e.g. (today): capital letters means "root", other are "leaves"

p1, p2, p3 (link)> o1[P3+p1+p2]

or

and

p4, p5, p6 (link)> o2[P6+p4+p5] or

o1, o2 (link)> o3[O2+o1] == o3[P6+p1+p2+p3+p4+p5]

or

You can see how, in the (second) linking process, the "P3 = root" information is lost to "flatten" the hierarchy into a single level. Thus, when unlinking, no one knows about the ancient "root" (P3) and all the prims are "leaves" (unlinking o3 leads to p1...p6 normal (unlinked) prims, where each prim is "root for itself").

The new way of linking

If a multilevel hierarchy can be preserved, linking o1 to o2 is the same to attach a tree to another tree:

o3[[P6+p4+p5]+[P1+p2+p3]] (yes: "lists into lists")

or

How the prims/objects will be linked depends to the objects/prims presents at the linking time, so linking o1 to p3, p4 and p6 will gives o2[P6+p4+p5+o1[P3+p1+p2]] and linking o1[P3+p1+p2] to o2[P6+p4+p5] leads to o3[o2[P6+p4+p5]+o1[P3+p1+p2]].

So, for instance, you can move a complex built engine, with e.g. 50 prims, apart of the whole car frame, which can be built by 200 prims, without unlinking all the 250 prims car (or, better, you can select the engine to move, as a "object", without have to unlink the car or move 50 pieces one at the time).

LSL POV

From the LSL point of view we'll have "main roots" and "sub roots": a main root is the root we already have (root of all object) and a sub-root is a root prim which is root for it's sub-prims but a leaf for the upper-level root (which may not be the top-level-root). Thus the messages from script to linked prim/scripts may complain of the multi-level hierarchy: "message to sub-hierarchy N" means "to the N.th root" (N.th root can then route the message to a local leaf if needed, the same way actual roots do).

general POV

I don't know if is a real need to have LSL multilevel hierarchy (even if, from the architectural POV, is a cleaner behaviour), but I'm sure that hierarchies are a major need from the building POV.

A real case: i built a not too complex cape with an "object" built up of really small particulars. Now: if I want to extract this object to change some of the small particulars it's impossible to me to move that "object" 'cause of, after linking, it's no more an "object": all it's prims are now parts of the whole cape... a real pain in the butt to edit old things. :-/

Progressive linking may help linking objects/prims at the "end" of flexy object also.
(now the linking is spatially referred to the center of mass of the whole object: only flexies have an end-attach-point to the "reference" but no prim can be linked to them)

Mercia

I think it would increase the possibilities in scripting too.
I think of the objects containing of more then one prim and can be rotated right unlinked.
But after linking the rotation don't work anymore because the object with root<->childs is gone. All prims of the objects are childs of the new root prim.
VOTE FOR IT!

http://jira.secondlife.com/browse/SVC-2631

Not they think to handle this request, nor they like to comment here (e.g. "don't think we'll do it" or similar).

The only way to change the things, which I seen "productive" for me, is to change the "environment" where to operate: e.g. in OpenSim we do HAVE c-like comments since a year or more... and I think we'll have this kind of things (hierarchical building) early or later.
Now I'll stop my rants on LLs and go to the world where a "resident" is listened (as a matter of fact I'm not only a "resident", there, but the owner of 32 sims nowadays ).

c ya