Description
Currently PDF parsing is done in sequential manner resulting in problems with stream parsing and skipping unused content. The solution is a conforming parser which first reads XREF tables and uses this information to only parse required objects and uses length information for stream parsing. A completely new implementation of such a parser is currently worked on in PDFBOX-1000. While this parser will be the long term solution a short term solution based on existing code would be desirable. A first incomplete solution was presented in PDFBOX-1104.
Starting from PDFBOX-1104 I have implemented an 'as much as possible' conforming parser, called 'non-sequential parser', which handles all PDF documents (even inlined, with object streams etc.). The parser can be used as a drop-in-replacement for PDFParser (subclass of PDFParser). It overwrites method parse and getPage method. The only restriction is currently the need to specify a file instead of an input stream. In order to efficiently read the file and use it with the existing object parsing code I developed a RandomAccessBufferedFileInputStream which allows InputStream operations in combination with seek operations and cached read data.
In order to use NonSequentialPDFParser small changes and additions on existing classes are needed. This includes changing some methods/fields from private to protected in PDFParser, add parsing of stream object information from XREF streams, store and get this information from XrefTrailerResolver (object ids are stored negated in order to distinguish them from offsets) and allow resetting offset in PushBackInputStream. All these changes do not change behavior of current parser. Another requirement is the long offset patch (PDFBOX-1196) which is excluded from the patch set provided here.
The provided parser currently works in a forceParsing=false mode resulting in an IOException if a parsing error occurs. In most cases this shouldn't be a problem since in my use cases exceptions typically occur trying to parse unused content or streams which with this new parser are no problems anymore. In my setup I use the new parser first and if a parsing error occurs, fall back to the sequential parser (a bit like Acrobat does it if XREF information is buggy):
try {
// ---- try first with (mostly) standard conform parsing
doc = PDDocument.loadNonSeq( PDF_FILE, raBuf );
handleDocument(doc);
} catch ( IOException ioe ) {
// ---- retry with sequential parser and force parsing
doc = PDDocument.load( new FileInputStream(PDF_FILE), raBuf, true );
handleDocument(doc);
}
For me this new parser works very well on large document collections and is a large step forward to parse all documents also accepted by common PDF tools. While its behavior is nearly 'conform' there is nevertheless a need for a clean 'real' conforming parser. For instance since the underlying object structure has no access to the parser it is necessary to first parse all objects before they can be used. This includes objects that might not be needed at all. Another normally not needed step is copying the content of a stream. Since we work on a file with random access there would be no need for it. However this parser should fill the hole until a full featured and clean conforming parser is available.
Attachments
Attachments
Issue Links
- relates to
-
PDFBOX-1238 Document stream used a undefined font resource
-
- Closed
-