Current component is not cleared after flush

[rdb]

Describe the bug
The current component is not being cleared at the end of the flush call. This means that lifecycle callbacks, like onMount, onDestroy, etc. are made available outside of Svelte code, with unpredictable effects (effectively operating on a random component, possibly one that is already destroyed).

This looks like a simple oversight. It looks like a call to set_current_component(null) is missing after this loop:

svelte/src/runtime/internal/scheduler.ts

Lines 43 to 48 in ec3589e

	for (let i = 0; i < dirty_components.length; i += 1) {
	const component = dirty_components[i];
	set_current_component(component);
	update(component.$$);
	}

To Reproduce

<script>
  import { get_current_component } from 'svelte/internal';

  let a = 0;

  // Do something to trigger a flush after half a second
  setTimeout(() => {
    a = 1;
  }, 500);

  // Print current component after 1 second - should give error
  // Instead, it prints the last-updated component
  setTimeout(() => {
    console.log(get_current_component());
  }, 1000);
</script>

<p>{a}</p>

Expected behavior
I expected an exception to be thrown after a second (which it indeed does if you comment out the
first setTimeout). Instead, it logs the most recently-updated component, which is not at all guaranteed to be the current one.

Severity
I'm currently stuck on this with a package I'm maintaining. I have found a workaround (importing tick from svelte-internal and then monkey-patching tick().then to reset the component after flush) but it is extremely hacky and I would not want to put this in a hosted library.

Additional context
I'm maintaining a helper package which enables closer integration between Svelte and Meteor. It registers an onDestroy callback to handle cleanup of certain resources, but the function in question needs to be able to work both inside and outside Svelte components, hence the need to be able to trust that the current component is cleared.

[Conduitry]

Duplicate of #4259. The fix does not seem to be that simple though, as seen in the failing tests in #4864.

[rdb]

I did see #4259, but it is specifically about onMount, and this issue is much more generic than that. If you are going to leave this closed, perhaps you can edit #4259 to include the scope that this bug
report covers?

If set_current_component(null) after the for loop introduces complications, this issue could still be trivially fixed by putting it at the end of the flush call, which would fix this issue while leaving the onMount bug in #4259 unaddressed.

[Conduitry]

That might be reasonable. Reopening.

[rdb]

To anyone else waiting for #4909 to be merged, it is possible to work around this using the following code:

import { set_current_component, tick } from 'svelte/internal';

const promise = tick();
const oldThen = promise.then;
promise.then = function (fn) {
  oldThen.call(promise, () => {
    fn();
    set_current_component(null);
  });
};

[Conduitry]

That component now throws an exception at runtime in 3.25.0.