Skip to contents

Changelog

Source: NEWS.md

bigcurve (development version)

  • The C++ engine src/bigcurve.h is frozen: bug fixes only. New engine features land first in the Rust crate bigcurve (hypertidy/bigcurve-rs), which becomes the canonical engine once it covers ellipsoid = TRUE (#14).

  • The C++ engine turns floating-point contraction (fused multiply-add) off with a pragma, so it gives the same vertices as the Rust engine on every platform. With contraction (clang on arm64, GCC with FMA targets) some midpoint tests went the other way and vertex counts changed. canon/check.cpp checks the header against the Rust crate’s canon tables (#15).

bigcurve 0.6.0

First CRAN release.

  • New densify(ellipsoid = TRUE) places new vertices on the WGS84 geodesic instead of the great circle, using the ‘geographiclib’ package (in Suggests). The sphere still chooses how many vertices each edge gets and at what fraction along it; each is then moved to the same fraction of the geodesic. Shared edges still get identical vertices.

  • Edges shared by neighbouring polygons now refine to identical vertices. The refinement arithmetic always runs in one canonical direction and the chain is reversed afterwards, so a -> b and b -> a give bitwise identical new vertices (previously some differed in the last bits, so wkpool::merge_coincident() could not fuse them). An edge that occurs more than once with the same vertices (a merged pool, a segment mesh) is refined once and its new vertices are reused.

  • densify() now really defaults to source = NULL, so a wkpool input’s own crs is used. Previously the documented default was not the one in the code, and a pool in projected units was read as degrees.

  • The default tolerance is now derived from the larger side of the projected extent, so a single parallel or meridian (zero width in one direction) no longer errors.

  • densify() now errors clearly when given projected input (coordinates outside lon [-360, 360] / lat [-90, 90]) instead of silently producing garbage or densifying nothing. The source must be geographic; treating a projected source’s edges as straight lines is a separate model that is not yet implemented.

  • Requires wkpool >= 0.4.0 (CRAN) for pool input; the fallbacks for older wkpool versions, and the vctrs and utils dependencies they needed, are gone. glue is no longer used.

  • New vignette, “bigcurve and s2”, contrasting bigcurve with s2.

  • Credit and licence notice for the d3-geo resampling method in inst/COPYRIGHTS; Mike Bostock added as contributor and copyright holder.

bigcurve 0.5.0.9001

  • densify() carries wkpool path provenance (.path and the paths table, wkpool >= 0.3.0.9003) through refinement: every refined segment inherits the path of the input segment it descends from, and added vertices are interior to their path, so cycles, ring roles and exact feature reconstruction (wkpool::cycles_to_wkb()) survive densification.

  • The default input coordinate system is now ‘EPSG:4326’ rather than ‘OGC:CRS84’ throughout (densify(), rproj_xy()). Authority codes are axis-normalized on the way into PROJ, so input remains lon,lat either way; ‘EPSG:4326’ is simply the more familiar name.

  • densify() gains crs awareness for wkpool input: source now defaults to NULL, and a pool built with wkpool >= 0.3.0.9000 supplies its own crs (as captured from the original geometry by wkpool::establish_topology()). An explicit source still wins, and inputs with no crs fall back to ‘EPSG:4326’.

  • The refined pool returned by densify() now carries the input pool’s crs and geodesic attributes, built through the exported wkpool::new_wkpool() constructor when available.

bigcurve 0.5.0

  • densify() now accepts wkpool objects (hypertidy/wkpool >= 0.3.0): any wk-handleable geometry decomposed with wkpool::establish_topology() can be densified directly, returning a wkpool with vertices appended (freshly minted .vx ids), directed segments refined in place, and .feature provenance carried onto every refined segment.

  • Added vertices are degree-2 by construction, so densification does not alter the arc-node structure of a pool: find_nodes() before and after refinement is identical.

  • The mesh engine now reports parent, the input segment each refined segment descends from. This is the hook that carries per-segment attributes (feature, ring, arc membership) through refinement.

  • Pools carrying z or m vertex values are refused with an error: densification is strictly 2D and will not invent values at inserted vertices. Rebuild the pool from xy geometry to proceed.

  • Added wkpool, wk, and vctrs to Suggests.

bigcurve 0.1.0

Complete rework of the package around a new adaptive resampling engine, following d3-geo (Bostock). The prototype bisect() interface is gone.

New user surface

  • New exported function densify(), the single verb of the package. Accepts a two-column lon,lat matrix (a path), a list of such matrices, or a segment mesh (rgl mesh3d style vb/is), and returns the same kind of object with vertices inserted only where the target projection demands them.

  • rproj_xy() is now exported: forward coordinate transformation with axis normalization (authority codes such as ‘EPSG:4326’ behave as lon,lat input) and NA rather than error for unprojectable points.

Breaking changes

  • Error is now controlled in the plane of the target projection: the tolerance argument is in projected units (usually metres), or derived from the projected extent as one part in pixels (default

    1. when unset. The old dist metres-on-the-sphere argument (which was accepted but not honoured) is gone.

  • Output coordinates are in the source coordinate system (geographic by default); project the result for plotting.

  • Points outside the domain of the target projection never error: segments touching them are left un-densified and refinement stops at unprojectable midpoints. Clipping is out of scope.

  • bisect(), curv_len(), project_segment() and reproj_mesh3d() are removed.

Internals and dependencies

  • New header-only C++ core with no R or PROJ dependency, driving refinement with a target-plane metric: perpendicular distance from the projected great circle midpoint to the planar chord, a parametric shear check, a forced split for arcs wider than 30 degrees, and a maximum depth. Each point is transformed exactly once and no inverse transforms are used, so refinement is dramatically faster than the 0.0.x loop.

  • Links directly to the system PROJ library (>= 8.0) via configure and pkg-config; the libproj package (removed from CRAN) is no longer used. Windows builds use the PROJ shipped with Rtools.

  • Dropped terra, geodist, geosphere, glue, reproj, and all undeclared dependencies; hard dependencies are now cpp11 (LinkingTo) and system PROJ only.

  • Spherical great circle midpoints are used (consistent with d3-geo); this is a display densification tool, not an ellipsoidal geodesy one.

  • README rewritten around densify(), with a graticule example using textures::segs() (textures and maps added to Suggests).

bigcurve 0.0.1

  • First version with workable C++ components. Logic needs work but the pieces are in place.

bigcurve 0.0.0.9000

  • Added a NEWS.md file to track changes to the package.