In prepping for our next release on Tika, we found ~1800 new exceptions caused by the added byte array (max picture length) check in HSLFSlideShowImpl.readPictures. I support the added check, but we should bump the value higher. I can't explain why this didn't turn up in earlier regression tests except that I must not have used the correct jars during the eval. :( The largest value in our regression corpus is 95,000,000 bytes. The limit is currently set at 10,000,000.
r1913004