Possibly broken in IE11

[Miladiir]

I noticed that JSBI is broken in a test application when running in IE11.

When I exclude JSBI in webpacks babel-loader exclude (jsbi is parsed like the rest of my code):

This possibly points to:

if (!_(t) || i(t) !== t) throw new RangeError("The number " + t + " cannot be converted to BigInt because it is not an integer");

I use core-js@3 and regenerator-runtime as polyfills. Maybe I need something else as well?

When I do not exlude JSBI and treat it like any other dependency (jsbi is included in babel-loader):

This time it is unfortunately absolutely impossible to tell what upsets poor IE. Maybe its the same statement, who knows 🤷‍♂️.

Is the UMD version tested? Works fine in Firefox 80, but not in IE11.

I also noticed #42. I tested importing the common-js and mjs versions directly, however I still get similar results. IE is broken while Firefox works.

[glr0221]

@Miladiir

I was about to use this library for a project of mine. Did this library ever work for IE11? I mean if there was a version of this library that worked for IE11? Currently i need support for two core objects/functions BigInt() and getBigUint64(). Thank you.

[Miladiir]

@Miladiir

I was about to use this library for a project of mine. Did this library ever work for IE11? I mean if there was a version of this library that worked for IE11? Currently i need support for two core objects/functions BigInt() and getBigUint64(). Thank you.

I do not know whether it ever worked. However currently definetly not. https://github.com/GoogleChromeLabs/jsbi/blob/5e757a2d1c51e3722fe7c59edbecf1be93ff187e/.babelrc which is the version on master as of today does not include IE11 as a target. The issue is related to the fact, that I cannot even get to get it working, even if I set the target to include IE11 myself. Maybe I can get it working. In that case I will ping you.

[jakobkummerow]

Looks like IE11 doesn't support Math.imul and Math.clz32, which JSBI uses. At least, that's the result I get when I comment out "minify()" in rollup.config.js, regenerate the UMD version, and test var a = JSBI.BigInt('12345678901234567890') with that on an IE11 VM.

As a short-term workaround, you can provide a very simple polyfill (which does not have the proper semantics, but is good enough for JSBI):

if (!Math.imul) {
  Math.imul = function(a, b) {
    return ((a & 0xFFFF) * (b & 0xFFFF)) | 0;
  }
}
if (!Math.clz32) {
  Math.clz32 = function(x) {
    if (x === 0) return 32;
    return 31 - (Math.log(x >>> 0) / Math.LN2 | 0) | 0;
  }
}

If you're worried about breaking other modules that might want to use these functions, you can find functionally fully correct polyfills on MDN.

I know next to nothing about Babel. Presumably it has a way to polyfill Math functions, but I don't know how to make it do that.

[Yaffle]

@jakobkummerow ,

if (!Math.clz32) {
  Math.clz32 = function(x) {
    if (x === 0) return 32;
    return 31 - Math.floor(Math.log((x >>> 0) + 0.5) * Math.LOG2E);
  }
}

[jakobkummerow]

@Yaffle : interesting fix, thanks! I actually copied that line from MDN; turns out different translations of the Math.clz32 page provide different polyfills, at least two of them omit the +0.5 (which I can confirm is buggy e.g. for x = 0x10000000), a few others use / Math.LN2 instead of * Math.LOG2E, which also avoids the issue (yay for finite precision math!). Oh well, or one could implement bit fiddling.

I'll edit my earlier post to avoid any confusion.

[Miladiir]

Does that mean that the target only has transformed syntax, but does not include the proper polyfills? Thats really strange. Is that a misconfiguration or on purpose. Maybe a maintainer can clarify. I thank you for figuring out the problem.

One thing I'd like to add though, is that polyfills on MDN are not alway according to standard (or correct at all). For proper polyfills for ECMA standard behaviour, please take a look at core-js. One does not need to import the whole project, but it is quite modular.

As I wrote in the original post:

I use core-js@3 and regenerator-runtime as polyfills. Maybe I need something else as well?

Which I still find really strange. For some reason my configuration did not pick up on the missing functionality in my target, or the required shims are missing in my libs. I will try to find the issue sometime this week hopefully. I'll keep you updated.

[mathiasbynens]

Looks like IE11 doesn't support Math.imul and Math.clz32, which JSBI uses. At least, that's the result I get when I comment out "minify()" in rollup.config.js, regenerate the UMD version, and test var a = JSBI.BigInt('12345678901234567890') with that on an IE11 VM.

As a short-term workaround, you can provide a very simple polyfill (which does not have the proper semantics, but is good enough for JSBI):

if (!Math.imul) {
  Math.imul = function(a, b) {
    return ((a & 0xFFFF) * (b & 0xFFFF)) | 0;
  }
}
if (!Math.clz32) {
  Math.clz32 = function(x) {
    if (x === 0) return 32;
    return 31 - (Math.log(x >>> 0) / Math.LN2 | 0) | 0;
  }
}

If you're worried about breaking other modules that might want to use these functions, you can find functionally fully correct polyfills on MDN.

IMHO this would be the best approach, except preferably we wouldn’t install on Math, but just keep these as local helper utilities that we don‘t expose globally or on any built-ins.

[glr0221]

I just wanted to say, I have successfully used JSBI with IE. Thank you so much @jakobkummerow @mathiasbynens @Miladiir