ConstHold the limit perimeter of a surface network's girth rings, the rings
that declare perimeter, by solving a radial factor along each sweep.
A girth ring runs once round one sweep at one u. Along a row of
valence-four vertices whose loop neighbours lie opposite, beside rows of
such vertices, the Catmull-Clark limit is the tensor-product cubic
B-spline, so the limit along the row is the periodic uniform cubic B-spline
whose control points are (A + 4 P + B) / 6 for each vertex P and its
two neighbours A and B across the row; its perimeter is integrated by
five-point Gauss-Legendre quadrature on every span. Each sweep's factor is
the sampled-overshoot-policy spline of interpolateAutoMovieSpline through 1 at the sweep's
other rings at one u, which meet neighbouring parts and stay, and through
a value at each girth ring, holding its end values beyond the outermost
knots; IAutoMovieSweepRig.girth folds it into the scale. The girth
values solve the system "every measured perimeter equals its declaration"
jointly by Broyden's method (Broyden 1965, Mathematics of Computation
19(92):577-593), one rest fit per step: the Jacobian starts diagonal, each
perimeter in proportion to its own factor, which is exact where the fit
interpolates the row and no neighbouring sample reads another factor, and
each step's rank-one update learns how neighbouring rings couple through
the spline, cell averages or a least-squares fit, so rings close together
converge together where updating each alone would oscillate or crawl; a
step that would make a factor non-positive is halved. It stops when every
perimeter has relative error at most 1e-9 and refuses after 24 fits.
Non-finite measurements/steps and zero representable progress are refused
before a factor update can enter nonterminating positivity backtracking.
The initial diagonal uses each measured perimeter directly, without
reconstructing it from a potentially cancellation-dominated residual.
Cage quad vertex-index lists, borrowed to establish regular row neighbours for limit-perimeter integration.
Read-only rest declaration selecting rings with explicit perimeter targets. Each girth ring must run once around one sweep station; the solver rejects unsupported cage/ring incidence.
Writable packed rest control XYZ buffer in metres, with 3*vertices entries. Overwritten by rest evaluations and read by the limit-perimeter measurement; caller retains ownership. A failed solve can leave the rigs and this buffer at its last evaluated trial; there is no whole-solve rollback.
Re-evaluate and fit the complete network at rest using current girth factors, overwriting the supplied packed metre control positions. Called once per solve step; exceptions propagate.
Ring names to cage vertex indices in closed-loop order, produced by the sampler. Each index addresses positions at three times its value.
Return the sealed compiled rig for an existing network sweep name. The solve installs radial girth factors on these same rigs; do not return unrelated new rigs on successive calls.
Number of cage vertices. The caller supplies indices within this population and position buffers containing three scalars per vertex.
Error when a girth ring's perimeter is not positive and finite, it also has positions, it does not run once round one sweep at one u, a vertex on or beside it is not a regular interior vertex, two rings of one sweep share a u, a measured perimeter is non-positive or non-finite, a computed step is non-finite, no representable progress is possible, or the solve does not converge.
Hold the limit perimeter of a surface network's girth rings, the rings that declare
perimeter, by solving a radial factor along each sweep.A girth ring runs once round one sweep at one
u. Along a row of valence-four vertices whose loop neighbours lie opposite, beside rows of such vertices, the Catmull-Clark limit is the tensor-product cubic B-spline, so the limit along the row is the periodic uniform cubic B-spline whose control points are(A + 4 P + B) / 6for each vertexPand its two neighboursAandBacross the row; its perimeter is integrated by five-point Gauss-Legendre quadrature on every span. Each sweep's factor is the sampled-overshoot-policy spline ofinterpolateAutoMovieSplinethrough 1 at the sweep's other rings at oneu, which meet neighbouring parts and stay, and through a value at each girth ring, holding its end values beyond the outermost knots;IAutoMovieSweepRig.girthfolds it into the scale. The girth values solve the system "every measured perimeter equals its declaration" jointly by Broyden's method (Broyden 1965, Mathematics of Computation 19(92):577-593), one rest fit per step: the Jacobian starts diagonal, each perimeter in proportion to its own factor, which is exact where the fit interpolates the row and no neighbouring sample reads another factor, and each step's rank-one update learns how neighbouring rings couple through the spline, cell averages or a least-squares fit, so rings close together converge together where updating each alone would oscillate or crawl; a step that would make a factor non-positive is halved. It stops when every perimeter is within 1e-9 of its declaration and refuses after 24 fits. Non-finite measurements/steps and zero representable progress are refused before a factor update can enter nonterminating positivity backtracking. The initial diagonal uses each measured perimeter directly, without reconstructing it from a potentially cancellation-dominated residual.