Andre Poenitz wrote: > > But maybe time for some comments on the last of todays patches (the one > I won't apply). > > This is changing ParagraphList to a std::vector<Paragraph>.
Man, where do you get all that energy? ;-) > After playing around a bit I am convinced this is the way to go. First > of all, there is really no performance problem withit whatsoever. E.g. > splitting the first paragraph of the UserGuide.lyx leads to no > recognizable delay although it moves all the contents of the document. > > You can check this yourself by changing the typedef and commenting out > the 'splice' line in CutAndPaste.C which seemingly is the only time we > use a std::list specific feature. I would expect big time improvements once the getPar stuff dies. This doesn't necesarily is an argument in favour. (rather against paroffsets everywhere) > Secondly, it gives us a trivial 'stable iterator' implementation. In > fact, I even think we get drop the DocumentIterator::inset_ 'cache' as > recreating this is O(nesting level) with the vector<> based par list, > i.e. effectively O(1) on real world documents. Not counting the insetlist stuff yes, but I agree. > You will notice, however, that after changing the typedef things go > foobar for 'complex' operations. So we seem to rely on the fact that > list iterators are 'semi-stable' (compared to vector iterators) under > deletion/insertion. > > The easy way out is of course, to use offsets everywhere, and this is > what patch 8 does. I am not done with it. Although most changes are > purely mechanical, some are not. > > I also think this it is a good idea to let everybody getting used to the > idea of a vector based paragraph list. Sigh. > PS: A rule of thumb I came up with during the 'playing around': O(n) > is acceptable anywhere anytime as response to a user interaction. O(n^2) > is not. With the vector based solution we never get worse then O(n), > for list based this is tricky (if possible at all). I don't know if I agree with you here. This is like this because of the getPar stuff, that is really needed only when moving paragraphs around (i.e. c&p, undo) but is used everywhere. In any case, I'm not really against the move. If only we could get more people fixing cvs, it wouldn't be unwise. Right now, I don't know. Alfredo
