Add BigInt64Array shim for Safari 14

[hoodmane]

As discussed in #16693, -s WASM_BIGINT requires both BigInt (present since Safari 14.0 in 90.91% of users' browsers) and BigInt64Array (present since Safari 15.0 in 87.67% of users' browsers). We want to target Safari 14.0 as our minimal supported version, so we need this shim to work.

I didn't shim all of the TypedArray APIs, just the ones that I think are most important. It would be pretty easy to shim the rest of them if you want.

[kripken]

Is this fast enough to be useful?

We usually put polyfills under src/polyfill/, see examples there. It should also be gated behind a check to avoid the constant code size overhead. Perhaps it could check MIN_SAFARI_VERSION perhaps.

[hoodmane]

Is this fast enough to be useful?

I dunno, probably? Do you have a specific benchmark in mind? The main goal is to be able to use -sWASM_BIGINT so we can avoid patches like #16693, but without dropping support for Safari 14 entirely. It's okay if some code paths are a bit slow for Safari 14 I just don't want it to be raising a ReferenceError and failing on startup.

[sbc100]

This looks good to me as it stands.

We have not used globalThis in any of our other polyfills that I know of... but if it works in this case I guess that is fine.

[hoodmane]

Oh I can just use var because it doesn't respect scopes:

function f(){
  if(true) {
    var x = 2;
  }
  console.log(x); // 2
}
console.log(x);  // ReferenceError

[hoodmane]

This raises:

parseTools.js preprocessor error in preamble.js:289: "#if !POLYFILL"!
Internal compiler error in src/compiler.js!
Please create a bug report at https://github.com/emscripten-core/emscripten/issues/
with a log of the build and the input files used to run. Exception message: "ReferenceError: POLYFILL is not defined" | ReferenceError: POLYFILL is not defined

[hoodmane]

Oh that may only be a problem in the backport.

[hoodmane]

Okay I added a fairly thorough test suite and caught a few bugs. There are still a few cases where the real thing throws an error but my shim just does something, but I don't think it matters.

[hoodmane]

Can this be merged?

[hoodmane]

Bump

[sbc100]

For the other polyfils we have tended to test them in place ? By running emscripten -sXXX_VERSION=YY to enable them. But this direct testing you have implemented seems reasonable too.

[hoodmane]

Bump

[kripken]

@sbc100 , any remaining comments from you? I think everything was resolved but didn't want to merge before asking.

[sbc100]

lgtm