> But I still did not understand the reasoning in the codes of ublas where (it seems)
> there is not a check for the completion of the fill of the elements of a matrix...
Hmmm, I don't understand your statement. Sorry. Anyway ...
> I wrote a matrix market interface to export matrices to MATLAB however I iterate
> over the nonzeros now, which is pretty fast,
I would also prefer to iterate over the nonzeros with appropiate iterators. That's
'cleaner' than working directly with the underlying arrays.
Are you doing the FE assembly process with a compressed matrix? If so, check
Gunter's hompepage for faster alternatives (general vector of vectors or so).