Struct MeasuredTolerance 

pub struct MeasuredTolerance { /* private fields */ }

A deviation measured between a constructed entity and the geometry that entity was built to reproduce, recorded next to the derived band it was judged against.

Why this type exists

Everything above this line in this module is derived: a band predicted from a size (model_scale), a policy field, or a constant, computed before the geometry exists. Nothing above this line ever looks at what the construction actually produced. A collocation fit, a polyline interpolation or a marched section can miss its intended locus by far more (or, far more often, far less) than the derived band assumed, and today that number is either thrown away or never taken.

This is the other half, and it is the half OCCT has and we did not: BRepOffset_SimpleOffset::FillEdgeData sets an edge’s tolerance to the measured maximum distance between its new 3D curve and its pcurve on the offset surface, over every adjacent face (BRepOffset_SimpleOffset.cxx:296-310), and UpdateTolerance (BRepOffset_MakeOffset.cxx:4196-4302) re-measures and raises after the whole pipeline has run. Their model throughout is measure, do not assume: no tolerance in that pipeline is predicted from a size heuristic, each is measured off the geometry that was actually built and then propagated up the entity hierarchy.

The direction rule — this may never loosen anything

A MeasuredTolerance is a record of what happened, not a budget to spend. The measured number may flow into

• a diagnostic,
• a refusal message,
• a comparison against a derived band that tightens an existing gate,

and nowhere else. It must never be max-ed into an acceptance band, a weld radius or a search radius, because a construction would then certify its own error: the sloppier the fit, the wider the band that judges it. That is the self-certification failure per-entity-tolerances.md’s invariant I1 excludes structurally, and it is why this type deliberately exposes no f64-producing “band to use” accessor — only Self::deviation (what happened), Self::band (what was demanded), and the verdict between them.

Where it lives

Alongside the constructed entity, in the transient result struct the construction already returns — never as a field on crate::BrepSolid’s records. BrepSolid is serialized by io/snapshot.rs, which is a documented durable format; a record field would be a format change with a reader obligation. Persisting per-entity tolerances is a real and planned piece of work with its own design (docs/developer/kernel-plans/per-entity-tolerances.md, slice S3) — this type is the measurement layer beneath it, and lands with no format churn at all.

Implementations

impl MeasuredTolerance

pub fn new(deviation: f64, band: f64) -> Self

Record deviation against the derived band that was in force.

Both arguments are normalised in the FAIL-SAFE direction, so a measurement that went wrong reads as “worse than the band”, never as “fine”: a non-finite or negative deviation becomes +inf, and a non-finite or negative band becomes 0.0. Either way Self::exceeds_band answers true.

pub fn exact(band: f64) -> Self

A construction with no fit to measure — a closed-form analytic result, or an affine offset that shifts a control net rigidly. Records 0.0 deviation as a claim, which is why it is a named constructor: reading MeasuredTolerance::exact(band) at a call site says “this lane has no approximation error by construction”, where new(0.0, band) would read as an unmeasured default.

pub fn deviation(&self) -> f64

What the construction actually did.

pub fn band(&self) -> f64

What the size-derived policy demanded of it.

pub fn within_band(&self) -> bool

The construction met the band it was judged against.

pub fn exceeds_band(&self) -> bool

The construction was WORSE than the derived band assumed — the case worth acting on. A site that gates on this refuses; a site that only reports names both numbers.

pub fn utilisation(&self) -> f64

deviation / band — how much of the derived band the construction spent. > 1 is Self::exceeds_band; a value orders of magnitude below 1 says the band is loose here, which is worth knowing and is exactly what the corpus distribution reports. A zero band answers 0.0 for a zero deviation and +inf otherwise.

pub fn worse_of(self, other: Self) -> Self

The worse of two records: the larger deviation against the tighter band.

Both halves move in the fail-safe direction, so folding a set of per-coedge or per-edge measurements can only ever make the summary harder to pass, never easier.

pub fn worst(records: impl IntoIterator<Item = Self>) -> Option<Self>

Fold a set of measurements into their worst, or None when empty.

pub fn describe(&self) -> String

measured 1.234e-7 against band 4.000e-3 — the shared wording for refusal messages and diagnostics, so every site that reports a measured deviation reports both numbers in the same form.

Trait Implementations

impl Clone for MeasuredTolerance

fn clone(&self) -> MeasuredTolerance

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for MeasuredTolerance

impl Debug for MeasuredTolerance

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl PartialEq for MeasuredTolerance

fn eq(&self, other: &MeasuredTolerance) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for MeasuredTolerance