Fix: clearSceneAndBustCache when rust panics because it can load a stale cache

[nadr0]

closes #6808

Issue

When rust panics the rust cache can become stale and prevent you from rendering other KCL files in your project.

Implementation

• added new custom unreachable panic error message matching
• added generic stack track panic error message for rust runtime
• clearSceneAndBustCache if rust panics to prevent the cache from not re executing when it needs to
• Tried grabbing a specific enough keyword from app.zoo.dev when it crashes in the stack trace for the generic handler
  

Video walkthrough

2025-05-09.12-26-07.mp4

KCL code that causes panic

/* Generated by Text-to-CAD:
Create a cube with sides of length 20mm.  On the top face, add a cylindrical boss of radius 6mm and height 2mm. Then, shell the body to 2mm thickness, leaving the top face of the cylindrical boss open. */
@settings(defaultLengthUnit = mm)

// Define the dimensions of the cube
cubeSide = 20

// Define the dimensions of the cylindrical boss
bossRadius = 6
bossHeight = 2

// Define the shell thickness
shellThickness = 2

// Define the fillet radius
filletRadius = 2

// Create a sketch for the cube
cubeSketch = startSketchOn(XY)
  |> startProfile(at = [-cubeSide / 2, -cubeSide / 2])
  |> xLine(length = cubeSide)
  |> yLine(length = cubeSide, tag = $seg01)
  |> xLine(length = -cubeSide)
  |> yLine(length = -cubeSide, tag = $seg02)
  |> close()

// Extrude the cube
cube = extrude(
       cubeSketch,
       length = cubeSide,
       tagEnd = $topFace,
       tagStart = $capStart001,
     )

// Apply fillets to all edges of the cube
filletedCube = fillet(
  cube,
  radius = filletRadius,
  tags = [
    getNextAdjacentEdge(seg01),
    getNextAdjacentEdge(seg02),
    getNextAdjacentEdge(getNextAdjacentEdge(seg01)),
    getNextAdjacentEdge(getNextAdjacentEdge(seg02))
  ]
)

// Create a sketch for the cylindrical boss on the top face of the cube
bossSketch = startSketchOn(filletedCube, face = topFace)
  |> circle(center = [0, 0], radius = bossRadius)

// Extrude the cylindrical boss
boss = extrude(bossSketch, length = bossHeight, tagEnd = $bossTopFace)

// Shell the cube, leaving the top face of the cylindrical boss open
shell([filletedCube, boss], thickness = shellThickness, faces = [bossTopFace])

[qa-wolf]

QA Wolf here! As you write new code it's important that your test coverage is keeping up.
Click here to request test coverage for this PR!

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
modeling-app	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	May 9, 2025 5:57pm

[nadr0]

what the fuck is this?

cargo fmt does nothing

[jtran]

It's inside a Rust string, so cargo fmt shouldn't modify your strings.

[jessfraz]

ah smart! perfect use of that function but also why is rust panicing hahaa we should never panic

[nadr0]

hahaa we should never panic

Well that is something we are confused about. I've talk to Jon and Max about this and I've been able to come across 3 types of panics in Rust. That is why we have this global window event handler as well as the WASM module reload.

I took a valid main.kcl file, asked t2c edit to add fillets and the result produced a file that causes a rust panic. When rust panics its breaks out of the Javacript runtime which will prevent the cleanup workflow.

^--- this is the specific panic from the KCL code I listed above.

Seeing that this can randomly happen I want to get this catch all handler in place to prevent users from getting in an unrecoverable state and they only way to fix it is by refreshing.

The root cause has not been debugged on the Rust side. Since this is the 2nd one I've found while using the app I am worried we will have another oddly specific panic and then the user will have an unrecoverable state so I want to prevent that.

[codspeed-hq]

CodSpeed Instrumentation Performance Report

Merging #6809 will not alter performance

_{Comparing nadro/gh-6808/rust-panics-cache-bricks (a7ee874) with main (923fead)}

Summary

✅ 54 untouched benchmarks

[jtran]

Suggested change

	return stack.indexOf(substringError) !== -1 ? true : false
	return stack.indexOf(substringError) !== -1

[jtran]

Do we send error reports to the server to notify us when stuff like this gets triggered in the wild? If not, we should.

[jessfraz]

@jtran we could add honeycomb tracing or something just wanna be mindful when we do we only track dire shit or we will get so much noise! but would be good to do!

[jessfraz]

@jtran have you looked at that args.rs line hes talking about does it make sense why its panic-ing

[jessfraz]

its this line:

modeling-app/rust/kcl-lib/src/std/args.rs

Line 411 in a7ee874

Ok(T::from_kcl_val(&arg).unwrap())

we should make that not unwrap, i can fix

[jessfraz]

i opened a fix for the panic here: #6825
restarting the wasm isnt gooing to fix this, it will just panic again