compiler: Fix bug in snapshotting

[hunger]

Do not take an ElementRc to move it into the Component's root_element.

@ogoffart spotted this! Thanks!

[ogoffart]

I think there is still a problem here that snapshot_element calls snapshot_component again for the enclosing component

I think you'll have to either

1. use Rc::new_cyclic, and pass the enclosing component in snapshot_element, or
2. keep the enclosing element empty, and set it later once the component is complete.

I think i'd go with option 1

[hunger]

I tried the new_cyclic approach. I have the problem that some of the elements are first discovered elsewhere than via the component they belong to (I had actually not expected that!). So I see elements created as ElementRcs... which is a problem if one of those is later discovered to be root element.

I have been playing with separating having a create method for components/elements (just creates a basically empty object (and asserts if the object exists already), trying to create elements/components referenced in the original only), as well as a use method (asserts if some component/element is used that has not been created yet) and a final snapshot step that copies the rest of the data (properties and such that use elements and such).

I get lots of asserts triggered with that approach :-/

[hunger]

I think it works now. After way too many stupid bugs and some surprising behavior in how and when elements are created...

[hunger]

/me sighs. Left in a variable I had added to debug things.

[hunger]

I used the stable order of struct fields (#5561) to debug-print the components and elements I snapshot and to compare that output to the same from the original components and elements. Nothing asserts anymore, they all produce the same debug strings.

That's a step into the right direction... BUT I see the base_type of some elements as <error> -- in both the source and the target! I m poking at that problem now.

[hunger]

I Hate this part, hard-coding some "special" components!

[hunger]

Funny... I find elements with a base_type of Error in the master branch as well when taking a snapshot there. They do show up in the property editor with the expected types though! I do the Error in the Property editors type display as well when using this PR :-(