Declaration AST placement

[maximiliankaul]

This is just a reminder to a discussion between @oxisto and myself: 9d02f6c

[oxisto]

Currently, we place a declaration both on the AST property declarations directly as well as indirectly through the statements property and a DeclarationStatement. This could potentially be a bad idea, since the AST should be a tree, not a graph.

[konradweiss]

Yes, the ast should always be a tree. If it has more connections this can break several things, in the worst case lead to cycles in some algorithms

[oxisto]

Also for this discussion: Currently we model a a: Int or a simple a (the statement, not the expression) as an assignment with no rhs. This works sort-of and the assignment expression will also add a declaration, however the question would be whether we want to directly translate this into a declaration statement / variable declaration.

[oxisto]

This just came back to haunt me in #2004 and needs a solution :( The current implementation basically prevents us from having EOGs of nested functions, but if we do it correctly everything else fails because we assume things in the tests that are simply not "true", e.g., that a record declaration has fields (aka field declarations) (which is very seldom has, if not at all).