Tesselation is broken

[isovector]

Filing a new bug to act as a little notepad for myself.

The following object has a huge erroneous spike when rendered with resolution=2:

translate (V3 0.0 (-27.5) 0.0) (union [translate (V3 62.0 0.0 33.0) (union [translate (V3 (-124.0) 0.0 0.0) (cubeR 0.0 False (V3 124.0 124.0 2.0)),translate (V3 (-56.0) 62.0 (-33.0)) (union [translate (V3 0.0 0.0 35.0) (union [translate (V3 0.0 0.0 (-17.5)) (difference (translate (V3 0.0 0.0 (-17.5)) (cylinder 42.0 35.0)) [translate (V3 0.0 0.0 (-15.5)) (cylinder 40.0 33.0)]),union [translate (V3 0.0 0.0 (-2.0)) (cylinder 48.0 2.0),emptySpace]]),emptySpace])]),translate (V3 (-3.5) (-50.0) 0.0) (union [translate (V3 0.0 (-9.5) 0.0) (union [translate (V3 0.0 0.0 8.5) (rotate3 (3.141592653589793,0.0,0.0) (translate (V3 (-62.0) (-9.5) (-8.5)) (cubeR 0.0 False (V3 124.0 19.0 17.0)))),union [translate (V3 61.0 0.0 0.0) (union [translate (V3 0.0 (-4.0) 0.0) (cubeR 0.0 False (V3 8.0 8.0 2.0)),translate (V3 (-2.0) (-4.0) 0.0) (cubeR 0.0 False (V3 2.0 8.0 8.0))]),emptySpace]]),translate (V3 0.0 52.0 0.0) (union [translate (V3 (-52.0) (-54.0) 0.0) (cubeR 0.0 False (V3 104.0 54.0 17.0)),translate (V3 0.0 (-2.0) 0.0) (difference (extrudeR 0.0 (polygonR 0.0 [V2 (-48.0) 0.0,V2 48.0 0.0,V2 0.0 60.0]) 17.0) [translate (V3 0.0 0.0 2.0) (extrudeR 0.0 (polygonR 0.0 [V2 (-46.0) 0.0,V2 46.0 0.0,V2 0.0 58.0]) 13.0)])])])])

[isovector]

The STL contains the following triangles, which have X and Y components 10x bigger than anything else in the file. Presumably these are the culprits:

facet normal 7.808541e-1 -6.247135e-1 0.0e0
outer loop
vertex -3.200224e1 -3.1188211e0 -1.3e1
vertex 1.1662979e2 1.8266225e2 -1.3e1
vertex 4.2313778e1 8.977171e1 -1.2e1
endloop
endfacet
facet normal 7.808541e-1 -6.247135e-1 0.0e0
outer loop
vertex 1.1662979e2 1.8266225e2 -1.3e1
vertex 1.1662979e2 1.8266225e2 -1.1e1
vertex 4.2313778e1 8.977171e1 -1.2e1
endloop
endfacet
facet normal 7.808541e-1 -6.247135e-1 0.0e0
outer loop
vertex 1.1662979e2 1.8266225e2 -1.1e1
vertex -3.200224e1 -3.1188211e0 -1.1e1
vertex 4.2313778e1 8.977171e1 -1.2e1
endloop
endfacet
facet normal -7.846233e-1 6.1997277e-1 0.0e0
outer loop
vertex 1.1662979e2 1.8266225e2 -1.1e1
vertex 1.1662979e2 1.8266225e2 -1.3e1
vertex 4.289338e1 8.934312e1 -1.2e1
endloop
endfacet
facet normal -7.846233e-1 6.1997277e-1 0.0e0
outer loop
vertex 1.1662979e2 1.8266225e2 -1.3e1
vertex -3.0843039e1 -3.9760044e0 -1.3e1
vertex 4.289338e1 8.934312e1 -1.2e1
endloop
endfacet
facet normal -7.846233e-1 6.1997277e-1 0.0e0
outer loop
vertex -3.0843039e1 -3.9760044e0 -1.3e1
vertex -3.0843039e1 -3.9760044e0 -1.1e1
vertex 4.289338e1 8.934312e1 -1.2e1
endloop
endfacet
facet normal -7.846233e-1 6.1997277e-1 0.0e0
outer loop
vertex -3.0843039e1 -3.9760044e0 -1.1e1
vertex 1.1662979e2 1.8266225e2 -1.1e1
vertex 4.289338e1 8.934312e1 -1.2e1
endloop
endfacet

(search for e2 in the above to spot the vertices)

[isovector]

The spike from another angle. It's not just a single triangle:

[isovector]

Rendering the same object with resolutions {3,4,5,6,9,10} also produce crazy, but different, spikes. It doesn't (visibly) occur at any of {1, 2.5, 7, 8}.

[isovector]

At resolution 10:

The vertex here is at vertex -3.4342197e3 1.4210855e-14 -1.6341694e1, which is 10x further away from the model than in the resolution 2 case. Maybe the resolution has something to how crazy these triangles get? If so, it would explain why we don't see them when we decrease the resolution.

[isovector]

The spike remains all the way until resolution 10.0695652173913030580365557398181408. At 10.0695652173913030580365557398181409, we get a drastically different mesh:

[isovector]

I instrumented the triangle generation with this:

mkTriangle :: (ℝ3, ℝ3, ℝ3) -> Triangle
mkTriangle (a, b, c)  =
  if abs a >= 3400
     then markDebugger $ Triangle (a, b, c)
     else Triangle (a, b, c)

markDebugger :: Triangle -> Triangle
markDebugger = id

and then added a breakpoint to markDebugger. The resulting stack trace is this:

> :hist
-1  : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:140:11-43)
-2  : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:139:6-10)
-3  : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:139:6-18)
-4  : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:(139,3)-(141,28))
-5  : tesselateLoop (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:113:29-48)
-6  : rnf (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Definitions.hs:192:28-38)
-7  : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:141:11-28)
-8  : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:139:6-10)
-9  : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:139:6-18)
-10 : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:(139,3)-(141,28))
-11 : tesselateLoop (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:113:29-48)
-12 : rnf (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Definitions.hs:192:28-38)
-13 : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:141:11-28)
-14 : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:139:6-10)
-15 : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:139:6-18)
-16 : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:(139,3)-(141,28))
-17 : tesselateLoop (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:113:29-48)
-18 : rnf (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Definitions.hs:192:28-38)
-19 : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:141:11-28)
-20 : mkTriangle (/home/sandy/prj/ImplicitCAD/Graphics/Implicit/Export/Render/TesselateLoops.hs:139:6-10)

[isovector]

This is the offending line:

https://github.com/colah/ImplicitCAD/blob/f04ebb57a7809067a36cf97f223891d49babdb4f/Graphics/Implicit/Export/Render/TesselateLoops.hs#L93

[isovector]

some printf of the values in scope (ghci doesn't track this for some stupid reason)

mid= V3 (-318.298453912407) (-5.033029043219576) 2.365568688888259
mid'= V3 (-65.49999999494523) 24165.67241446491 155.07559049902787
midval= 252.798453912407
deriv= 1.045809707875378e-2
normal= V3 (-1.0458097078962893e-2) (-0.9999253558637543) (-6.317507913012856e-3)

[isovector]

The weird point comes from path, which is defined:

https://github.com/colah/ImplicitCAD/blob/f04ebb57a7809067a36cf97f223891d49babdb4f/Graphics/Implicit/Export/Render/TesselateLoops.hs#L78

so maybe it's shrinkLoop that's broken

[sorki]

Maybe simpler repro:

res = 2
s = 6.79
obj = rotate3 (V3 0 (0.9 * 2 * pi) 0) $ union $ [
    translate (V3 0 0 (-s)) (cylinder2 1 3 7) 
  , translate (V3 0 0 s) $ rotate3 (V3 pi 0 0) $  cylinder2 1 3 7
  ]  

With res = 1 and others it's fine

[isovector]

Found it! e7b2cbd

interpolate doesn't do a very good job of finding roots.

[isovector]

#355 is a better step here, but it's still not right