I think properly fixing this bug may involve doing something that
probably ought to be done anyway: to invert the relationship between
CellData and InsetTableCell in Tabular.
Right now, a CellData has an InsetTableCell, whereas one might have
thought it should be the other way around. It is how it is because the
inset used to be an InsetText, and of course that has what it has, and
in particular not a CellData. But now this could be inverted, so that
cell_vector, for example, would become a vector<InsetTableCell>, and the
members of the CellData class (really, struct) could just become members
of InsetTableCell.
In a way, this is easy, but let me ask a couple questions: (i) Why is
CellData::inset a boost::shared_ptr? What is it shared with? (ii) Why is
CellData::operator=() implemented the way it is? That looks very
strange, since we're modifying the rhs! Why do we need a non-default
implementation here at all? (iii) In InsetTabular::pasteClipboard(), we
have this:
shared_ptr<InsetTableCell> inset(
new InsetTableCell(*paste_tabular->cellInset(r1, c1)));
tabular.setCellInset(r2, c2, inset);
Could you do this instead?
ParagraphList const & paras =
*paste_tabular->cellInset(r1,c1).inset->paragraphs();
tabular.cellInset(r2,c2).inset->paragraphs() = paras;
Do we have to copy the whole inset? Also, why aren't we copying the
CellData?
rh
