Say how each laid piece samples its material, in the UV transform the PBR
record already carries.
A pattern is not a texture repeat, but what a laid piece finally shows is
still a texture, and the way to show it is the one the material record
already has: an offset, a scale, and a rotationDeg on its texture
reference. Nothing new is invented here and no second texturing path is
opened; what this adds is the arithmetic that turns a piece's own place,
size, rotation, grain, and flip into that transform, so a book-matched pair
is a real mirrored image rather than two slabs a viewer cannot tell apart.
The mesh UV this is applied to is a point's place inside the module rectangle
the occurrence was generated at, normalized: the binding this fills in is a
"normalized" coordinate source and not the "surface-metres" one an
atlas-bearing procedural surface emits, so a prototype has to be a unit square
by construction rather than by luck, and the binding has to say which
arithmetic it was written in. A whole piece therefore spans
[0, 1] on both axes, which is what a unit-square prototype gives, and a cut
piece spans the sub-rectangle the cut left it rather than being renormalized
over its own outline, because a renormalized cut piece would show the whole
image squeezed into the surviving sliver. The sheet is turned by the piece's
own IAutoMoviePatternPlacement.grainDeg, so a board whose grain runs
along it samples straight while a slab set across the grain samples across
it.
One call states one material's sheet. A run whose zones carry different
materials calls it once per material and keeps the occurrences belonging to
that material's zones, exactly as the prototype table is stated per call.
At zero texture centre, the viewer/Three matrix is diag(scale) · R(-rotationDeg) · uv + offset: it rotates the UV column vector first, then applies axis scales. Orthogonal rows admit this ideal factorization; the implementation accepts their relative dot product within 1e-9 and reports larger residual skew by occurrence id.
Every placement of the run lands in exactly one of the two lists, in
placement order. Throws when a tile length is not a finite number above zero
or a face sheet's origin is not finite.
Say how each laid piece samples its material, in the UV transform the PBR record already carries.
A pattern is not a texture repeat, but what a laid piece finally shows is still a texture, and the way to show it is the one the material record already has: an
offset, ascale, and arotationDegon its texture reference. Nothing new is invented here and no second texturing path is opened; what this adds is the arithmetic that turns a piece's own place, size, rotation, grain, and flip into that transform, so a book-matched pair is a real mirrored image rather than two slabs a viewer cannot tell apart.The mesh UV this is applied to is a point's place inside the module rectangle the occurrence was generated at, normalized: the binding this fills in is a
"normalized"coordinate source and not the"surface-metres"one an atlas-bearing procedural surface emits, so a prototype has to be a unit square by construction rather than by luck, and the binding has to say which arithmetic it was written in. A whole piece therefore spans[0, 1]on both axes, which is what a unit-square prototype gives, and a cut piece spans the sub-rectangle the cut left it rather than being renormalized over its own outline, because a renormalized cut piece would show the whole image squeezed into the surviving sliver. The sheet is turned by the piece's own IAutoMoviePatternPlacement.grainDeg, so a board whose grain runs along it samples straight while a slab set across the grain samples across it.One call states one material's sheet. A run whose zones carry different materials calls it once per material and keeps the occurrences belonging to that material's zones, exactly as the prototype table is stated per call.
At zero texture centre, the viewer/Three matrix is
diag(scale) · R(-rotationDeg) · uv + offset: it rotates the UV column vector first, then applies axis scales. Orthogonal rows admit this ideal factorization; the implementation accepts their relative dot product within1e-9and reports larger residual skew by occurrence id.Every placement of the run lands in exactly one of the two lists, in placement order. Throws when a tile length is not a finite number above zero or a face sheet's origin is not finite.