkcheck now reads lychee’s exit code from the environment, since zizmor v1.32.0 flags expanding it into the shell as a template injection. (@claude)

  • #2922: geovista.geodesic.line() is fast again. geovista.geodesic.npoints() and geovista.geodesic.npoints_by_idx() now read an end given as two plain numbers without building masked arrays, so a line of 2,000 vertices takes 0.14 s, against 0.39 s when they read each end through masked arrays. This lands row 7 of positions spec §9. (@claude)

  • #2925: geovista now benchmarks its geodesic module on every pull request and every push to main. CodSpeed counts the instructions each benchmark runs, and pytest-memray holds each one to a memory ceiling, as row 1 of bench spec §9 sets out. See Benchmarks. (@claude)

  • #2927: geovista now benchmarks building meshes and point clouds with Transform, and projecting them with transform_mesh() and transform_points(), as row 2 of bench spec §9 sets out. (@claude)

  • #2934: A benchmark’s memory ceiling now holds on any CI runner: OpenBLAS’s one-off buffers, 128 MiB under every kernel older than SkylakeX, are set up before any benchmark runs. (@claude)

  • #2936: geovista now benchmarks from_cartesian(), slicing, combining and resizing meshes in geovista.core, the KDTree and find_nearest_cell(), and the graticule of geovista.gridlines. With this last row, bench spec §9 is complete. (@claude)

  • #2941: Recorded in docs/AGENTS.md that conf.py changes a configuration value in config-inited, never builder-inited. Sphinx compares the configuration with the last build’s between the two events, so a value changed later reads as changed on every run and every page is read again, as #2904 found. (@claude)

  • #2974: Added a floors environment, which resolves each direct dependency at the lowest release its floor admits on the oldest supported Python, and a tests floors check that runs the suite in it on every pull request, landing row 1 of floors spec §9. The pyarrow floor rises to 18.0.0, its first release with a wheel for Python 3.13. (@claude)

  • #2999: The reading of the order between roadmap rows moved from the conventions test into the shared citation grammar, which the conventions test, the documentation build and the verification report all load. The grammar can now write a Mermaid diagram of the rows and their orderings, with its key, and say which rows are ready to start. This lands row 1 of diagram spec §9. (@claude)

  • #989: Added adRise/update-pr-branch Action to automatically update stale auto-merge pull-requests. (@tkoyama010)

  • #996: Enabled require_passed_checks for adRise/update-pr-branch Action. (@bjlittle)

  • #999: Merged back 0.5.x release feature branch into main for 0.5.1. (@bjlittle)

🌱 Community#

🛠️ Contributor Infrastructure#

  • #1005: Added the ci-joss workflow to automatically compile our JOSS paper from Markdown to PDF. (@tkoyama010)

  • #1008: Enabled the nbQA pre-commit hook to ensure tutorial jupyter notebook quality assurance against ruff rule compliance. (@bjlittle)

  • #1064: Dropped nbQA pre-commit hook, as ruff lints and formats Jupyter notebooks by default from v0.6.0. (@tkoyama010)

  • #1124: Dropped the use of tox to render lock file updates within the ci-locks Action due to incompatibilites between the pinned tox-conda plugin and the modern runner environment. (@bjlittle)

  • #1146: Optimised the documentation building workflow by providing control of whether tutorial jupyter notebook code cells are executed. By default, code cells will only be executed when building the documentation with either the make html or make html-tutorial commands. Closes #1030. (@bjlittle)

  • #1147: Adopted caching of tutorial jupyter notebook outputs when building the documentation. The cache will only be cleared with either the make clean-cache or make clean-all commands. The make clean command will not clear the cache. Closes #1018. (@bjlittle)

  • #1160: Updated dependabot to orchestrate updates for the pip package-ecosystem. (@bjlittle)

  • #1162: Triggered the ci-tests-pypi Action on a pull_request event in order to provide test coverage of dependabot package updates. (@bjlittle)

  • #1177: Extended ci-locks Action to support py313. (@bjlittle)

  • #1207: Added taplo pre-commit hook to format and sort toml files. (@tkoyama010)

  • #1211: Complied with ruff flake8-boolean-trap (FBT) and flake8-todos (TD) rules. (@bjlittle)

  • #1212: Complied with ruff flake8-annotations (ANN) rules. (@bjlittle)

  • #1214: Complied with ruff pylint-refactor (PLR) rules. (@bjlittle)

  • #1215: Complied with ruff pylint-warning (PLW) rules. (@bjlittle)

  • #1216: Complied with ruff flake8-self (SLF) rules. (@bjlittle)

  • #1217: Dropped the use of .ruff.toml to capture outstanding ruff rule exceptions. Non-compliances have been addressed or moved to the pyproject.toml. (@bjlittle)

  • #1225: Dropped sort-all pre-commit hook, as ruff 0.8.0 stabilized the unsorted-dunder-all (RUF022) preview rule. (@bjlittle)

  • #1227: Dropped the use of setup.py, given the already configured PEP 517 build-backend with setuptools. (@bjlittle)

  • #1241: Dropped use of latest tag and explicitly pinned to ubuntu-22.04 Action runner-image. (@bjlittle)

  • #1254: Updated the ci-label Action to migrate from nep-29 to SPEC-0. (@bjlittle)

  • #1257: Enabled the zizmor pre-commit hook for security static analysis of Actions. Closes #1248. (@tkoyama010)

  • #1277: Rationalised the use of sphinx cross-references. (@bjlittle)

  • #1279: Restructured the Contributing guidelines. (@bjlittle)

  • #1298: Added developer Documentation guidelines. (@bjlittle)

  • #1304: Added sphinx cross-references to the Examples. (@bjlittle)

  • #1313: Mirrored the pyvista PyPI package management strategy by adopting a rolling latest maximum pin, and rely on dependabot to orchestrate ecosystem updates. (@bjlittle)

  • #1317: Detect missing author :user: directive within a changelog news fragment. (@bjlittle)

  • #1324: Configured the conda-remove-defaults = true option for the setup-miniconda Action. (@bjlittle)

  • #1325: Converted pypi-publish Action options from snake_case to kebab-case. (@bjlittle)

  • #1356: Extended the validate-project pre-commit hook to use additional schema compliance checks for pyproject.toml. (@bjlittle)

  • #1388: Updated the testing theme background, font and colormap. (@bjlittle)

  • #1414: Updated the pygrep-hooks pre-commit hook to include additional coverage. (@bjlittle)

  • #1458: Enabled sphinx-lint pre-commit hook to quality check .rst and .py files. (@bjlittle)

  • #1495: Consolidated cartopy-cache composite Action to use a single common cache for all Actions. (@bjlittle)

  • #1497: Created the PyPI optional dependencies devs group and renamed the pixi 🧚 feature dev to devs. Also included the zizmor package to devs. (@bjlittle)

  • #1500: Migrated to actions/create-github-app-token Action from tibdex/github-app-token Action. (@bjlittle)

  • #1509: Adopted publishing binary wheels to PyPI with a Trusted Publisher via the OpenID Connect (OIDC) Standard. (@bjlittle)

  • #1513: Resolved various ruff ==0.12.0 rule non-compliances. (@bjlittle)

  • #1567: Created the PyPI optional dependencies all group. (@bjlittle)

  • #1578: Added the pixi 🧚 workflow tasks clean-all, clean-cache and doctest for the docs feature. (@bjlittle)

  • #1585: Added the pixi 🧚 workflow tasks test and test-clean for the test feature. (@bjlittle)

  • #1595: Migrated readthedocs to build with pixi 🧚. (@bjlittle)

  • #1609: Added the direnv .envrc configuration file to .gitignore. (@bjlittle)

  • #1612: Added developer Cross-References guidelines. (@bjlittle)

  • #1620: Updated the ci-label Action to auto-label pixi branches. (@bjlittle)

  • #1621: Added the latest supported distribution of python to the pixi 🧚 default environment. (@bjlittle)

  • #1634: Added developer Packaging guidelines. (@bjlittle)

  • #1655: Added developer Branding guidelines. (@bjlittle)

  • #1668: Enabled ruff flake8-copyright preview rule (CPY001) to enforce file copyright preamble. (@bjlittle)

  • #1671: Added developer Copyright & License guidelines. (@bjlittle)

  • #1703: Added developer Tagging guidelines. (@bjlittle)

  • #1718: Updated developer Packaging guidelines. (@bjlittle)

  • #1727: Updated developer Documentation guidelines. (@bjlittle)

  • #1770: Added the html-docstring-inline task as a documentation Build option to build and render only docstring and inline images and scenes within the documentation. (@bjlittle)

  • #1771: Building documentation will now automatically check, verify and create the docs/image_cache symbolic link the appropriate pooch.Pooch documentation reference test images cache. (@bjlittle)

  • #1776: Added the doc-images pixi 🧚 task to the docs feature which performs documentation image testing of pyvista-plot directive static scenes. (@bjlittle)

  • #1783: Rebranded the doc-images pixi 🧚 task as tests-docs. Also added Pixi Tasks to the Packaging guidelines. (@bjlittle)

  • #1787: Added descriptions to the pixi 🧚 tasks, available via the pixi task list command line interface (CLI). (@bjlittle)

  • #1810: Added embedded mermaid graphs of pixi 🧚 solve-group environment hierarchies to the Packaging guidelines. (@bjlittle)

  • #1820: Reduced the dependabot schedule interval for the pip ecosystem to be daily now that the cooldown period is configured. (@bjlittle)

  • #1824: Added @geovista-bot as a generic user independent codeowner. (@bjlittle)

  • #1830: Introduced the @geovista-bot auto-approve workflow to approve pull-requests by trusted contributors and thus permitting elevated main branch protection by enforcing CODEOWNERS reviews. (@bjlittle)

  • #1847: Reduced max-concurrency to a single thread when performing URL linkchecks with the lychee Action. This is to avoid exceptions caused by sporadic server-side connection resets. (@bjlittle)

  • #1858: Updated the Cross-References guidelines to include the use of the :name: option for a figure or table, and recommended the :alt: option for figure accessibility best practice. (@bjlittle)

  • #1876: Added developer Testing guidelines. (@bjlittle)

  • #1883: Relaxed dependabot to ignore patch releases for the PyPI ecosystem in order to reduce the update cadence and closer align to the default pixi 🧚 major/minor maximum capping-pin strategy for explicit dependency versions. (@bjlittle)

  • #1890: Capture failed image unit tests for the ci-tests Action as a workflow artifact, and also when using the tests-unit pixi 🧚 task. (@bjlittle)

  • #1965: Added the ci-first-contribution Action to automatically acknowledge and welcome first time Issue and Pull-Request contributions from the community. (@bjlittle)

  • #2008: Extended the @geovista-bot auto-approve workflow behaviour to allow trusted collaborators to request approval for bot sourced pull-requests. (@bjlittle)

  • #2020: Added developer Codecraft 🚧 maintenance guidelines. (@bjlittle)

  • #2061: Added developer Generative AI guidelines. (@bjlittle)

  • #2337: Added AGENTS.md for AI-assisted development guidelines across the project root, docs/ and tests/ directories. (@bjlittle)

  • #2338: Added the agentic auto-label for feature-branch names agentic, agent or ai. (@bjlittle)

  • #2537: Declared make as a dependency of the docs pixi 🧚 feature. Every task in that feature shells out to it, as do the tests-docs and tests-docs-browser tasks by way of depends-on, so a contributor whose host lacked make could not build, serve or test the documentation at all: each task exited 127 from its clean step. The runner images ship make, which is why CI never saw it. See #2535. (@claude)

  • #2550: Added the tests/docs/test_tooltips.py unit tests, which hold the sphinx-tippy configuration in place. Its two halves fail differently, and only one of them leaves a trace in the build: tippy_skip_urls is applied as a tooltip is generated, so a leak is visible in the payload of a page, whereas tippy_skip_anchor_classes is applied in the browser as a tooltip is attached, so a dropped class leaves the payload untouched and only a browser can tell. The unit tests read the built payload for the first and drive playwright for the second. They also cover the tooltip runtime vendored by #2549, which has to be served from the build itself and accompanied by its upstream licence notice. See Tooltips. (@claude)

  • #2551: Added the tests/docs/test_readingtime.py unit tests, which hold the readingtime sphinx extension in place. The reading-time model moved out to docs/src/_ext/reading.py, which imports nothing beyond the standard library, so the rate, the word pattern and the argument grammar are covered wherever pytest runs. The directive around it is covered by building a throwaway documentation set in-process, as tests/docs already does for intersphinx.

    The extension now declares itself safe for a parallel read and write. Without that, sphinx-build --jobs warns twice and drops back to a serial read, which --fail-on-warning turns into a build failure. It also notes both of its own modules as dependencies of every page it estimates, as neither is a document source that sphinx watches, and an incremental build would otherwise serve a pickled doctree carrying the previous figure. docs/src/conf.py puts _ext ahead of the environment on sys.path rather than behind it, since the modules there are imported by bare top-level name and an installed package of the same name would otherwise win. See Documentation. (@claude)

  • #2557: Separated the work of blacken-docs and ruff in .pre-commit-config.yaml. blacken-docs leaves >>> doctests untouched, so the docstring-code-format setting of ruff format was already the only thing formatting those, and blacken-docs is the only thing reaching a code block in a standalone .rst or .md file. The two met on a single construct, an rst code block inside a Python docstring, where the pinned black==24.* had drifted from the style ruff follows: "\N{snowman}" is normalised to "\N{SNOWMAN}" by one and left alone by the other. blacken-docs is now limited to markdown and rst, with its black pin on the current major.

    ruff also picks up docs/src/tutorials/region-manifold-extraction.ipynb, which a types override had been narrowing out of the types_or: [python, pyi, jupyter] default of the hook. That notebook was already clean.

    Six hooks join them: check-json, for the badge endpoint under docs/assets/badge/ that nothing validated; detect-private-key; check-executables-have-shebangs and check-shebang-scripts-are-executable, since ci-changelog.yml invokes .github/scripts/changelog.py by path and the execute bit carries it; and the check-hooks-apply and check-useless-excludes meta hooks, which hold every filter in this file to matching something. All six pass as they stand. (@claude)

  • #2558: Renamed tests/themes/test.py to tests/themes/test_themes.py. The old name matched neither half of the pytest python_files default, test_*.py and *_test.py, so none of its eleven tests had been collected since they landed.

    Collecting them exposed a real failure. test_default_theme asserts that the global pyvista theme is geovista, and importing tests/plotting loads the pyvista testing theme over it to prepare off-screen rendering. That happens at import, so collecting the image tests displaces the theme whether or not they are then selected to run. The test now reinstates the theme it is about through a fixture, and puts back whatever it found afterwards. (@claude)

  • #2559: Added tests/docs/test_assets.py, which asserts that no built page loads a script or a stylesheet from another host, and that every vendored icon is published beside its upstream licence notice. An extension that defaults its asset URLs to a content delivery network writes them into the page as it is read, so no build log carries them and no build fails when that host is unreachable. The three such dependencies closed so far were each found by hand and after the fact. (@claude)

  • #2561: Added a test that holds every declaration of python support to the trove classifiers in pyproject.toml: requires-python, the pyXYZ pixi features, the environments and solve-groups built upon them, the conda environment that ci-locks.yml exports, and the workflow matrices. Nine places declare which versions are supported and nothing else ties them together, so a bump reaching only some of them leaves geovista advertising one version of python and testing another. The test reads the source tree as text, so it requires neither a built artefact, a network, nor any particular interpreter.

    The same test now also records a convention that had only ever been followed by hand: the ci-docs.yml and ci-tests-docs.yml workflows build the documentation against the latest supported distribution of python alone, as does the codecov report raised by ci-tests.yml. Adopting python 3.14 had left all three on the oldest instead. (@claude)

  • #2563: mypy is now told which distribution of python to assume, and it is the oldest one supported. It reads no floor from requires-python, as ruff does; it assumes the interpreter running it, which for the pre-commit hook is whichever one that isolated environment happens to have been built with. A stdlib API added after the floor therefore type checked clean here, and would have raised for the oldest version the classifiers advertise. Pinning python_version makes a tenth place declaring python support, so tests/test_python_support.py now holds it to the classifiers alongside the other nine.

    The gallery carousel walks its images with Path.walk again. #870 swapped it for os.walk when the floor was still python 3.10, leaving a TODO waiting on 3.12; the floor has been past it since #2561. (@claude)

  • #2564: mypy has run in strict mode since the configuration was written, and has never seen a third-party type. The pre-commit hook is mirrors-mypy, whose isolated environment installs mypy and nothing else, so imports from pyvista, numpy, pyproj and the rest resolve to Any. Run instead against the locked geovista environment, the same configuration reports 642 errors across 22 of the 25 library modules.

    The design for closing that is the type coverage specification, the first specification this repository carries: a local hook invoking the locked environment’s own mypy, a companion CI job for pre-commit.ci which has no pixi, and a per-module ratchet so the check can go live before the 206 lines behind those errors are corrected.

    Specifications here are living documents, maintained alongside the code and cited by section from the files they govern, so where the two diverge it is the specification that gets corrected. This one is built but not yet listed in a toctree, and carries orphan: true until the specifications gain an index. Note that myst_parser registers .md whatever source_suffix names, so any markdown page added under docs/src is built from the day it lands, and warns unless a toctree holds it or another page includes it. (@claude)

  • #2565: The mypy pre-commit hook is now a local one invoking the locked geovista pixi environment, where the third-party types are real, rather than mirrors-mypy, whose isolated environment resolved every one of them to Any. Contributors therefore need pixi on PATH for the hook to run. pre-commit.ci has none, so the hook is listed under ci.skip and ci-typing.yml carries the identical command.

    Going live would otherwise have meant correcting 642 errors in a single change, so the 22 library modules carrying them are listed under ignore_errors alongside the gallery, and tests/test_typing_ratchet.py only ever lets that list shrink. The check is live today for everything outside it, and each module is retired on its own. sp-repo-review’s PC140 is ignored along the way: it matches a fixed set of repository URLs, so a local hook is invisible to it whatever it runs. (@claude)

  • #2567: Neither design specification in this repository can be reached from the documentation. Both are built, since myst_parser registers .md whatever source_suffix names, but nothing links to them, so each carries orphan: true to keep the --fail-on-warning build quiet. The seven citations of the form typing spec 3.1 in pyproject.toml, .pre-commit-config.yaml and ci-typing.yml therefore point a reader at a page they cannot find.

    A second specification, the publication specification, now records how that is closed and what keeps it closed. Specifications are published under an index, while plans are withheld by exclude_patterns, since a plan is frozen when it is written and a published plan is wrong by design. Sections are addressed by an anchor keyed to the section number: the slug docutils derives from heading text discards the number and resolves a collision to id1, which silently becomes id2 when anything is inserted above it. A citation carries its section sign, so a later transform can find it in source, tests, configuration and workflows without a pattern that also matches every decimal version number.

    The roadmap records the remaining citation machinery as candidates, each with the condition that would promote it, and a conventions test asserts every condition that can be observed. The one that cannot is marked as unwatched, because the failure is already on the page: the type coverage specification said its orphan: true marker would come off with the same change, nothing was watching, and nothing came off.

    Statuses now cite the work that produced them. A landed roadmap row carries its date and its pull request, and an open item carries Open, Resolved, Deferred or Abandoned with the issue or pull request that settled it, so a reader can see which decisions are still live and read what argued for the ones that are not.

    Raising issues for the type coverage specification’s two open items found both of them misstated, which is the argument for citing in miniature. One of the mypy codes disabled for the gallery is an upstream defect, reported in 2024 and fixed on pyvista main in September 2026. The other is not upstream at all. It comes from one annotation in geoplotter.py that names the camera property where a type belongs, which mypy cannot bind, and which shadows the attribute pyvista provides. The comment on the override says both are upstream. The first correction blamed lazy.load instead, and the same check, run before taking a related request upstream, disproved that too. The section now carries the measured cause and counts, and the two halves retire on separate triggers. (@claude)

  • #2573: The design specifications are now published. developer/specs/index.rst lists them under a new Specifications card in the developer guide, with a table of the citation prefix each one owns, and both lose the orphan: true they carried while nothing listed them. The implementation plans stay in the repository but out of the build, since a plan records what was intended and is not updated once the work lands.

    tests/test_spec_conventions.py holds them to their conventions: an anchor keyed to every numbered heading, a section sign in every citation, a {ref} role that displays the section it opens, and a status that names the work behind it. Its corpus is every text file in the repository, so a specification, or a citation of one, is governed from the day it lands. It also watches the roadmap. The citation machinery tephpy built becomes due once citations outside the collection pass thirty-five, and the test fails at that point instead of leaving the decision to somebody’s memory.

    The seven typing spec 3.1 citations gained their section sign. Two of the publication specification’s own rules changed on the way: the trigger of row 2 of docs spec §4 counted citation prefixes and fired on the very document that declared it, and the status rule could not cite an issue in another repository. (@claude)

  • #2577: Each end of a citation range is now a citation of its own (docs spec §3.2), as in typing spec §3.2–typing spec §3.4. A dash ends a run, so a bare far end cites the containing document, and a far end with no section sign cites nothing. tests/test_spec_conventions.py fails both, even in a range that a line break has split. (@claude)

  • #2578: The tests-unit pixi task and the test workflows now run the unit tests in parallel with pytest-xdist. To run the task serially, pass 0 as its second argument. (@claude)

  • #2585: The sphinx-gallery examples now run in parallel with up to four workers, which halves the CI docs build that feeds the browser tests. The GEOVISTA_SPHX_GLR_SERIAL environment variable still runs them one at a time. The test workflows now use pytest-xdist with one worker per logical CPU. (@claude)

  • #2589: In a planar CRS, transform_mesh has lifted a mesh by zlevel * zscale times a quarter of the mesh’s own extent, so equal levels sit at different heights in a projected scene. The design that fixes it, tying the length to the Earth’s radius in the CRS’s own units, is the planar z-levels specification, cited as zlevel spec §…. (@claude)

  • #2590: The sphinx-gallery examples with the largest interactive scenes now render a static image instead, which takes about 380 MB and 35 seconds of scene export off each documentation build. Add an example to STATIC in docs/src/_ext/gallery_scenes.py to do the same for it. (@claude)

  • #2600: Every unit test that renders is now marked image, so pytest -m "not image" runs safely on a machine without a display. A new unit test, tests/test_image_marker.py, fails any test that draws without the marker, and any test marked example but not image. Closes #2596. (@claude)

  • #2609: The work that each design specification status cites (docs spec §3.6) is now checked against GitHub. A landed roadmap row must cite merged pull requests on its date, a Resolved item must cite real work, and each night a status left behind by its work is reported. The ci-spec-status workflow runs the check, and pixi run -e test check-spec-status runs it locally. Closes #2575. (@claude)

  • #2659: Row 6 of the type coverage roadmap is now rows 6a to 6c of typing spec §4, one for each pull request it was split into. Once 6a landed, no single status was true of the row, and the nightly status check reported it as in progress against a merged pull request. Each row now has its own status and counts the lines its change clears, measured again: 11, 38 and 19. They total 68, one fewer than the specification said. Closes #2655. (@claude)

  • #2660: The tests-doc pixi task is now tests-docs, in line with tests-docs-browser and tests-docs-browser-install. Run the documentation image tests with pixi run tests-docs. (@claude)

  • #2672: Read the Docs now reads the documentation pages in parallel, with one process per CPU. The new parallel_build extension in docs/src/_ext keeps such a build within the memory Read the Docs allows, and its output the same as that of a serial build. It also mends the links from the geovista.themes module code back to the documentation of its objects. (@claude)

  • #2681: The example gallery is type checked in full, and the typing ratchet that held its errors is retired, so mypy now checks every module in src/geovista. tests/test_ci_typing.py keeps the ratchet’s guards on the CI workflow, and refuses any override that would ignore a module’s errors again. (@claude)

  • #2696: The browser tests of ci-tests-docs install chromium without --with-deps. On the runner image that flag only added fonts and a libnss3 upgrade, and fetching them from a slow apt mirror made the install step take anywhere from 17 seconds to 12 minutes. (@claude)

  • #2699: The test and docs workflows no longer run pyvista/setup-headless-display-action. Each job already wrapped its tests in xvfb-run, and the runner image ships xvfb and mesa, so the action only added an apt install and a second X server. The workflows set PYVISTA_OFF_SCREEN themselves, which saves about 15 seconds per job. (@claude)

  • #2700: The tests-docs-browser task runs tests/docs in parallel with pytest-xdist, taking the number of workers as its third argument, where 0 runs serially. Each worker launches its own browser over the one build. The browser tests took about 28 seconds on four workers, against 74 serially. (@claude)

  • #2703: tests/test_ci_typing.py now fails on a # mypy: comment anywhere in the package, and on any per-module relaxation of strict mode beyond the three that typing spec §3.1 and typing spec §3.4 give reasons for, so neither can silence mypy unseen. (@claude)

  • #2705: Published the repository checks specification, which gives every check geovista enforces on its own repository a section on one page of the developer guide, and has each failure link to its section. (@claude)

  • #2707: Added the repository checks page to the developer guide. It gives every check geovista enforces on its own repository a section saying why the rule exists and how to declare an exception, and a failure of any of them now ends with a link to its section. (@claude)

  • #2713: The repository checks gate of checks spec §3.4 now reports a section left with only one of its labels, and a CHECK it cannot read because it is unpacked, bound inside a block or imported. The nightly specification status issue links to its check’s section, and the status workflow runs when the breadcrumb changes. (@claude)

  • #2716: The changelog fragment check, .github/scripts/changelog.py, is now tested against towncrier itself, which joins the test dependencies, for each rule a fragment can break and each way the check can fail to run. (@claude)

  • #2717: Published the globe-first CRS foundation specification, which designs how geovista takes a mesh from the globe into any projection it supports, starting from a measurement of every PROJ projection method and cartopy class on a global mesh. (@claude)

  • #2732: Published the globe, heights and layers design specification, which designs how geovista gives data a height in metres and keeps one layer of a scene from hiding another, to replace zlevel and zscale, starting from a measurement of how layers render on the globe and on a map. (@claude)

  • #2752: Required that a mesh keeps its CRS, globe, heights and record through any file format VTK writes, as crs spec §2 decision 14, and recorded in crs spec §3.4 what each format keeps today. (@claude)

  • #2757: Published the globe on the scene design specification, which puts a Globe in place of every radius, records it on each mesh, reads meshes saved before it existed, and retires the fields geovista no longer writes, and extended the globe, heights and layers design with a vertical mapping split at sea level, a display lift for exploded views, and a record of the gallery’s use of the vertical. (@claude)

  • #2765: The browser tests of ci-tests-docs keep the output of Xvfb, which xvfb-run discards by default, and run it with -audit 2 so that it logs each client that connects or disconnects. When the job fails, it uploads the log as a workflow artifact, to show why VTK sometimes cannot open the display ( #2764). (@claude)

  • #2771: Published the heights as data design specification, which gives a mesh the height of each of its points as data, takes heights and depths where a mesh is built or added to a scene, draws a missing height at sea level, and refuses a height given with a z-level. (@claude)

  • #2785: Published the decision ledger design specification, which makes each decision of a specification citeable and permanent, records who agreed it and when, and gives every specification a Verification table showing the test or the reviewer’s sign-off behind each decision. (@claude)

  • #2786: Made each decision of a specification citeable, as heights spec §2 decision 3, with an anchor the build gives it (ledger spec §3.1). The docs, citation and decision ledger specifications now record each decision with who agreed it and when, and end with a Verification table naming the test or the sign-off behind each one. The conventions test holds every specification in the ledger to it, and the specifications index explains how to add, retire, verify and sign off a decision. (@claude)

  • #2792: Added pixi run -e test verification-report, the report of ledger spec §3.8, which lists each specification decision with its state and every place in the repository that cites it, and flags a verified decision whose test no longer cites it and a decision in force that nothing cites. (@claude)

  • #2796: Brought the type coverage, planar z-level and repository checks specifications into the decision ledger, landing row 4 of ledger spec §8. Each of their decisions is now numbered, with who agreed it and when, and each specification ends with a Verification table. Eleven of the sixteen decisions are verified by tests that now cite them, and the other five are signed off. A new check keeps # type: ignore comments out of the gallery scripts. (@claude)

  • #2799: Brought the globe-first CRS specification into the decision ledger, landing row 5 of ledger spec §8. Each of its 14 decisions now records who agreed it and when, and the specification ends with a Verification table. One decision is verified by tests that now cite it, and two are signed off. The other eleven wait on roadmap rows that haven’t started, and each pending row names the rows that will deliver it. (@claude)

  • #2803: Published the open items design specification. Each gap found while delivering a specification becomes one of its open items, with an issue of its own labelled spec: open item, and names the roadmap row that owes it. A row can’t land past an item it owes unless its pull request settles, moves or promotes the item, and an item that has moved twice graduates to a roadmap row of its own. (@claude)

  • #2807: Delivered the first row of the open items specification, row 1 of items spec §8. Each open item now starts with its state and records where it was found, the roadmap row that owes it and the issue that tracks it. The suite refuses an item owed by a row that has landed, or one that has moved more than twice. CI checks that each item’s issue carries the spec: open item label, and the nightly run reports a labelled issue that no item cites. Items and roadmap rows can be cited like sections and decisions, and the docs build links each citation to what it names. A citation in a near-miss form is refused, with the form to write instead, and the renumber guard now freezes an untitled item once something cites it. (@claude)

  • #2826: Brought the globe, scene and heights specifications into the decision ledger, which now covers every specification, and their open items, with the ledger specification’s own, into the open items grammar, landing row 6 of ledger spec §8 and row 5 of items spec §8. Each decision records who agreed it and when, and each specification ends with a Verification table: 16 decisions are verified by tests, 3 are signed off, 7 wait on later work, and one is retired, replaced by a decision that says what changed. Each of their open items has an issue of its own and the roadmap row that owes it. (@claude)

  • #2846: pixi run -e test verification-report now shows each specification’s roadmap before its decisions, and its open items after them (items spec §3.9), with each item’s state, the row that owes it and how often it has moved. It also flags an item on its last move, which must settle or graduate when its row lands, and an item owed by a row in progress, which that row’s pull request must answer. (@claude)

  • #2848: pixi run -e test verification-report, the report of ledger spec §3.8, now puts each specification’s ledger summary under a Decisions heading of its own, so it no longer reads as part of the roadmap. (@claude)

  • #2850: Brought the type coverage specification’s open items into the open items grammar, landing row 3 of items spec §8. Each item now records the pull request that first wrote it, and each of the seven still open has an issue of its own and the roadmap row that owes it. Those rows are four new candidates, rows 8 to 11 of typing spec §4, each waiting on an upstream release, and tests/test_ci_typing.py fails when one arrives, naming the row to land: a pyvista floor of 0.50.0, a typed lazy_loader or a Python floor of 3.15, rasterio stubs that take an array, or annotations for pooch, geopy, cartopy, cmocean or h3. (@claude)

  • #2859: The nightly check of the specification statuses no longer reports a labelled open item issue that an open pull request names, settling items spec §7 item 13. The pull request that raises an item raises its issue too, and the item reaches main only when it merges, so until then the issue is in flight rather than a gap. (@claude)

  • #2863: Recorded three gaps in the trigger tests of the type coverage specification’s candidate rows as open items, items 11 to 13 of typing spec §8, each with its own issue, owed by a new roadmap row that hardens those tests, row 12 of typing spec §4. (@claude)

  • #2873: Brought the globe-first CRS specification’s open items into the open items grammar, the last specification to come in, so every specification is now held to the same rules for its open items (row 4 of items spec §8). Each of the nine still open has an issue of its own and the roadmap row that owes it, two of them new: one deletes today’s seam and transform path once every class of projection has landed, and a candidate waits on the regions design. (@claude)

  • #2874: Every roadmap state in the design specifications now carries an emoji (ledger spec §2 decision 16), as a landed row always has: 🚧 in progress, 📋 not started and 💡 candidate. The conventions test refuses a bare state with the form it should take, and with it the decision ledger specification has nothing left to deliver. (@claude)

  • #2876: Settled items spec §7 item 11 and items spec §7 item 7, two of the open items specification’s own: an item owed by another specification’s row is always owed by a row of its own specification instead, landing with the other, and the check that two items never share an issue now reads every specification. A new open item, and a roadmap row to deliver it, cover checking that such a row really lands with the one it names. (@claude)

  • #2877: Hardened the renumber guard of citation spec §4.6, settling items spec §7 item 4, items spec §7 item 8 and items spec §7 item 12. Its advice now follows the direction of a move, so an item or row pulled up by a removal is told to keep the removed one in place. A cited item that only its number holds, while the list moved before it, is reported. A retired decision is paired at its own number first, so a decision agreed under its title is no longer read as it moving. (@claude)

  • #2878: Hardened the citation and item grammar the specifications are checked against, landing row 6 of items spec §8. A citation such as item 9a is read alike before or after its section, settling items spec §7 item 5. An item’s state is read whole, settling items spec §7 item 6. A middle dot inside a link or a code span no longer splits an item’s provenance, settling items spec §7 item 9. No list after an unnumbered subheading is anchored, agreeing with the build and settling items spec §7 item 10. A graduation to a row that had already landed is refused, settling items spec §7 item 3. (@claude)

  • #2879: The conventions test now checks that a roadmap row scoped as landing with another specification’s row lands with it, in the same pull request, settling items spec §7 item 14 and landing row 7 of items spec §8, the last row of the open items specification. (@claude)

  • #2882: Each row of a design specification’s Verification table now names its decision by its citation, §2 decision 3, in place of a bare number that could be read as a row, an item or another decision cited beside it (ledger spec §2 decision 17). The docs build links each row to its decision, and the conventions test refuses a bare number with the citation to write. (@claude)

  • #2888: Published the missing positions design specification. A position is missing when its longitude or latitude is masked or NaN, and the specification says what each public function that takes positions does with one: a query answers that it has no neighbour, a point cloud drops it with a warning, a geodesic shape refuses it, and a mesh drops the faces that use it. (@claude)

  • #2897: Added the How We Work page to the developer guide. It holds the rules geovista is developed by, for people and AI agents alike, numbered within their sections. Each rule says whether it applies to everyone, to people only or to agents only, and opens with its status: when, by whom and where it was agreed, or retired. The root AGENTS.md imports the page, so every agent session reads it. A new repository check keeps the page in form, and the renumber guard keeps each rule’s number for good. (@claude)

  • #2898: The How We Work page gains three rules that were written down elsewhere: confirm a roadmap row is bounded before planning it, record deferred work as an open item of its specification, and settle the specification items of an issue in the pull request that closes it. A fourth says when a change to a rule retires it. Under it, the rule on changelog fragments is retired for one that also lets a pull request which only records open work in a specification skip its fragment. (@claude)

  • #2919: The How We Work page gains a rule to check what pyvista has adopted before choosing a tool or a workflow, and to follow it unless there is a reason not to, written down where the choice is made. (@claude)

  • #2923: Added the benchmarks specification. It measures geovista’s speed with CodSpeed, as pyvista does, on every pull request and every push to main, and gives each benchmark a memory ceiling checked by pytest-memray. A regression is reported on the pull request that causes it, including one from a change of dependency, and never blocks a merge. Three roadmap rows build it. (@claude)

  • #2939: The How We Work page gains a rule that a reviewer reports each finding inline on the pull request, and leaves it to the pull request’s author to decide whether it becomes an issue. (@claude)

  • #2954: Published the layers of a scene specification, which orders a scene’s layers by named tiers, clips each layer at its horizon behind an opaque base layer, gives layers below sea level a tier beneath the base layer, and restores on the web the depth bias that an exported scene loses. It is the first specification to carry figures. (@claude)

  • #2957: The How We Work page gains two rules: a design choice that can be measured is measured with a throwaway spike before it is agreed, and a design or finding about what a scene looks like is shown with a render or a diagram beside the words. (@claude)

  • #2961: The documentation build now shrinks each interactive .vtksz scene it exports, resampling a texture wider than 2,048 pixels to that width and storing points as float32. A full build’s scenes drop from 1,693 MB to 535 MB, under the size limit Read the Docs sets for a build’s HTML. See #2960. (@claude)

  • #2969: Added the dependency floors specification. It tests geovista at the minimum versions it declares for its dependencies, using pixi’s lowest-direct solve strategy: on every pull request, in a weekly build of the documentation and install from PyPI, and on the built wheel before a release is published to PyPI. Every pip requirement gains a floor equal to pixi’s. Five roadmap rows build it. (@claude)

  • #2972: Seven tests built their inputs with np.empty, whose memory is uninitialised. In five of them it could hold an infinity, which transform_points refuses before the error each test expects, so they failed now and then. All seven now use np.zeros. (@claude)

  • #2973: The tests may no longer call numpy.empty or numpy.empty_like, whose memory is uninitialised, after five tests failed whenever it held an infinity. A new tests/ruff.toml adds both to ruff’s banned API for the tests alone, since the library writes every element of the arrays it builds this way before reading them. (@claude)

  • #2978: A failed weekly run of ci-locks.yml now opens an issue, or comments on the open one, quoting the pixi error that names the environment it could not solve. It used to fail silently, which in September left every environment’s lock four weeks behind. (@claude)

  • #2981: Working on geovista from source now needs pixi 0.84.2 or later, declared as requires-pixi. An older pixi misread the lock as stale and silently rewrote it; it now stops with an error naming the version it needs, which pixi self-update installs. (@claude)

  • #2984: The tests job of ci-tests.yml, ci-tests-pypi.yml and ci-tests-lock.yml no longer set ALLOW_PLOTTING. pyvista reads it only to decide whether to warn that no X server is running, and skips that warning when DISPLAY is set, which xvfb-run does in every one of these jobs. (@claude)

  • #2986: Posts and comments from AI agents on GitHub now name the agent that wrote them. This includes pull request descriptions, reviews and comments, issues and comments, and discussion posts and replies (workflow rule 3.6). (@claude)

  • #2987: Published the roadmap row order specification, which gives the order between roadmap rows one form: an ordering follows a row’s scope as · 🔗 after row 1, in one of four relations, so a check can read every roadmap’s order into one graph and refuse a cycle, a row that can never land, or a status the order contradicts. (@claude)

  • #2995: A changelog fragment written by an AI agent now credits that agent by its own GitHub account (workflow rule 4.9). Before, every agent’s fragment credited Claude. (@claude)

  • #2998: Published the roadmap row order diagrams specification, which draws the order between roadmap rows. The build will draw each specification’s rows and orderings under its roadmap table, and the open work of every roadmap on the specifications index, with the rows ready to start in green. Each pull request’s job summary will show the diagrams of the specifications it changes. (@claude)

  • #3003: Added workflow rule 2.10 to How We Work: a design decision that building and reviewing its work shows to be wrong is retired and replaced openly, its retirement saying what the review showed, and never edited in place or kept only because it was agreed. (@claude)

🧰 Miscellaneous#