Maximum call stack size exceeded triggered by importImageFileToBase64

[fetimo]

Hello!

When passing an ArrayBuffer to scribe.importFiles it throws RangeError: Maximum call stack size exceeded.

My test buffer is 236087 bytes (~236 KB) which isn't massive but I'm running on a shared hosting environment (Vercel) and I suspect that has something to do with it. It works locally with much larger files so I think it is impacted by the environment it's running in.

The line that seems to be causing it is

scribe.js/js/utils/imageUtils.js

Line 61 in 0fc036e

const binary = String.fromCharCode(...imageUint8);

I believe that because it's spreading in the array it's assigning a lot of resources quickly and is less efficient than some alternatives, like:

const imageUint8 = new Uint8Array(file.buffer, file.byteOffset, file.byteLength);

// Detect format from just the header
const format = detectImageFormat(file.subarray(0, 64));

const binary = file.toString('latin1');

resolve(`data:image/${format};base64,${btoa(binary)}`);
return;

Thanks for your time!

[fetimo]

I've done some more research and it looks like it's probably running into this: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Function/apply#:~:text=The%20consequences%20of%20calling%20a,coded%20argument%20limit%20of%2065536.). There's a limit on the number of arguments that can be passed to a function.

[Balearica]

Thanks for reporting and looking into this.

I've done some more research and it looks like it's probably running into this: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Function/apply#:~:text=The%20consequences%20of%20calling%20a,coded%20argument%20limit%20of%2065536.). There's a limit on the number of arguments that can be passed to a function.

This documentation indicates that this limit impacts JavaScriptCore. I was under the impression that Vercel primarily uses V8 rather than JavaScriptCore, and I am not finding any evidence that V8 has a similar limitation.

Can you provide more information regarding exactly what environment you are using in Vercel and/or if it's possible to recreate the issue without using Vercel? E.g. are you trying to run this as an edge function? In theory, it does not require pinning down the exact mechanism to fix the line of code you identified (assuming that is indeed the issue), however it could be useful to better understand what's going on. Additionally, certain environments impose such low resource limits that it is essentially impossible to run OCR, and so fixing would simply result in a failure at a later step.

[fetimo]

Great points.

I wrote a small script to test this:

let n = 0;
try {
  for (let i = 0; i < Infinity; ++i) {
    String.fromCharCode(...Array.from(Array(i).keys()));
    n = i;
  }
} catch (error) {
  console.log(`Max arity is ${n}`);
}

However, the mystery thickens with the results!

On Vercel ('Fluid Compute', 800s timeout, 1vCPU, 2GB RAM, Node.js 22.15.1, V8 12.4.254.21-node.24) I get a maximum value of 122,979.
On my local computer it's lower, at 105,567 (M1 Pro, 16GB RAM, Node.js 22.15.1, V8 12.4.254.21-node.24)!

The Vercel function is capable of doing OCR as I have successfully got it working with Tesseract.js and using Scribe.js by writing to an image file instead. This issue only presents itself when using ArrayBuffers.

Edit: Re-ran it with the same versions.