Re: Question on anonymous table objects

I have additional question regarding anomymous table objects.
In particular it is unclear how anonymous table objects should
interact with generated content. 

Consider for example the following markup:

<table>
<cell>Cell</cell>
<cell>Cell</cell>
</table>

with CSS

table {display:table;}
cell {display:table-cell;}
table:before, table:after {display:table-cell;content:'Cell';}

Since spec (12.1) says that
'The :before and :after pseudo-elements elements interact with
other boxes, such as run-in boxes, as if they were real elements inserted 
just inside their associated element'
it is logical to assume that this sample should be treated as

<table>
<anonymous-row>
<generated-cell>Cell</generated-cell>
<cell>Cell</cell>
<cell>Cell</cell>
<generated-cell>Cell</generated-cell>
</anonymous-row>
</table>

However in the same time spec (17.2.1) says that anonymous row must span
only consecutive siblings in the document tree:

"If the parent P of a 'table-cell' element T is not a 'table-row', an
object corresponding to a 'table-row' will be generated between P and T.
This object will span all consecutive 'table-cell' siblings (in the
document tree) of T."

Usually generated content is not considered to be part of document tree 
so what we get is:

<table>
<anonymous-row>
<generated-cell>Cell</generated-cell>
</anomymous-row>
<anonymous-row>
<cell>Cell</cell>
<cell>Cell</cell>
</anonymous-row>
<anonymous-row>
<generated-cell>Cell</generated-cell>
</anonymous-row>
</table>        

Now which interpretation is right? 
I think it is worth to clarify specs since different browsers handle this
sample in different way.