Implement transform support for using declarations

[nicolo-ribaudo]

Q	A
Fixed Issues?
Patch: Bug Fix?
Major: Breaking Change?
Minor: New Feature?
Tests Added + Pass?	Yes
Documentation PR Link	babel/website#2792
Any Dependency Changes?
License	MIT

Missing:

• exec tests
• support for using at the top-level of modules
• async support

REPL demo

cc @rbuckton in case you track implementations of the proposal

[babel-bot]

Build successful! You can test your changes in the REPL here: https://babeljs.io/repl/build/54520/

[nicolo-ribaudo]

Ready for review!

[liuxingbaoyu]

Maybe use value != null?

Unrelated to this PR: Why do we always use === in helpers? == seems to be more compatible and shorter.🤔

[nicolo-ribaudo]

It's because of document.all (document.all == null, nut it's neither null nor undefined). We have an assumption option to use == instead of checking with both null and undefined (and maybe we could even consider enabling that assumption by default), but assumptions do not affect the code used in the helper.

[liuxingbaoyu]

I have another thought.
Is it possible for us to have the whole try...catch as a helper?
This greatly reduces code size and improves readability.
But I'm not sure what the performance impact would be, less code caching vs more likely JIT? Maybe it doesn't matter much. :)
Also this has some limitations, especially in variable scope.

function helperX(fn) {
  var _stack = [], _error, _hasError;
  try {
    fn(function (obj, isAwait) {
      babelHelpers.using(_stack, obj, isAwait);
    });
  } catch (_) {
    _error = _;
    _hasError = true;
  } finally {
    return babelHelpers.dispose(
      _stack,
      _error,
      _hasError,
      typeof SuppressedError !== "undefined" && SuppressedError
    );
  }
}

helperX(function (using) {
  const x = using(obj, 1);
  stmt;
  const y = using(obj, 1),
    z = using(obj, 1);
  doSomethingWith(x, y);
});

[nicolo-ribaudo]

Unfortunately that gets really complex whenever the block contains return/continue/break/var.

[JLHwung]

The StaticBlock does not have body as BlockStatement, we may need extra handling here.

[nicolo-ribaudo]

This AST shape feels like a bug, StaticBlocks should have had a BlockStatement as their body 🤦
Thanks for catching this!

[JLHwung]

The current design was borrowed from TSModuleBlock, to compress the node memory size, though in this case, a BlockStatement body will simplify the handling here.

[JLHwung]

Can we make it a helper? The minifier can't do much thing about global variables.

[nicolo-ribaudo]

The reason I put this here and not inline code is so that if you ran babel-plugin-polyfill-corejs3 in "usage-pure" mode it has a change of injecting the polyfill. If the helper relied on a global, it wouldn't work with "pure" polyfills.

[liuxingbaoyu]

Personally I prefer to create a variable _SuppressedError at the top level and then always reference it.
By the way, will we merge it soon? I want to try it #15633 (comment) :)

[nicolo-ribaudo]

I just realized that thanks to #15531, when using "usage-pure" users can get imports to the polyfilled version of the helper. We don't need to pass SuppressedError from the user code.

[nicolo-ribaudo]

This is a side effect of using babel-plugin-polyfill-corejs2 directly, but it's good: there is one less "proxy file" that does nothing other than re-exporting another file.

[nicolo-ribaudo]

Now this is polyfilled :)

[liuxingbaoyu]

Is it possible we're doing dead code elimination here?

[nicolo-ribaudo]

We minify helpers before injecting polyfills and not after, however any users bundling our helpers in a client-side bundle will usually re-minify them (all bundlers minify the end bundle, and not just the individual chunks) and this will be very easily eliminated :)

[JLHwung]

lgtm except minor spec compliance questions.

[JLHwung]

For reviewers: I added a REPL demo link in the PR description.