a) Why PrintAnnotationInfo() and no PrintSelf() ?
It is not VTK compliant: http://www.cmake.org/Wiki/VTK_Coding_Standards
There are even some tests in VTK (tcl based) that ensure that a PrintSelf
exists in each vtkObject subclass.

b) vtkMRMLAnnotationControlPointsNode::WriteCLI
It doesn't seem to be used anywhere in Slicer. Is it obsolete ?
Should it be removed from code? (as well as in vtkMRMLDisplayableNode)

c) vtkMRMLAnnotationNode::CreatePolyData() is not virtual
However it is overloaded in subclassess (e.g. vtkMRMLAnnotationControlPointsNode)

d) vtkMRMLAnnotationNode::Initialize() disables modified event unnecessarily
The node is not yet added into the scene, so preventing events from being
fired is useless.
(I have some local work (few month old) that attempts to remove them)

e) headers (with copyright and license) are missing from most annotation nodes

f) Are Backups still in use ?
It is not "MRML" style. It should either be extended to all MRML nodes or
not be in use at all. I personally think it is not the role of the node to
keep a copy of itself, but more the role of the application (eventually the
logic).
0002608 could be a way to move the "backup" mechanism to "Copy"

g) Having an icon per node type is great
But it should be a MRML wide feature. It should probably be done in
qMRMLSceneModel. E.g. while you register a node to the scene, you register
an icon into a static qMRMLSceneModel.

h) Expose qMRMLSceneAnnotationModel for extensions/modules access
Some modules might want to reuse the model.

• remove Qt/GUI dependency from Logic CMakeLists

i) Support writing ROI nodes with no text display node (see 0002620)
It could be a bit more robust.

j) Move vtkMRMLAnnotation*Node::Initialize() into vtkMRMLAnnotationLogic ?
Initialize() is not standard in MRML. Either all the nodes use it, or none of them.
Generally, MRML nodes don't create other nodes (with the exception of the CreateXXXNode methods which solely create a node , they don't add them into scene nor observe them automatically))
Maybe some part could be moved into vtkMRMLAnnotationLogic or in CreateXXXNode ?

k) Add context menu entry in annotation tree view to create an annotation
On right click, it would be handy to have a menu that allows the user to create(place) a new annotation in the clicked hierarchy.
Some kind of a shortcut to the "place" toolbar

l) Make scene annotation model columns customizable
Following the other qMRMLSceneXXXModels style, each column can be moved.
If -1, the column gets hidden.

m) Export MRML displayable managers
Modules who want to implement their own annotation displayable manager should have access to the Annotation displayable manager.

n) vtkMRMLAnnotationNode::Initialize(vtkMRMLScene) fails if scene has not added prior
vtkMRMLAnnotationNode::Initialize() calls CreateAnnotationTextDisplayNode() which returns because no scene is set.
Maybe Initialize() should be split and add a utility function that creates all the display nodes:
void CreateAnnotationDisplayNodes(vtkMRMLScene scene){
vtkMRMLDisplayNode* displayNode = this->CreateXXXDisplayNode();
scene->AddNode(displayNode);
this->AddAndObserveDisplayNodeID(displayNode->GetID());
this->Superclass::CreateAnnotationDisplayNodes(scene);
}
-> CreateXXX() methods should return a node that is not observed yet: https://github.com/Slicer/Slicer/blob/master/Libs/MRML/Core/Documentation/MRMLCore.dox

o) should vtkMRMLAnnotationDisplayNode be visible by default?
For the same reason vtkMRMLDisplayNode and vtkMRMLAnnotationNode are visible by default,
it should be the same for vtkMRMLAnnotationDisplayNode. No ?
vtkMRMLAnnotationNode::Visible could be renamed into Visibility to be consistent with vtkMRMLDisplayNode.

p) qSlicerMouseModeToolBar should extract name from XML tag and not by a custom node class name method.
When overloading annotations, the class name might not follow the format vtkMRMLAnnotationXYZNode

q) Add option to draw annotations on top of existing objects in scene
Sometimes you might want to have the annotations "always on top" and not be obfuscated by the volumes/models. I believe it could be fixed by either disabling depth test or by rendering the widgets in a separate layer.

r) Consolidate Click Counter with the seed widget vector in the annotations manager?
Isn't the click counter keeping track of the number of seeds ? If it is the same, the information is probably redundant.