The hook should not be child of curve which it is deforming, it is causing cyclic dependency.

@Olson Burry (olson):
I cannot subscribe to that opinion.
It is necessary to move empty together when I move curve object.
Therefore it is easy to set relations.
This method works definitely in 2.49b.

2.49b blend file

I made these two files in the same procedure.
I made untitled249b.blend in 2.49b.
I made untitled274.blend in 2.74(sub5).
It follows that I expected it when I carry out untitled249b.blend in 2.49b or 2.74(sub5).
It is different from the result that I expected when I carry out untitled274.blend in 2.74(sub5).

The empty object moves together when I drag the curve object of the parent in untitled249b.blend.
(the result agrees even if I confirm it in 2.74)
However, the results are different when they do the same operation in untitled274.blend.
When I drag empty object, it is a result like the said article.

The curve object is relations of the parents of the empty object, but there is not the empty object concerning a parent of the curve object.
The empty object only maintains relative position relations for origin position of curve Object.

@Antonis Ryakiotakis (psy-fi):
When a position of two object is decided, they do not have the circulation dependence.
I think that you understand this.

However, it understands it that there is circulation dependence there when I consider it as object not position relations.

@Bastien Montagne (mont29):
I have already used that method as a substitute plan, but that method is not an intuitive method.

The present behavior understood the claim to be a limit.
I hope that blender comes to react to it sometime like 2.49b.
Thanks.

@perfection cat (sindra1961), It's something which is expected to be solved by depsgraph_refactor branch which i hope will go into master (as an option) this week.

@Sergey Sharybin (sergey):
really?
Thanks for informing me of that.