Fix conflicted indeterminate state in asset importer

[rakkarage]

This pull request fixes an issue in the asset importer where the indeterminate state of conflicted items was not being properly resolved. It adds a new signal propagate_check_finished to the Tree class and a new method _fix_conflicted_indeterminate_state to the EditorAssetInstaller class to handle this.

Closes: #84021

Currently in Godot, in the asset import dialog (editor_asset_installer) it deals with indeterminate like this.

void EditorAssetInstaller::_check_propagated_to_item(Object *p_obj, int p_column) {
	TreeItem *affected_item = Object::cast_to<TreeItem>(p_obj);
	if (!affected_item) {
		return;
	}

	Dictionary item_meta = affected_item->get_metadata(0);
	bool is_conflict = item_meta.get("is_conflict", false);
	if (is_conflict) {
		affected_item->set_checked(0, false);
		affected_item->propagate_check(0, false);
	}
}

This is a problem because normally a indeterminate parent when clicked will become checked, but because some of the children are is_conflict they are unchecked again? which causes parent to go indeterminate again and repeat. So it prevents toggling from folder.

[KoBeWi]

Why is the new signal needed? Can't you just

tree->propagate_check();
_fix_conflicted_indeterminate_state()

?

[rakkarage]

Thanks. I tried it like that first but _check_propagated_to_item is already being called multiple times during a:

void TreeItem::propagate_check(int p_column, bool p_emit_signal) {
	bool ch = cells[p_column].checked;

	if (p_emit_signal) {
		tree->emit_signal(SNAME("check_propagated_to_item"), this, p_column);
	}
	_propagate_check_through_children(p_column, ch, p_emit_signal);
	_propagate_check_through_parents(p_column, p_emit_signal);
}

Which is how it gets to _check_propagated_to_item so modifying the tree while it was in the process of propagating a check did not seem to work even when limiting it to only calling when affected_item == root.

Seems much easier to let it finish propagating then fix. Idk.

Thanks.

[KoBeWi]

I mean you can fix the conflicts after calling propagate_check() in EditorAssetInstaller. The function is not async, so right after the call is when it has already finished. The signal is redundant.

[rakkarage]

I (dialog) don't call propagate_check tree does, when something gets checked or unchecked or added or...
The existing code just uses this handler to uncheck conflict when checked...
I need a new handler to know when tree has finished propagating checks and dialog has finished flipping them back off again. propagate_check_finished idk

void EditorAssetInstaller::_check_propagated_to_item(Object *p_obj, int p_column) {
	TreeItem *affected_item = Object::cast_to<TreeItem>(p_obj);
	if (!affected_item) {
		return;
	}

	Dictionary item_meta = affected_item->get_metadata(0);
	bool is_conflict = item_meta.get("is_conflict", false);
	if (is_conflict) {
		affected_item->set_checked(0, false);
		affected_item->propagate_check(0, false);
	}
}

I am probably just misunderstanding. Thanks, I will take another look at it.

[KoBeWi]

But I do see propagate_check() calls in EditorAssetInstaller. It's in item checked callback, but it's the installer calling it, not the Tree itself. And your new signals is emitted right at the end of this method, which is the same as if the method's caller did something after the call.

I (dialog) don't call propagate_check tree does

Where? Maybe I'm missing something.

[rakkarage]

I apologize for the confusion. :)
I was focused (tunnel vision) on the one propagate_check call inside _check_propagated_to_item which is signaled/called from TreeItem::propagate_check, but ya adding fix after all other propagate_check calls works.
Thanks.

[KoBeWi]

Shouldn't the conflict check in _check_propagated_to_item() be removed now?

Also the branch with conflicts appears as checked, while I think it should be indeterminate

Other than that it seems to work correctly.

EDIT:
I see the new behavior is intended. Not sure which behavior would be better tbh.

[akien-mga]

Thanks!