Method assumed that not much attribution happens below data and started from
the end of attribution, in the case of not much attribution it worked fine.
For the case of many differently formatted areas with a sufficiently large
area of visibly equal attribution near data end it was a bottle neck looping
over unnecessarily many comparisons. Start at data end instead. For the case
of not much attribution it doesn't really matter, and for the case of no
sufficiently large area below data end it doesn't matter at all and compares
the same number of entries. The drawback would be a large area near
attribution end with many small areas between data end and the large area.

Observed with test case of fdo#46160