ConstA radius/height meridian for an authored lateral surface around local Y.
Each profile point is x = radius and y = height, in metres. Unlike
extrusion and sweep, the meridian is taken as authored rather than hulled, so
an arbitrary silhouette polyline is allowed here; only its radii must be
non-negative. Profile order owns the surface orientation, and every output
buffer is new.
Closure follows the meridian and is the caller's to declare: a meridian that starts and ends on the axis closes into a solid whose pole rings collapse to zero-area triangles, and one that does not is an open tube with a rim at each end. Neither is repaired.
The surface carries a metric atlas measured the way its siblings measure
theirs: distance travelled around the section against distance travelled
along the path. Around is the arc the vertex actually sits on, theta times
its own radius, so a baluster turned to a 0.3 m radius shows 1.88 m of grain
around it rather than one turn of an image stretched to fit it. Along is the
meridian's own polyline length, which is slant distance on a cone and
developed height on a cylinder rather than the shortcut through the axis.
Around runs against the direction the lattice is built in, and that is the
whole reason it is written 2 * pi - theta rather than theta. The outward
normal this winding produces, the meridian running up, and theta running
round form a left-handed triple, so an atlas keyed on theta directly comes
out mirrored against every other builder here: a letter reads backwards, and
a normal map's tangent basis is handed the wrong way, which tilts the
lighting rather than merely the image. Reversing the axis that closes on
itself costs nothing an author can name, because which way round a surface of
revolution is traversed carries no authored meaning, while reversing the
meridian instead would put v = 0 at whichever end of the profile was typed
last.
The lattice carries a duplicate ring at theta = 2 * pi, which is what makes
the seam an honest cut: one edge of the seam reads zero and the other reads
the full circumference at that radius, so the closing quad never
interpolates backwards through the atlas. Those two phases coincide only
when the declared texture repeat divides that circumference; the kernel does
not stretch a finish to hide a mismatched seam. A changing radius also makes
the surface non-developable in general. Scaling by radius preserves local
circumference scale on every parallel and therefore shears the flat atlas
between unequal parallels instead of pretending one global isometry exists.
A meridian point on the axis is a pole, and its whole ring reads zero around,
because a circle of no radius has no arc to travel. Its collapsed triangles
remain the topology this operator declares; their UVs stay finite, and they
are the one place the atlas has no handedness to report. The duplicate seam
vertices share their accumulated face normals before normalization, and every
duplicate of an axis pole shares that pole's normal as well, so the UV cut
does not introduce a lighting discontinuity.
How far that shear goes decides what finish the surface can carry. The map
is exactly equiareal, its Jacobian determinant one everywhere, so the whole
distortion is pure shear of 2 * pi - theta times dr / ds. Worst
anisotropy over a full turn is therefore
(sqrt(k * k + 4) + k) / (sqrt(k * k + 4) - k) at k = 2 * pi * |dr / ds|:
1.0 on a cylinder, 2.0 at 6.5 degrees off the axis, 4.0 at 13.8 degrees, 21.5
on a 45-degree cone, and 41 where the meridian runs level, which is what a
dome's pole and a cone's tip approach. A tile that reads square beside the
seam is drawn into a ribbon a quarter turn away, and the curve of constant
u spirals instead of following a meridian.
That is the price of the metric statement rather than a defect awaiting
repair. Measuring true arc around forces u to be theta * r plus some
c(s), whose derivative would have to be -theta * dr / ds to cancel the
shear, and no function of s alone does that at every theta. Choosing
c(s) only moves where the shear vanishes, which is why it sits at the seam
here; centring it would halve the worst case to 11.8 at a pole, still far
past readable. The pole is not a second fault: cut the tip off a cone and the
frustum that remains shears exactly as hard, so the collapsed pole triangles
above are a topology note rather than the cause.
A directional finish therefore wants a meridian within a few degrees of the
axis, which is a drum, a shaft, a turret barrel. A sloped or domed form that
must carry one is authored as flat faces through
buildAutoMoviePolyhedron, whose per-face frame is isometric, or as a
flat cap through extrudeAutoMovieRegion, and a spire built as facets
is how a spire is built anyway. meshAutoMovieSweep is not the
escape here that it is for the coordinate-less builders. It measures this
same pair, a tapered loft shears the same way, and it gives up something this
one keeps: the map here is exactly equiareal, while a loft's v is path
distance rather than surface distance, so a taper or a bend costs it texel
density on top of the angle. Where the form has to stay a true surface of
revolution, the finish on it stays non-directional.
Borrowed ordered meridian with at least two finite radius/height points in metres; radius x is nonnegative. Authored order sets winding; ends off the axis remain open rims.
Safe integer of at least three angular intervals over 2 pi; checked before allocation. One extra duplicate ring carries the UV seam.
A new unskinned mesh of (segments + 1) * profile.length vertices and 2 * segments * (profile.length - 1) triangles, with smooth normals and the metric atlas above.
Revolve a radius/height profile around local Y into a closed surface.
Each profile point is
x= radius andy= height, in metres. Unlike extrusion and sweep, the meridian is taken as authored rather than hulled, so an arbitrary silhouette polyline is allowed here; only its radii must be non-negative. Profile order owns the surface orientation, and every output buffer is new.Closure follows the meridian and is the caller's to declare: a meridian that starts and ends on the axis closes into a solid whose pole rings collapse to zero-area triangles, and one that does not is an open tube with a rim at each end. Neither is repaired.
The surface carries a metric atlas measured the way its siblings measure theirs: distance travelled around the section against distance travelled along the path. Around is the arc the vertex actually sits on,
thetatimes its own radius, so a baluster turned to a 0.3 m radius shows 1.88 m of grain around it rather than one turn of an image stretched to fit it. Along is the meridian's own polyline length, which is slant distance on a cone and developed height on a cylinder rather than the shortcut through the axis.Around runs against the direction the lattice is built in, and that is the whole reason it is written
2 * pi - thetarather thantheta. The outward normal this winding produces, the meridian running up, andthetarunning round form a left-handed triple, so an atlas keyed onthetadirectly comes out mirrored against every other builder here: a letter reads backwards, and a normal map's tangent basis is handed the wrong way, which tilts the lighting rather than merely the image. Reversing the axis that closes on itself costs nothing an author can name, because which way round a surface of revolution is traversed carries no authored meaning, while reversing the meridian instead would putv = 0at whichever end of the profile was typed last.The lattice carries a duplicate ring at
theta = 2 * pi, which is what makes the seam an honest cut: one edge of the seam reads zero and the other reads the full circumference at that radius, so the closing quad never interpolates backwards through the atlas. Those two phases coincide only when the declared texture repeat divides that circumference; the kernel does not stretch a finish to hide a mismatched seam. A changing radius also makes the surface non-developable in general. Scaling by radius preserves local circumference scale on every parallel and therefore shears the flat atlas between unequal parallels instead of pretending one global isometry exists. A meridian point on the axis is a pole, and its whole ring reads zero around, because a circle of no radius has no arc to travel. Its collapsed triangles remain the topology this operator declares; their UVs stay finite, and they are the one place the atlas has no handedness to report. The duplicate seam vertices share their accumulated face normals before normalization, and every duplicate of an axis pole shares that pole's normal as well, so the UV cut does not introduce a lighting discontinuity.How far that shear goes decides what finish the surface can carry. The map is exactly equiareal, its Jacobian determinant one everywhere, so the whole distortion is pure shear of
2 * pi - thetatimesdr / ds. Worst anisotropy over a full turn is therefore(sqrt(k * k + 4) + k) / (sqrt(k * k + 4) - k)atk = 2 * pi * |dr / ds|: 1.0 on a cylinder, 2.0 at 6.5 degrees off the axis, 4.0 at 13.8 degrees, 21.5 on a 45-degree cone, and 41 where the meridian runs level, which is what a dome's pole and a cone's tip approach. A tile that reads square beside the seam is drawn into a ribbon a quarter turn away, and the curve of constantuspirals instead of following a meridian.That is the price of the metric statement rather than a defect awaiting repair. Measuring true arc around forces
uto betheta * rplus somec(s), whose derivative would have to be-theta * dr / dsto cancel the shear, and no function ofsalone does that at everytheta. Choosingc(s)only moves where the shear vanishes, which is why it sits at the seam here; centring it would halve the worst case to 11.8 at a pole, still far past readable. The pole is not a second fault: cut the tip off a cone and the frustum that remains shears exactly as hard, so the collapsed pole triangles above are a topology note rather than the cause.A directional finish therefore wants a meridian within a few degrees of the axis, which is a drum, a shaft, a turret barrel. A sloped or domed form that must carry one is authored as flat faces through buildAutoMoviePolyhedron, whose per-face frame is isometric, or as a flat cap through extrudeAutoMovieRegion, and a spire built as facets is how a spire is built anyway. meshAutoMovieSweep is not the escape here that it is for the coordinate-less builders. It measures this same pair, a tapered loft shears the same way, and it gives up something this one keeps: the map here is exactly equiareal, while a loft's
vis path distance rather than surface distance, so a taper or a bend costs it texel density on top of the angle. Where the form has to stay a true surface of revolution, the finish on it stays non-directional.