For some reasons, we interest only some particular slides (eg. first slide to generate thumbnail of document). We can load partially a (or some) slide to optimize performance and avoid failed parsing error on un-used slides.
HSLF reads the whole binary stream before parsing and the offsets are determined from the back of that stream (via CurrentUserAtom and PersistPtrHolder). As this is so unlikely that we change the API to open the slideshow with only a predefined set of slides, I close this issues as resolved-later ... If you can't generated thumbnails because of failed parsing, please open a new entry and attach those files please.