Vincent Hennebert wrote:
> Hi Simon,
> Thank you for giving a go at testing the branch. I could reproduce the
> hyphenation and flow_changing-ipd_4.xml problems. Regarding the stretch
> issue, could you send me a sample FO file? I suspect that this may be
> normal in the sense that the layout policy is slightly different when
> there is changing IPD: Instead of being full total-fit, it’s ‘n-pages
> total fit’ i.e., total-fit over the number of pages that have the same
> ipd. When the ipd changes, the best node /so far/ is selected, which may
> be different from the node that is part of the optimal solution, had the
> ipd remained the same.
I had a look at the file Simon sent me off-list, and this is indeed what
happens. Basically, the first page contains n lines with the ChangingIPD
version, and n + 2 with the Trunk. The n-lines version is the one that
is closest to the spaces’ natural lengths. In the n + 2 version spaces
are shrunk quite a bit. The n-lines version has less demerits than the
n + 2 one, so it’s the one that is chosen by the algorithm before
re-generating the Knuth elements for the next page with a different
width. But it appears that chosing the n + 2 version leads to less total
demerits, so this is why it is selected by the Trunk.
I committed yesterday fixes for the other bugs that Simon rose.
Thanks again for testing,