Guys thanks a lot for doing these posts. I also feel a big downgrade using the current Houdini manipulators coming from XSI and have thought about doing a similar thread myself.

Hopefuly SESI can pick some!

What I like about XSI is that things work with general rules on different contexts. So is the case of Transformations where some things apply to selected polygons as they apply to selected objects. For example: “Reset Translation” will put your object into 0,0,0, but if you apply it to a polygon selection it will put will move them to the 0,0,0 position of the object containing them (center the polygons).

The advantage of XSI's “simple rules that apply and combine everywhere” tendency is that the user doesn't need to learn a set of rules for each environment; by learning his way around a environment he is already learning his way around other enviorments and the muscular memory kicks in and “Oh this works here too, nice” kind of thing happens often. That along that some of the rules combine also results in many more outcomes, than a fixed set.
Cheers

Just another thing to point out. In XSI you can model two object at the same time. As far as I understand I can't be in edit the topology of meshes in 2 objects in the same object in Houdini. Not saying it's something major but here and there I've noticed it.

Cheers

I completely agree with what Probiner is mentionning.
However, I do think it is something major not being able to work on multiple geometries at the same time.
I've asked a few modellers about this and they all say they need to work on multiple geometries at the same time to be efficient. And for rigging it is also very important for weight editing and shape modelling.

I've logged an RFE about this : 73643

There is also a discussion about this here:
http://www.sidefx.com/forum/topic/43179/ [sidefx.com]

Cheers!

Just when I thought I finished pointing out all aspects of the Houdini's insufficiency in the transformations system, I've stumbled upon another one while trying to pose a rigged character: no transformation space memory.
For example I sometimes need to move a rig ctrl (or any other object) with the handle aligned to world, but rotate it in local/object mode.
The switching of these modes is cumbersome enough, so switching between align modes every time one switches the transformation type is incredibly debilitating.