CT: assertion failure in Graph.computeWindingMap #72
Comments
|
Still present |
|
It fails at this point in 1459 assert && assert( forwardFace !== reversedFace ); /**
* Computes the winding map for each face, starting with 0 on the unbounded face (for each shapeId).
* @private
*/
computeWindingMap() {
const edges = this.edges.slice();
// Winding numbers for "outside" are 0.
const outsideMap = {};
for ( let i = 0; i < this.shapeIds.length; i++ ) {
outsideMap[ this.shapeIds[ i ] ] = 0;
}
this.unboundedFace.windingMap = outsideMap;
// We have "solved" the unbounded face, and then iteratively go over the edges looking for a case where we have
// solved one of the faces that is adjacent to that edge. We can then compute the difference between winding
// numbers between the two faces, and thus determine the (absolute) winding numbers for the unsolved face.
while ( edges.length ) {
for ( let j = edges.length - 1; j >= 0; j-- ) {
const edge = edges[ j ];
const forwardHalf = edge.forwardHalf;
const reversedHalf = edge.reversedHalf;
const forwardFace = forwardHalf.face;
const reversedFace = reversedHalf.face;
assert && assert( forwardFace !== reversedFace );
const solvedForward = forwardFace.windingMap !== null;
const solvedReversed = reversedFace.windingMap !== null;
if ( solvedForward && solvedReversed ) {
edges.splice( j, 1 );
if ( assert ) {
for ( let m = 0; m < this.shapeIds.length; m++ ) {
const id = this.shapeIds[ m ];
assert( forwardFace.windingMap[ id ] - reversedFace.windingMap[ id ] === this.computeDifferential( edge, id ) );
}
}
}
else if ( !solvedForward && !solvedReversed ) {
continue;
}
else {
const solvedFace = solvedForward ? forwardFace : reversedFace;
const unsolvedFace = solvedForward ? reversedFace : forwardFace;
const windingMap = {};
for ( let k = 0; k < this.shapeIds.length; k++ ) {
const shapeId = this.shapeIds[ k ];
const differential = this.computeDifferential( edge, shapeId );
windingMap[ shapeId ] = solvedFace.windingMap[ shapeId ] + differential * ( solvedForward ? -1 : 1 );
}
unsolvedFace.windingMap = windingMap;
}
}
}
} |
|
This happens rather irregularly on Bayes. My initial assumption is that it is a corner case. Possibly when the spotlight gets so small that it becomes a point on the edge of the projector screen. This would create an a graph union that could potentially fail. This is pure speculation on my part. I have not been able to reproduce it. |
In a setup like below, I can at least get the sim to freeze up (from poor performance, I assume), but I get no error. Until such time that I determine if this screenshot is at all related to this issue, or this is its own issue, I will just leave it here. I got this on Win 10 FF.
|
|
I tried to setup cases where the intersections of two shapes might fail, say a circle "kissing" a rectangle but that does not trigger an assertion. |
|
Even the case of disk of radius 0 on the edge of a rectangle is handled correctly. |
|
@jonathanolson , I think this is a Kite bug that involves Shape.intersection. We are passing two shapes and want to determine their intersections. ScreenShape is a fixed, well defined parallelogram. I added some assertions to diskShape to make sure its shape is legit. However, I have not been able to trigger any assertions within geometric optics itself. The assertion that is triggered is within kite. I must admit that I dont understand what computeWindingMap is doing. It would be useful to get some insight into what is going on. Unfortunately, even though this assertion is triggered on a rather regular basis on Bayes, we dont have very clear procedure to trigger it. |
|
Assigning @jonathanolson to look into it. |
|
I added assertions to make sure that the two shapes that are passed are appropriate. We will monitor Bayes to see if that proves useful. |
|
Using BinaryResult instead of the more general intersection method still trigger assertions. |
|
(I will be looking into this) |
|
This is really reproducible ... |
|
Thanks @kathy-phet. Reduce the slider until you reach the index of refraction 1.36.
|
|
In the above case, it is the "second source" spotlight that fails. |
|
I tried reproducing using the above instructions in master, and couldn't reproduce. I tried fuzzing for 30 minutes, both with master and a dev version (https://phet-dev.colorado.edu/html/geometric-optics/1.0.0-dev.9/phet/geometric-optics_all_phet_debug.html?fuzz). Thoughts on how I can reproduce? Is it potentially dependent on screen size? Alternatively, if you could toString() both of the shapes being intersected and report that, I could presumably reproduce the kite error from that. |
|
We have moved the initial position of the source since, and this affects the reproduction instructions. |
|
I tried to co back to our previous initial positions for the sources but unfortunately we also changed the size and position of the projector screen during a refactor, so I have not been able to reproduce it. |
|
Considering the comment from @jonathanolson , I'll attempt to go back to a previous commit where KP was able to reproduce it by merely moving the sliders. |
|
Since this is not in the prototype, unassigning until I start working on public production issues. |
|
Note that Spotlight was renamed LightSpot, and SpotlightNode was renamed LightSpotNode, per #196. |
|
Still failing occassionally in CT: |
|
We'll need @jonathanolson to take at look at this. Like #163, this is a problem in kite.Graph. |
pixelzoom
changed the title
|
I'm going to close this because: (1) I haven't seen this in CT since 10/19/21, almost 4 months ago. |
|
Reopening. While this problem is rare, it's not resolved. I hit it while running the State Wrapper for ~5 minutes in my local copy with |
|
I fuzzed tested locally for 30 minutes and did not encounter this problem. Test URLs:
|
|
|
This is a non-blocking error. It won't cause the sim to crash, and is unlikely to result in any noticeable problems in practice. 1.1 was published with this as a known issue. |
|
In c8f18cd, the implementation changed. LightSpot.ts no longer computes intersection with the projection screen, |
The text was updated successfully, but these errors were encountered: