pdfsyntax 0.0.6

A Python PDF parsing library and tool built on top to browse the internal structure of a PDF file

PDFSyntax

A Python PDF parsing library and tool built on top to browse the internal structure of a PDF file

Introduction

The project is focused on chapter 7 ("Syntax") of the Portable Document Format (PDF) Specification.

PDFSyntax is lightweight (no dependencies) and written from scratch in pure Python.

1. CLI: It started as a command-line interface to inspect the internal structure of a PDF file: the PDF source is pretty-printed and augmented with hyperlinks in a static HTML file.
2. API: Now the internal functions are being exposed as a tooklit for PDF read/write operations.

It is mostly made of simple functions working on built-in types and named tuples. Shallow copying of the Doc object structure performed by pure functions offers some kind of - experimental - immutability.

PDFSyntax favors non-destructive edits allowed by the PDF Specification: by default incremental updates are added at the end of the original file.

Project status

WORK IN PROGRESS! This is ALPHA quality software.

CLI

Features

The generated HTML looks like the raw PDF file with the following additions:

• Pretty-print dictionary object
• Extract an object contained in an object stream and insert it in the flow like a regular object
• Decompress stream and display a small part of it
• Turn indirect object reference into hyperlink
• Turn offset reference (for example a /Prev entry) into hyperlink
• Put some color on key names (for example /Type)
• Display offset of an object

Screenshot

Usage

Generate the HTML file and open it in your browser:

python3 -m pdfsyntax inspect file.pdf > inspection.html

API

Usage

Most functions are pure and are exposed both as basic functions and as instance methods of a Doc object: in the function signatures found in the following sections, a doc first argument can read as self. For example both samples are equivalent:

>>> #Function pattern
>>> from pdfsyntax import read, metadata
>>> doc = read("samples/simple_text_string.pdf")
>>> m = metadata(doc)

>>> #Method pattern
>>> import pdfsyntax as pdf
>>> doc = pdf.read("samples/simple_text_string.pdf")
>>> m = doc.metadata()

File information

structure and metadata are functions showing general information about the document.

>>> #File structure
>>> structure(doc)
{'Version': '1.4', 'Pages': 1, 'Revisions': 1, 'Encrypted': False, 'Paper of 1st page': '215x279mm or 8.5x11.0in (US Letter)'}

>>> #File metadata
>>> metadata(doc)
{'Title': None, 'Author': None, 'Subject': None, 'Keywords': None, 'Creator': None, 'Producer': None, 'CreationDate': None, 'ModDate': None}

Low-level access to object tree

trailer and catalog give access to the starting point of the object tree.

>>> #Access to document trailer
>>> doc.trailer()
{'/Root': 1j, '/Size': 8, '/Prev': 603}

>>> #Access to document catalog as specified in the /Root entry of the trailer
>>> doc.catalog()
{'/Pages': 3j, '/Outlines': 2j, '/Type': '/Catalog'}

1j is a complex number (!) representing indirect reference 1 0 R. Why? Because the approach is to map PDF object types to Python basic built-in types as much as possible, and it is a concise way to show both the object number (as the imaginary part) and the generation number (as the real part). Moreover the generation is very often equal to zero, so the real part is not shown. You may think of the j as a "jump" to another object :)

get_object gives direct access to indirect objects.

>>> #Access to document catalog, given that the trailer redirects to 1j for root
>>> #(equivalent to catalog fonction)
>>> doc.get_object(1j)
{'/Pages': 3j, '/Outlines': 2j, '/Type': '/Catalog'}

Pages

Page index is a tree structure where attributes can be inherited from parent nodes. For convenience flat_page_tree returns an ordered list of document pages and specifies inherited attributes that should apply to each page.

>>> #Each item of the list is a tuple with the page object reference and its inherited attributes
>>> doc = pdf.read("samples/simple_text_string.pdf")
>>> pdf.flat_page_tree(doc)
[(4j, {})]
>>> #(In this example, nothing is inherited from upper nodes)

The page function goes further by merging inherited attributes with local attributes of each page and giving the result in a list.

>>> #Equivalent list with computed page attribues
>>> pdf.pages(doc)
[{'/Resources': {'/Font': {'/F1': 7j}, '/ProcSet': 6j},
  '/Contents': 5j,
  '/MediaBox': [0, 0, 612, 792],
  '/Parent': 3j,
  '/Type': '/Page'}]

Incremental updates

PDFSyntax tracks document incremental updates made possible by appending new or updated objects at the end of an original PDF file (and the matching XREF section). The Revisions entry of the structure function result, if greater than 1, indicates that incremental updates have been appended. By default, a newly opened document by PDFSyntax is ready to write modifications in the next revision. The rewind function rolls back to the previous revision.

>>> import pdfsyntax as pdf
>>> doc = pdf.read("samples/add_text_annotation.pdf")
>>> doc.structure()
{'Version': '1.4', 'Pages': 1, 'Revisions': 2, 'Encrypted': False, 'Paper of 1st page': '215x279mm or 8.5x11.0in (US Letter)'}

>>> #This file contains 2 revisions and PDFSyntax has initialized the doc object for a future revision 3

>>> doc.get_object(4j)
{'/Annots': 8j, '/Resources': {'/Font': {'/F1': 7j}, '/ProcSet': 6j}, '/Contents': 5j, '/MediaBox': [0, 0, 612, 792], '/Parent': 3j, '/Type': '/Page'}

>>> #In its current state, the page (object 4) contains an annotation
>>> #Let's rewind to revision 1

>>> doc = doc.rewind() # to revision 2
>>> doc = doc.rewind() # to revision 1

>>> doc.get_object(4j)
{'/Resources': {'/Font': {'/F1': 7j}, '/ProcSet': 6j}, '/Contents': 5j, '/MediaBox': [0, 0, 612, 792], '/Parent': 3j, '/Type': '/Page'}

>>> #The annotation was not present in the initial revision of the file

Transformation

rotate turns pages relatively to their current position by multiples of 90 degrees clockwise. NB: It takes into account the inherited attributes from the page hierarchy.

>>> #Default rotation applies 90 degrees to all pages
>>> doc90 = rotate(doc)

>>> #Apply 180 degrees to first two page
>>> doc180 = doc.rotate(180, [1, 2])

TO BE CONTINUED