tolerances.hpp File Reference

Unified geometric epsilon / tolerance policy (see docs/STANDARDS.md §4). More...

Go to the source code of this file.

Namespaces
namespace	reusex

Variables
constexpr double	reusex::kEpsilonLengthM = 1e-9
	Absolute length zero-check, in meters.
constexpr double	reusex::kEpsilonAngleRad = 1e-6
	Absolute angular tolerance, in radians.
constexpr double	reusex::kEpsilonRelative = 1e-12
	Dimensionless / normalized tolerance.
constexpr double	reusex::kToleranceCoplanarM = 0.01
	Coplanarity / spatial-adjacency tolerance, in meters.

Detailed Description

Unified geometric epsilon / tolerance policy (see docs/STANDARDS.md §4).

This header is the single source of truth for degenerate-geometry epsilons. It replaces the scattered ad-hoc 1e-6/1e-9/1e-15 literals that used to live in geometry/utils.cpp, CoplanarPolygon.cpp, Solidifier_to_mesh.cpp and friends.

Which constant to use when

• kEpsilonLengthM — absolute length comparisons expressed in meters. Use it to decide whether a length/distance/norm is effectively zero (degenerate edge, zero-area face, near-zero direction vector). Because it is absolute and metric, it only makes sense on quantities that live in the scene's metric coordinate frame. Prefer recentering coordinates near the origin (see #216) before applying it so that floating-point resolution is uniform across the scene.
• kEpsilonAngleRad — absolute angular comparisons in radians. Use it for "is this angle effectively zero / are these two directions parallel" tests when the quantity you have is an angle (e.g. std::acos(dot)), not a raw dot product.
• kEpsilonRelative — dimensionless / already-normalized comparisons. Use it for dot products of unit vectors, squared norms of normals that are supposed to be unit length, and any ratio/relative test where the operands carry no physical unit. This is the correct guard for divisions by a squared norm (a normalized normal has squared norm ~1, so a value below kEpsilonRelative means the "normal" is degenerate).
• kToleranceCoplanarM — coplanarity / adjacency tolerance in meters. Two planes whose signed offsets differ by less than this (and whose normals are parallel) are considered the same physical surface. Also the default gap under which two disjoint inlier patches count as spatially adjacent.

Rules (STANDARDS §4)

• No new magic numbers in algorithm code — reach for one of these constants, or add a tunable to the relevant options struct.
• Absolute tolerances state their unit (meters) in the name (...M) or a comment. Angular tolerances state radians.
• When comparing two vectors for parallelism, prefer converting to an angle and using kEpsilonAngleRad, or compare 1 - |dot| against kEpsilonRelative; do not invent a per-call-site literal.

Definition in file tolerances.hpp.