> I would suggest one more: > - remove eqn. numbers from previews; they tend to be wrong anyway
Isn't this too specific as a solution? I think there are several other things wrong we'll probably never know. > Yet these are just workarounds. I try to catch attention of other > developers on how best to tackle this. It seems that ideally, the > previews would be broken down into several .tex files, each to be > processed by the intended backend. venom? The things works like this: call latex on the .tex file (containing all the IP insets) and create a unique .dvi file. Then we find all the pages (IP insets) containing postscript literals (which dvipng can't handle) and create a new tex file with those pages and give it to the legacy method (latex+dvips+gs). Maybe we can add a third fallback level based on pdflatex. I think it would be much better to manage all the IP generation in a unique place (that is Python scripting). This would work with all the possible input, even if the last cases would be a little slow. venom00
