Text extraction ignores bookmarks

As reported by Noam S. on the user mailing list:

My problem is that when trying to getText(doc) form a certain section of the pdf using setStartBookmark(item) and setEndBookmark(item) I get all the text rather than just the text from the specified section.

WhiIe trying to resolve this I realized that the writeText(doc, outputStream) method always calls resetEngine() method. That will reset all the parameters and delete the bookmarks I set.

The two lines that reset the bookmarks were added to resetEngine in PDFBOX-1808 in [ https://svn.apache.org/r1553175 ] in an attempt to save some memory.

Another weird segment can be found in the trunk:

I also found another weird piece of code in the trunk, which would mean that text extraction would fail if start and end bookmarks are identical:

        if (startPage != null && endPage != null &&
            startBookmark.getCOSObject() == endBookmark.getCOSObject())
        {
            // this is a special case where both the start and end bookmark
            // are the same but point to nothing.  In this case
            // we will not extract any text.
            startBookmarkPageNumber = 0;
            endBookmarkPageNumber = 0;
        }

earlier, that segment was:

       if( startBookmarkPageNumber == -1 && startBookmark != null &&
                endBookmarkPageNumber == -1 && endBookmark != null &&
                startBookmark.getCOSObject() == endBookmark.getCOSObject() )
        {
            //this is a special case where both the start and end bookmark
            //are the same but point to nothing.  In this case
            //we will not extract any text.
            startBookmarkPageNumber = 0;
            endBookmarkPageNumber = 0;
        }

which makes more sense. The change was made last year in rev [ https://svn.apache.org/r1634252 ] as part of the pagetree refactoring.

I am writing a test to prevent this from breaking in the future.