From 4a3098f1f27594772129ccc393e5d4f44f2f1921 Mon Sep 17 00:00:00 2001
From: Terry Jia <terryjia88@gmail.com>
Date: Mon, 8 Dec 2025 20:41:23 -0500
Subject: [PATCH] performance fix: prevent gcd infinite loop with
 floating-point step values (#7258)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

## Summary

report and fix https://github.com/Comfy-Org/ComfyUI_frontend/issues/3919

- Convert recursive gcd to iterative to avoid stack overflow
- Add epsilon tolerance (1e-10) for floating-point precision issues

This fixes workflow loading hangs when node trying merge values like
0.01 and 0.001, which caused the original recursive gcd to run
indefinitely due to floating-point modulo never reaching exactly zero.

please notice, we need both iterative and epsilon together to fix this
gcd issue

Call Chain

PrimitiveNode.onAfterGraphConfigured
  → #mergeWidgetConfig
    → #isValidConnection
      → mergeIfValid
        → mergeInputSpec
          → mergeNumericInputSpec
            → lcm(step1, step2)
              → gcd(a, b)  ← Problem here

Why It Happened
When some nodes connect to multiple nodes, it may merge values using
LCM, which internally calls GCD.

Original recursive implementation:
```
export const gcd = (a: number, b: number): number => {
   return b === 0 ? a : gcd(b, a % b)
}
```

Issues:
1. Stack Overflow: Recursive calls with many nodes exhausted the call
stack.
2. Floating-Point Precision: For values like gcd(0.01, 0.001):
 ` 0.01 % 0.001 = 0.0009999999999999994  // Not exactly 0!`
3. Due to Ifloating-point representation, the modulo never reaches
exactly zero, causing hundreds or thousands of iterations.

## Screenshots
### before


https://github.com/user-attachments/assets/cca4342c-a882-4590-a8d4-1e0bea19e5b7

### fix with only iterative, without epsilon


https://github.com/user-attachments/assets/1dc52aa4-a86a-40b5-8bac-904094c4c36b


### final fix with iterative and epsilon

https://github.com/user-attachments/assets/7b868b50-c3c9-4be4-8594-27cecbc08a26

┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7258-performance-fix-prevent-gcd-infinite-loop-with-floating-point-step-values-2c46d73d3650818cbe8cf455c934a114)
by [Unito](https://www.unito.io)
---
 src/utils/mathUtil.ts                 | 28 +++++++++++++++++++++++++--
 tests-ui/tests/utils/mathUtil.test.ts | 13 +++++++++++++
 2 files changed, 39 insertions(+), 2 deletions(-)

diff --git a/src/utils/mathUtil.ts b/src/utils/mathUtil.ts
index 8655451c5..4dad40df5 100644
--- a/src/utils/mathUtil.ts
+++ b/src/utils/mathUtil.ts
@@ -2,14 +2,38 @@ import type { ReadOnlyRect } from '@/lib/litegraph/src/interfaces'
 import type { Bounds } from '@/renderer/core/layout/types'
 
 /**
- * Finds the greatest common divisor (GCD) for two numbers.
+ * Finds the greatest common divisor (GCD) for two numbers using iterative
+ * Euclidean algorithm. Uses iteration instead of recursion to avoid stack
+ * overflow with large inputs or small floating-point step values.
+ *
+ * For floating-point numbers, uses a tolerance-based approach to handle
+ * precision issues and limits iterations to prevent hangs.
  *
  * @param a - The first number.
  * @param b - The second number.
  * @returns The GCD of the two numbers.
  */
 export const gcd = (a: number, b: number): number => {
-  return b === 0 ? a : gcd(b, a % b)
+  // Use absolute values to handle negative numbers
+  let x = Math.abs(a)
+  let y = Math.abs(b)
+
+  // Handle edge cases
+  if (x === 0) return y
+  if (y === 0) return x
+
+  // For floating-point numbers, use tolerance-based comparison
+  // This prevents infinite loops due to floating-point precision issues
+  const epsilon = 1e-10
+  const maxIterations = 100
+
+  let iterations = 0
+  while (y > epsilon && iterations < maxIterations) {
+    ;[x, y] = [y, x % y]
+    iterations++
+  }
+
+  return x
 }
 
 /**
diff --git a/tests-ui/tests/utils/mathUtil.test.ts b/tests-ui/tests/utils/mathUtil.test.ts
index c4cb16dd8..47a70083d 100644
--- a/tests-ui/tests/utils/mathUtil.test.ts
+++ b/tests-ui/tests/utils/mathUtil.test.ts
@@ -12,6 +12,19 @@ describe('mathUtil', () => {
       expect(gcd(0, 5)).toBe(5)
       expect(gcd(5, 0)).toBe(5)
     })
+
+    it('should handle negative numbers', () => {
+      expect(gcd(-48, 18)).toBe(6)
+      expect(gcd(48, -18)).toBe(6)
+      expect(gcd(-48, -18)).toBe(6)
+    })
+
+    it('should not cause stack overflow with small floating-point step values', () => {
+      // This would cause Maximum call stack size exceeded with recursive impl
+      // when used in lcm calculations with small step values
+      expect(() => gcd(0.0001, 0.0003)).not.toThrow()
+      expect(() => gcd(1e-10, 1e-9)).not.toThrow()
+    })
   })
 
   describe('lcm', () => {