DeloneCommons · IvanChernyshov · Mar 4, 2026 · Feb 16, 2026 · Mar 4, 2026 · Mar 4, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -4,6 +4,25 @@ All notable changes to this project are documented in this file.
 
 The format is based on *Keep a Changelog*, and this project follows *Semantic Versioning*.
 
+## [0.4.2] - 2026-03-04
+
+### Changed
+
+- Vendored Voro++: updated the vendored snapshot to include the upstream numeric robustness fix for
+  power/Laguerre (radical) pruning (fixes rare cross-platform edge cases in fully periodic power
+  tessellations).
+- Removed the previously vendored local `nextafter`-based `max_radius` inflation patch (no longer needed).
+
+## [0.4.1] - 2026-02-16
+
+### Fixed
+
+- Vendored Voro++: inflate the stored global `max_radius` by 1 ULP (via `nextafter`) in
+  power/Laguerre mode to make radical pruning robust across platforms.
+- Removed a Python-side workaround that recomputed fully periodic orthorhombic power tessellations
+  via the periodic (triclinic) backend when periodic face-shift assignment failed.
+- Updated documentation to reflect the patched vendored Voro++ snapshot.
+
 ## [0.4.0] - 2026-02-15
 
 Initial public release.

diff --git a/NOTICE.md b/NOTICE.md
@@ -9,3 +9,7 @@ This project vendors and links against the following third-party software:
 - License: See `vendor/voro++/LICENSE`
 
 pyvoro2 is licensed under the MIT License (see `LICENSE`). The included Voro++ code remains under its original license.
+
+Local modifications:
+
+- None. The vendored Voro++ snapshot is kept unmodified (aside from being vendored into this repository).
diff --git a/README.md b/README.md
@@ -1,7 +1,6 @@
 # pyvoro2
 
 [![CI](https://github.com/IvanChernyshov/pyvoro2/actions/workflows/ci.yml/badge.svg)](https://github.com/IvanChernyshov/pyvoro2/actions/workflows/ci.yml) [![Docs](https://github.com/IvanChernyshov/pyvoro2/actions/workflows/docs.yml/badge.svg)](https://github.com/IvanChernyshov/pyvoro2/actions/workflows/docs.yml) [![PyPI](https://img.shields.io/pypi/v/pyvoro2.svg)](https://pypi.org/project/pyvoro2/) [![Python Versions](https://img.shields.io/pypi/pyversions/pyvoro2.svg)](https://pypi.org/project/pyvoro2/) [![License](https://img.shields.io/pypi/l/pyvoro2.svg)](https://github.com/IvanChernyshov/pyvoro2/blob/main/LICENSE)
-
 **Documentation:** https://IvanChernyshov.github.io/pyvoro2/
 
 
@@ -19,7 +18,7 @@ matter physics — especially **periodic boundary conditions** and **neighbor gr
 
 pyvoro2 is designed to be **honest and predictable**:
 
-- it vendors and wraps **unmodified** upstream Voro++;
+- it vendors and wraps an upstream Voro++ snapshot (with a small numeric robustness patch for power/Laguerre diagrams);
 - the core tessellation modes are **standard Voronoi** and **power/Laguerre**.
 
 ## Quickstart
@@ -100,19 +99,10 @@ For stricter post-hoc checks, see:
 - `pyvoro2.validate_tessellation(..., level='strict')`
 - `pyvoro2.validate_normalized_topology(..., level='strict')`
 
-## Platform note (macOS)
-
-On some macOS builds, fully periodic **power/Laguerre** tessellations can
-occasionally produce a non-reciprocal face/neighbor graph. In that case,
-requesting `return_face_shifts=True` may raise a `ValueError` because pyvoro2
-cannot assign consistent periodic shifts.
-
-Workarounds:
-
-- Avoid requesting shifts (`return_face_shifts=False`), or
-- disable shift validation (`validate_face_shifts=False`, shifts may be
-  unreliable), or
-- run strict periodic power workflows on Linux/Windows.
+Note: pyvoro2 vendors a Voro++ snapshot that includes the upstream numeric robustness fix for
+*power/Laguerre* mode (radical pruning). This avoids rare cross-platform edge cases where fully
+periodic power tessellations could yield a non-reciprocal face/neighbor graph under aggressive
+floating-point codegen.
 
 ## Why use pyvoro2
 

diff --git a/docs/guide/domains.md b/docs/guide/domains.md
@@ -60,18 +60,13 @@ Notes:
 - `return_face_shifts=True` is supported whenever at least one axis is periodic.
 - For non-periodic axes, query points outside the bounds are treated as out-of-domain.
 
-Platform note (macOS):
+Robustness note:
 
-- On some macOS builds (compiler/CPU-dependent), fully periodic **power/Laguerre**
-  tessellations can occasionally produce a non-reciprocal face/neighbor graph.
-  In that case, requesting `return_face_shifts=True` may raise a `ValueError`
-  because pyvoro2 cannot assign consistent periodic shifts.
-
-  Workarounds (choose what matches your workflow):
-
-  - Disable shift validation: `validate_face_shifts=False` (shifts may be unreliable).
-  - Avoid requesting shifts: `return_face_shifts=False`.
-  - Use a non-macOS build for strict periodic power workflows.
+- pyvoro2 vendors a Voro++ snapshot that includes the upstream numeric robustness fix for fully
+  periodic **power/Laguerre** tessellations (radical pruning).
+- If `return_face_shifts=True` fails with a `ValueError`, it is typically due to a
+  genuine geometric degeneracy (for example, nearly co-spherical sites) rather than
+  a platform-specific issue.
 
 ## `PeriodicCell` (triclinic, fully periodic)
 

diff --git a/docs/index.md b/docs/index.md
@@ -12,7 +12,7 @@ matter physics — especially **periodic boundary conditions** and **neighbor gr
 
 pyvoro2 is designed to be **honest and predictable**:
 
-- it vendors and wraps **unmodified** upstream Voro++;
+- it vendors and wraps an upstream Voro++ snapshot (with a small numeric robustness patch for power/Laguerre diagrams);
 - the core tessellation modes are **standard Voronoi** and **power/Laguerre**.
 
 ## Quickstart
@@ -93,6 +93,11 @@ For stricter post-hoc checks, see:
 - `pyvoro2.validate_tessellation(..., level='strict')`
 - `pyvoro2.validate_normalized_topology(..., level='strict')`
 
+Note: pyvoro2 vendors a Voro++ snapshot that includes the upstream numeric robustness fix for
+*power/Laguerre* mode (radical pruning). This avoids rare cross-platform edge cases where fully
+periodic power tessellations could yield a non-reciprocal face/neighbor graph under aggressive
+floating-point codegen.
+
 ## Why use pyvoro2
 
 Voro++ is fast and feature-rich, but it is a C++ library with a low-level API.

diff --git a/docs/project/about.md b/docs/project/about.md
@@ -20,8 +20,9 @@ At the core, pyvoro2 exposes only two mathematically standard tessellations:
 Voro++ is a widely used C++ library for computing Voronoi cells efficiently in 3D.
 It implements robust algorithms and is commonly used in computational physics and materials science.
 
-pyvoro2 vendors an **unmodified** snapshot of upstream Voro++ and builds its Python extension against it.
-This keeps the core geometry trustworthy and makes it easy to track upstream behavior.
+pyvoro2 vendors a snapshot of upstream Voro++ and builds its Python extension against it.
+The vendored snapshot includes the upstream numeric robustness fix for *power/Laguerre* (radical) pruning,
+which avoids rare cross-platform edge cases in fully periodic power tessellations.
 
 ## When should you use pyvoro2?
 

diff --git a/src/pyvoro2/__about__.py b/src/pyvoro2/__about__.py
@@ -5,4 +5,4 @@
 """
 
 # Keep this as a simple assignment so scikit-build-core can extract it via regex.
-__version__ = '0.4.0'
+__version__ = '0.4.2'
diff --git a/src/pyvoro2/api.py b/src/pyvoro2/api.py
@@ -342,85 +342,6 @@ def compute(
             else tuple(bool(x) for x in domain.periodic)
         )
         is_periodic = isinstance(domain, OrthorhombicCell) and any(periodic_flags)
-        full_periodic_ortho = (
-            isinstance(domain, OrthorhombicCell)
-            and periodic_flags == (True, True, True)
-        )
-
-        def _compute_full_periodic_ortho_via_periodic_backend(
-            *,
-            mode: Literal['standard', 'power'],
-            rr: np.ndarray | None,
-        ) -> list[dict[str, Any]]:
-            """Compute fully periodic OrthorhombicCell via Voro++ periodic containers.
-
-            This is a fallback path for rare platform-dependent issues in the
-            rectangular container backend (or its face bookkeeping).
-
-            It remaps points into the primary cell, shifts the origin to (0,0,0),
-            calls the periodic container backend, then shifts outputs back.
-            """
-            assert isinstance(domain, OrthorhombicCell)
-            (xmin, xmax), (ymin, ymax), (zmin, zmax) = bounds
-            origin = np.asarray((xmin, ymin, zmin), dtype=np.float64)
-            bx = float(xmax - xmin)
-            by = float(ymax - ymin)
-            bz = float(zmax - zmin)
-
-            # Wrap into the primary domain (half-open in periodic axes)
-            pts_w = domain.remap_cart(pts, return_shifts=False)
-            pts_i = pts_w - origin
-            cell_params = (bx, 0.0, by, 0.0, 0.0, bz)
-
-            if mode == 'standard':
-                cells_i = core.compute_periodic_standard(
-                    pts_i,
-                    ids_internal,
-                    cell_params,
-                    (nx, ny, nz),
-                    init_mem,
-                    opts,
-                )
-            elif mode == 'power':
-                if rr is None:
-                    raise ValueError('internal error: rr is required for power mode')
-                cells_i = core.compute_periodic_power(
-                    pts_i,
-                    ids_internal,
-                    rr,
-                    cell_params,
-                    (nx, ny, nz),
-                    init_mem,
-                    opts,
-                )
-            else:
-                raise ValueError(f'unknown mode: {mode}')
-
-            if include_empty:
-                # Voro++ remaps inserted points into the primary cell; mirror that
-                # for any empty-cell records we inject.
-                _add_empty_cells_inplace(cells_i, n=n, sites=pts_i, opts=opts)
-
-            # Shift outputs back to the original bounds origin.
-            if float(xmin) != 0.0 or float(ymin) != 0.0 or float(zmin) != 0.0:
-                ox, oy, oz = float(origin[0]), float(origin[1]), float(origin[2])
-                for cc in cells_i:
-                    s = cc.get('site')
-                    if s is not None and len(s) == 3:
-                        cc['site'] = [
-                            float(s[0]) + ox,
-                            float(s[1]) + oy,
-                            float(s[2]) + oz,
-                        ]
-                    if return_vertices:
-                        vv = cc.get('vertices')
-                        if vv:
-                            cc['vertices'] = [
-                                [float(x) + ox, float(y) + oy, float(z) + oz]
-                                for (x, y, z) in vv
-                            ]
-            return cells_i
-
         if return_face_shifts:
             if not is_periodic:
                 raise ValueError(
@@ -479,50 +400,21 @@ def _compute_full_periodic_ortho_via_periodic_backend(
         if return_face_shifts:
             assert isinstance(domain, OrthorhombicCell)
             a, b, cvec = domain.lattice_vectors
-            try:
-                _add_periodic_face_shifts_inplace(
-                    cells,
-                    lattice_vectors=(a, b, cvec),
-                    periodic_mask=periodic_flags,
-                    mode=mode,
-                    radii=(
-                        np.asarray(radii, dtype=np.float64)
-                        if radii is not None
-                        else None
-                    ),
-                    search=int(face_shift_search),
-                    tol=face_shift_tol,
-                    validate=bool(validate_face_shifts),
-                    repair=bool(repair_face_shifts),
-                )
-            except ValueError:
-                # Fallback for rare platform-dependent issues (observed on macOS)
-                # where the rectangular periodic backend can produce a face/neighbor
-                # graph that prevents consistent shift assignment.
-                if full_periodic_ortho and mode == 'power':
-                    cells = _compute_full_periodic_ortho_via_periodic_backend(
-                        mode=mode,
-                        rr=rr,
-                    )
-                    # Re-run shift assignment on fallback output.
-                    _add_periodic_face_shifts_inplace(
-                        cells,
-                        lattice_vectors=(a, b, cvec),
-                        periodic_mask=(True, True, True),
-                        mode=mode,
-                        radii=(
-                            np.asarray(radii, dtype=np.float64)
-                            if radii is not None
-                            else None
-                        ),
-                        search=int(face_shift_search),
-                        tol=face_shift_tol,
-                        validate=bool(validate_face_shifts),
-                        repair=bool(repair_face_shifts),
-                    )
-                else:
-                    raise
-
+            _add_periodic_face_shifts_inplace(
+                cells,
+                lattice_vectors=(a, b, cvec),
+                periodic_mask=periodic_flags,
+                mode=mode,
+                radii=(
+                    np.asarray(radii, dtype=np.float64)
+                    if radii is not None
+                    else None
+                ),
+                search=int(face_shift_search),
+                tol=face_shift_tol,
+                validate=bool(validate_face_shifts),
+                repair=bool(repair_face_shifts),
+            )
         if ids_user is not None:
             _remap_ids_inplace(cells, ids_user)