recovering from adding all native types

[sigmundch]

When we find code like js_util.getProperty during resolution, we assume that all native types may be used and enqueue them.

Would it be possible to treat native types again as part of the work-queue during codegen, so that if js_util.getProperty was tree-shaken, all those native types are also omitted?

[matanlurey]

To be clear, if @jmesserly's #35084 ends up "solving this", we don't consider this to be necessary. Just an FYI!

[rakudrama]

Even if js_util.getProperty is retained in the program, we should make it generic so that the given or inferred type parameter can constrain the type, something like:

getProperty(o, name) => JS('Object|Null', '#[#]', o, name);

-->

T getProperty<T>(o, name) => JS<T>('', '#[#]', o, name);

The trick is to defer the native impact of the inline JS code to when getProperty is used.
This could be achieved by summaries, i.e. make 'creates(T)' part of the summary of getProperty.
How T is bound by the uses of getProperty would constrain the native types retained.
This should work with inlining, since getProperty(o, 'x') is often inlined and compiles to o.x.

[matanlurey]

QQ @rakudrama or @sigmundch:

If I create the following "type":

@JS()
@anonymous
abstract class JSUnknown {}

... and I write something like:

JSUnknown p = getProperty(o, 'foo');

... will that be enough not to bring in native types?

[rakudrama]

@matanlurey Not today, it is 'game over' at JS('Object... in the current snippet.

If you know that o is a JavaScript object you might be able to do something like

@JS()
@anonymous
class JSUnknown {
  external JSUnknown get foo;
}
...
JSUnknown jsO = o;
JSUnknown p = jsO.foo;

That won't help if 'foo' is computed, but it might work for some uses.

[matanlurey]

In my particular case, I do know all of the static members, so I think I can do that. Thanks!