safari demo error (without-a-bundler-no-modules)

[cyanBone]

Summary

wasm-bindgen/examples/without-a-bundler-no-modules
This demo is successful in chrome, but there are errors in Safari

errors: Unhandled Promise Rejection: InvalidCharacterError: The string contains invalid characters.

Safari version 13.0.2

Additional Details

[alexcrichton]

Thanks for the report @cyanBone! I've updated the title of the issue here to reflect what's going on after debugging a bit. I turns out this JS snippet:

(new TextDecoder('utf-8', { ignoreBOM: true })).decode(new Uint8Array([112])).charCodeAt(0)

returns 65279 on Safari and 112 on Firefox. I believe the BOM is being included on the first use of TextDecoder on Safari whereas on other browsers it's only showing up if it was used. While this is likely a bug in Safari we should probably work around it ourselves locally.

@Pauan have you seen this before perchance? I figure the easiest thing to do would be to just decode something immediately to clear out the bom state on safari.

[Pauan]

@alexcrichton Uh, wow, that's an incredible bug in Webkit. It seems that it's always including the BOM even if the BOM doesn't exist in the Uint8Array.

I can't test Safari, but could you also try these?

(new TextDecoder('utf-8', { ignoreBOM: false })).decode(new Uint8Array([])).charCodeAt(0)

(new TextDecoder('utf-8', { ignoreBOM: true })).decode(new Uint8Array([])).charCodeAt(0)

var decoder = new TextDecoder('utf-8', { ignoreBOM: true });
console.log(decoder.decode(new Uint8Array([112])).charCodeAt(0));
console.log(decoder.decode(new Uint8Array([112])).charCodeAt(0));

[alexcrichton]

(new TextDecoder('utf-8', { ignoreBOM: false })).decode(new Uint8Array([])).charCodeAt(0)

Prints NaN

(new TextDecoder('utf-8', { ignoreBOM: true })).decode(new Uint8Array([])).charCodeAt(0)

Prints 65279

var decoder = new TextDecoder('utf-8', { ignoreBOM: true });
console.log(decoder.decode(new Uint8Array([112])).charCodeAt(0));
console.log(decoder.decode(new Uint8Array([112])).charCodeAt(0));

Prints 65279

then prints 112

---

Ok, sounds like we should probably just decode the empty array immediately to flush the bom out of the decoder?

[Pauan]

Ok, sounds like we should probably just decode the empty array immediately to flush the bom out of the decoder?

Yeah, though we should have some extensive tests for this (decoding with BOM, decoding without BOM, decoding with BOM after not-BOM, etc.)

[cyanBone]

@alexcrichton @Pauan

There is also wasm in golang. I can see from its JavaScript that it can run correctly. Here I give the JavaScript in golang.this javascritp has TextDecoder

https://github.com/golang/go/tree/master/misc/wasm

[Pauan]

@cyanBone Yeah, that's because they don't use ignoreBOM: true, the bug with Safari only happens with ignoreBOM: true.

[Pauan]

PR up: #1829 (I'd really like some unit tests though)