Any PDF which had wrong startxref was parsed with warning in PdfBox 1.8 thanks to self healing mechanism.
In version 2.0 the COSParser.parseXref (called from PDFParser) tries to seek to the startxref position.
If the position is wrong, but within the file size, the PDF is parsed with warning.
But when the startxref is over the file size, the parsing ended with exception.
We could either test if startxref is not over source.length(), or catch the exception.