Marco Patzer
Looking at the diffs, many of the differences are really tiny, but differences exist nevertheless. It will take me some time to sort out which ones are just tiny space shifting, and which ones give really different output.
That's why I use a header for each test file. You can specify “sensitivity=low|medium|high”, this sets a predefined tolerance for the page difference. I don't see any other option. There is no single value which is sensitive enough, for instance to detect if a footnote is subscripted or not and at the same time tolerant enough not to always mark a fifty page document as “failed”.
I'm not sure that exists a “fuzzy” algorithm for the tolerance. From my testing, a minimal shift, say, to the right, on a full page will generate a lot of noise, because each glyph will be “doubled”. I'm generating, with the visual diffs, a script to launch two PDF viewers and eventually promote the generated to the new reference. I guess there is no shortcuts to the manual ispection of the diffs. -- Marco