This is a pre-production deployment of Warehouse, however changes made here WILL affect the production instance of PyPI.
Latest Version Dependencies status unknown Test status unknown Test coverage unknown
Project Description

1   Introduction

pdfrw is a Python library and utility that reads and writes PDF files:

  • Version 0.2 is tested and works on Python 2.6, 2.7, 3.3, and 3.4.
  • Operations include subsetting, merging, rotating, modifying metadata, etc.
  • The fastest pure Python PDF parser available
  • Has been used for years by a printer in pre-press production
  • Can be used with rst2pdf to faithfully reproduce vector images
  • Can be used either standalone, or in conjunction with reportlab to reuse existing PDFs in new ones
  • Permissively licensed

pdfrw will faithfully reproduce vector formats without rasterization, so the rst2pdf package has used pdfrw for PDF and SVG images by default since March 2010.

pdfrw can also be used in conjunction with reportlab, in order to re-use portions of existing PDFs in new PDFs created with reportlab.

2   Examples

The library comes with several examples that show operation both with and without reportlab.

2.1   All examples

The examples directory has a few scripts which use the library. Note that if these examples do not work with your PDF, you should try to use pdftk to uncompress and/or unencrypt them first.

  • will shrink pages down and place 4 of them on each output page.
  • shows an example of modifying metadata, without altering the structure of the PDF.
  • shows an example of creating a 2-up output suitable for printing and folding (e.g on tabloid size paper).
  • shows an example of concatenating multiple PDFs together.
  • will extract images and Form XObjects (embedded pages) from existing PDFs to make them easier to use and refer to from new PDFs (e.g. with reportlab or rst2pdf).
  • increases the size of a PDF so it can be printed as a poster.
  • Allows creation of 8.5 X 5.5” booklets by slicing 8.5 X 11” paper apart after printing.
  • Rotates all or selected pages in a PDF.
  • Creates a new PDF with only a subset of pages from the original.
  • Takes a 2-up PDF, and splits out pages.
  • Adds a watermark PDF image over or under all the pages of a PDF.
  • rl1/ Another 4up example, using reportlab canvas for output.
  • rl1/ Another booklet example, using reportlab canvas for output.
  • rl1/ Another subsetting example, using reportlab canvas for output.
  • rl1/ Another watermarking example, using reportlab canvas and generated output for the document. Contributed by user asannes.
  • rl2 Experimental code for parsing graphics. Needs work.

2.2   Notes on selected examples

2.2.1   Reorganizing pages and placing them two-up

A printer with a fancy printer and/or a full-up copy of Acrobat can easily turn your small PDF into a little booklet (for example, print 4 letter-sized pages on a single 11” x 17”).

But that assumes several things, including that the personnel know how to operate the hardware and software. lets you turn your PDF into a preformatted booklet, to give them fewer chances to mess it up.

2.2.2   Adding or modifying metadata

The example will accept multiple input files on the command line, concatenate them and output them to output.pdf, after adding some nonsensical metadata to the output PDF file.

The example alters a single metadata item in a PDF, and writes the result to a new PDF.

One difference is that, since cat is creating a new PDF structure, and alter is attempting to modify an existing PDF structure, the PDF produced by alter (and also by should be more faithful to the original (except for the desired changes).

For example, the navigation should be left intact, whereas with it will be stripped.

2.2.3   Rotating and doubling

If you ever want to print something that is like a small booklet, but needs to be spiral bound, you either have to do some fancy rearranging, or just waste half your paper.

The example program will, for example, make two side-by-side copies each page of of your PDF on a each output sheet.

But, every other page is flipped, so that you can print double-sided and the pages will line up properly and be pre-collated.

2.2.4   Graphics stream parsing proof of concept

The script shows a simple example of reading in a PDF, and using the module to try to write the same information out to a new PDF through a reportlab canvas. (If you know about reportlab, you know that if you can faithfully render a PDF to a reportlab canvas, you can do pretty much anything else with that PDF you want.) This kind of low level manipulation should be done only if you really need to. decodegraphics is really more than a proof of concept than anything else. For most cases, just use the Form XObject capability, as shown in the examples/rl1/ demo.

3   pdfrw philosophy

3.1   Core library

The philosophy of the library portion of pdfrw is to provide intuitive functions to read, manipulate, and write PDF files. There should be minimal leakage between abstraction layers, although getting useful work done makes “pure” functionality separation difficult.

A key concept supported by the library is the use of Form XObjects, which allow easy embedding of pieces of one PDF into another.

Addition of core support to the library is typically done carefully and thoughtfully, so as not to clutter it up with too many special cases.

There are a lot of incorrectly formatted PDFs floating around; support for these is added in some cases. The decision is often based on what acroread and okular do with the PDFs; if they can display them properly, then eventually pdfrw should, too, if it is not too difficult or costly.

Contributions are welcome; one user has contributed some decompression filters and the ability to process PDF 1.5 stream objects. Additional functionality that would obviously be useful includes additional decompression filters, the ability to process password-protected PDFs, and the ability to output linearized PDFs.

3.2   Examples

The philosophy of the examples is to provide small, easily-understood examples that showcase pdfrw functionality.

4   PDF files and Python

4.1   Introduction

In general, PDF files conceptually map quite well to Python. The major objects to think about are:

  • strings. Most things are strings. These also often decompose naturally into
  • lists of tokens. Tokens can be combined to create higher-level objects like
  • arrays and
  • dictionaries and
  • Contents streams (which can be more streams of tokens)

4.2   Difficulties

The apparent primary difficulty in mapping PDF files to Python is the PDF file concept of “indirect objects.” Indirect objects provide the efficiency of allowing a single piece of data to be referred to from more than one containing object, but probably more importantly, indirect objects provide a way to get around the chicken and egg problem of circular object references when mapping arbitrary data structures to files. To flatten out a circular reference, an indirect object is referred to instead of being directly included in another object. PDF files have a global mechanism for locating indirect objects, and they all have two reference numbers (a reference number and a “generation” number, in case you wanted to append to the PDF file rather than just rewriting the whole thing).

pdfrw automatically handles indirect references on reading in a PDF file. When pdfrw encounters an indirect PDF file object, the corresponding Python object it creates will have an ‘indirect’ attribute with a value of True. When writing a PDF file, if you have created arbitrary data, you just need to make sure that circular references are broken up by putting an attribute named ‘indirect’ which evaluates to True on at least one object in every cycle.

Another PDF file concept that doesn’t quite map to regular Python is a “stream”. Streams are dictionaries which each have an associated unformatted data block. pdfrw handles streams by placing a special attribute on a subclassed dictionary.

4.3   Usage Model

The usage model for pdfrw treats most objects as strings (it takes their string representation when writing them to a file). The two main exceptions are the PdfArray object and the PdfDict object.

PdfArray is a subclass of list with two special features. First, an ‘indirect’ attribute allows a PdfArray to be written out as an indirect PDF object. Second, pdfrw reads files lazily, so PdfArray knows about, and resolves references to other indirect objects on an as-needed basis.

PdfDict is a subclass of dict that also has an indirect attribute and lazy reference resolution as well. (And the subclassed IndirectPdfDict has indirect automatically set True).

But PdfDict also has an optional associated stream. The stream object defaults to None, but if you assign a stream to the dict, it will automatically set the PDF /Length attribute for the dictionary.

Finally, since PdfDict instances are indexed by PdfName objects (which always start with a /) and since most (all?) standard Adobe PdfName objects use names formatted like “/CamelCase”, it makes sense to allow access to dictionary elements via object attribute accesses as well as object index accesses. So usage of PdfDict objects is normally via attribute access, although non-standard names (though still with a leading slash) can be accessed via dictionary index lookup.

4.3.1   Reading PDFs

The PdfReader object is a subclass of PdfDict, which allows easy access to an entire document:

>>> from pdfrw import PdfReader
>>> x = PdfReader('source.pdf')
>>> x.keys()
['/Info', '/Size', '/Root']
>>> x.Info
{'/Producer': '(cairo 1.8.6 (',
 '/Creator': '(cairo 1.8.6 ('}
>>> x.Root.keys()
['/Type', '/Pages']

Info, Size, and Root are retrieved from the trailer of the PDF file.

In addition to the tree structure, pdfrw creates a special attribute named pages, that is a list of all the pages in the document. pdfrw creates the pages attribute as a simplification for the user, because the PDF format allows arbitrarily complicated nested dictionaries to describe the page order. Each entry in the pages list is the PdfDict object for one of the pages in the file, in order.

>>> len(x.pages)
>>> x.pages[0]
{'/Parent': {'/Kids': [{...}], '/Type': '/Pages', '/Count': '1'},
 '/Contents': {'/Length': '11260', '/Filter': None},
 '/Resources': ... (Lots more stuff snipped)
>>> x.pages[0].Contents
{'/Length': '11260', '/Filter': None}
>>> x.pages[0]
'q\n1 1 1 rg /a0 gs\n0 0 0 RG 0.657436
  w\n0 J\n0 j\n[] 0.0 d\n4 M q' ... (Lots more stuff snipped)

4.3.2   Writing PDFs

As you can see, it is quite easy to dig down into a PDF document. But what about when it’s time to write it out?

>>> from pdfrw import PdfWriter
>>> y = PdfWriter()
>>> y.addpage(x.pages[0])
>>> y.write('result.pdf')

That’s all it takes to create a new PDF. You may still need to read the Adobe PDF reference manual to figure out what needs to go into the PDF, but at least you don’t have to sweat actually building it and getting the file offsets right.

4.3.3   Manipulating PDFs in memory

For the most part, pdfrw tries to be agnostic about the contents of PDF files, and support them as containers, but to do useful work, something a little higher-level is required, so pdfrw works to understand a bit about the contents of the containers. For example:

  • PDF pages. pdfrw knows enough to find the pages in PDF files you read in, and to write a set of pages back out to a new PDF file.
  • Form XObjects. pdfrw can take any page or rectangle on a page, and convert it to a Form XObject, suitable for use inside another PDF file. It knows enough about these to perform scaling, rotation, and positioning.
  • reportlab objects. pdfrw can recursively create a set of reportlab objects from its internal object format. This allows, for example, Form XObjects to be used inside reportlab, so that you can reuse content from an existing PDF file when building a new PDF with reportlab.

There are several examples that demonstrate these features in the example code directory.

4.3.4   Missing features

Even as a pure PDF container library, pdfrw comes up a bit short. It does not currently support:

  • Most compression/decompression filters
  • encryption

pdftk is a wonderful command-line tool that can convert your PDFs to remove encryption and compression. However, in most cases, you can do a lot of useful work with PDFs without actually removing compression, because only certain elements inside PDFs are actually compressed.

5   Library internals

5.1   Introduction

pdfrw currently consists of 19 modules organized into a main package and one sub-package.

The __init.py__ module does the usual thing of importing a few major attributes from some of the submodules, and the module supports logging and exception generation.

5.2   PDF object model support

The objects sub-package contains one module for each of the internal representations of the kinds of basic objects that exist in a PDF file, with the objects/ module in that package simply gathering them up and making them available to the main pdfrw package.

One feature that all the PDF object classes have in common is the inclusion of an ‘indirect’ attribute. If ‘indirect’ exists and evaluates to True, then when the object is written out, it is written out as an indirect object. That is to say, it is addressable in the PDF file, and could be referenced by any number (including zero) of container objects. This indirect object capability saves space in PDF files by allowing objects such as fonts to be referenced from multiple pages, and also allows PDF files to contain internal circular references. This latter capability is used, for example, when each page object has a “parent” object in its dictionary.

5.2.1   Ordinary objects

The objects/ module contains the PdfObject class, which is a subclass of str, and is the catch-all object for any PDF file elements that are not explicitly represented by other objects, as described below.

5.2.2   Name objects

The objects/ module contains the PdfName singleton object, which will convert a string into a PDF name by prepending a slash. It can be used either by calling it or getting an attribute, e.g.:

PdfName.Rotate == PdfName('Rotate') == PdfObject('/Rotate')

In the example above, there is a slight difference between the objects returned from PdfName, and the object returned from PdfObject. The PdfName objects are actually objects of class “BasePdfName”. This is important, because only these may be used as keys in PdfDict objects.

5.2.3   String objects

The objects/ module contains the PdfString class, which is a subclass of str that is used to represent encoded strings in a PDF file. The class has encode and decode methods for the strings.

5.2.4   Array objects

The objects/ module contains the PdfArray class, which is a subclass of list that is used to represent arrays in a PDF file. A regular list could be used instead, but use of the PdfArray class allows for an indirect attribute to be set, and also allows for proxying of unresolved indirect objects (that haven’t been read in yet) in a manner that is transparent to pdfrw clients.

5.2.5   Dict objects

The objects/ module contains the PdfDict class, which is a subclass of dict that is used to represent dictionaries in a PDF file. A regular dict could be used instead, but the PdfDict class matches the requirements of PDF files more closely:

  • Transparent (from the library client’s viewpoint) proxying of unresolved indirect objects
  • Return of None for non-existent keys (like dict.get)
  • Mapping of attribute accesses to the dict itself (pdfdict.Foo == pdfdict[NameObject(‘Foo’)])
  • Automatic management of following stream and /Length attributes for content dictionaries
  • Indirect attribute
  • Other attributes may be set for private internal use of the library and/or its clients.
  • Support for searching parent dictionaries for PDF “inheritable” attributes.

If a PdfDict has an associated data stream in the PDF file, the stream is accessed via the ‘stream’ (all lower-case) attribute. Setting the stream attribute on the PdfDict will automatically set the /Length attribute as well. If that is not what is desired (for example if the the stream is compressed), then _stream (same name with an underscore) may be used to associate the stream with the PdfDict without setting the length.

To set private attributes (that will not be written out to a new PDF file) on a dictionary, use the ‘private’ attribute: = 1

Once the attribute is set, it may be accessed directly as an attribute of the dictionary:

foo =

Some attributes of PDF pages are “inheritable.” That is, they may belong to a parent dictionary (or a parent of a parent dictionary, etc.) The “inheritable” attribute allows for easy discovery of these:

mediabox = mypage.inheritable.MediaBox

5.2.6   Proxy objects

The objects/ module contains the PdfIndirect class, which is a non-transparent proxy object for PDF objects that have not yet been read in and resolved from a file. Although these are non-transparent inside the library, client code should never see one of these – they exist inside the PdfArray and PdfDict container types, but are resolved before being returned to a client of those types.

5.3   File reading, tokenization and parsing contains the PdfReader class, which can read a PDF file (or be passed a file object or already read string) and parse it. It uses the PdfTokens class in for low-level tokenization.

The PdfReader class does not, in general, parse into containers (e.g. inside the content streams). There is a proof of concept for doing that inside the examples/rl2 subdirectory, but that is slow and not well-developed, and not useful for most applications.

An instance of the PdfReader class is an instance of a PdfDict – the trailer dictionary of the PDF file, to be exact. It will have a private attribute set on it that is named ‘pages’ that is a list containing all the pages in the file.

When instantiating a PdfReader object, there are options available for decompressing all the objects in the file. pdfrw does not currently have very many options for decompression, so this is not all that useful, except in the specific case of compressed object streams.

Also, there are no options for decryption yet. If you have PDF files that are encrypted or heavily compressed, you may find that using another program like pdftk on them can make them readable by pdfrw.

In general, the objects are read from the file lazily, but this is not currently true with compressed object streams – all of these are decompressed and read in when the PdfReader is instantiated.

5.4   File output contains the PdfWriter class, which can create and output a PDF file.

There are a few options available when creating and using this class.

In the simplest case, an instance of PdfWriter is instantiated, and then pages are added to it from one or more source files (or created programmatically), and then the write method is called to dump the results out to a file.

If you have a source PDF and do not want to disturb the structure of it too badly, then you may pass its trailer directly to PdfWriter rather than letting PdfWriter construct one for you. There is an example of this ( in the examples directory.

5.5   Advanced features contains functions to build Form XObjects out of pages or rectangles on pages. These may be reused in new PDFs essentially as if they were images.

buildxobj is careful to cache any page used so that it only appears in the output once. provides the makerl function, which will translate pdfrw objects into a format which can be used with reportlab. It is normally used in conjunction with buildxobj, to be able to reuse parts of existing PDFs when using reportlab. builds on the foundation laid by buildxobj. It contains classes to create a new page (or overlay an existing page) using one or more rectangles from other pages. There are examples showing its use for watermarking, scaling, 4-up output, splitting each page in 2, etc. contains code that can find specific kinds of objects inside a PDF file. The example uses this module to create a new PDF that places each image and Form XObject from a source PDF onto its own page, e.g. for easy reuse with some of the other examples or with reportlab.

5.6   Miscellaneous and contains compression and decompression functions. Very few filters are currently supported, so an external tool like pdftk might be good if you require the ability to decompress (or, for that matter, decrypt) PDF files. contains code to help manage the differences between Python 2 and Python 3.

6   Testing

The tests associated with pdfrw require a large number of PDFs, which are not distributed with the library.

To run the tests:

  • Download or clone the full package from
  • cd into the tests directory, and then clone the package into a subdirectory (also named static_pdfs).
  • Now the tests may be run from that directory using unittest, or py.test, or nose.
  • travisci is used at github, and runs the tests with py.test

7   Other libraries

7.1   Pure Python

  • reportlab

    reportlab is must-have software if you want to programmatically generate arbitrary PDFs.

  • pyPdf

    pyPdf is, in some ways, very full-featured. It can do decompression and decryption and seems to know a lot about items inside at least some kinds of PDF files. In comparison, pdfrw knows less about specific PDF file features (such as metadata), but focuses on trying to have a more Pythonic API for mapping the PDF file container syntax to Python, and (IMO) has a simpler and better PDF file parser. The Form XObject capability of pdfrw means that, in many cases, it does not actually need to decompress objects – they can be left compressed.

  • pdftools

    pdftools feels large and I fell asleep trying to figure out how it all fit together, but many others have done useful things with it.

  • pagecatcher

    My understanding is that pagecatcher would have done exactly what I wanted when I built pdfrw. But I was on a zero budget, so I’ve never had the pleasure of experiencing pagecatcher. I do, however, use and like reportlab (open source, from the people who make pagecatcher) so I’m sure pagecatcher is great, better documented and much more full-featured than pdfrw.

  • pdfminer

    This looks like a useful, actively-developed program. It is quite large, but then, it is trying to actively comprehend a full PDF document. From the website:

    “PDFMiner is a suite of programs that help extracting and analyzing text data of PDF documents. Unlike other PDF-related tools, it allows to obtain the exact location of texts in a page, as well as other extra information such as font information or ruled lines. It includes a PDF converter that can transform PDF files into other text formats (such as HTML). It has an extensible PDF parser that can be used for other purposes instead of text analysis.”

7.2   non-pure-Python libraries

  • pyPoppler can read PDF files.
  • pycairo can write PDF files.
  • PyMuPDF high performance rendering of PDF, (Open)XPS, CBZ and EPUB

7.3   Other tools

  • pdftk is a wonderful command line tool for basic PDF manipulation. It complements pdfrw extremely well, supporting many operations such as decryption and decompression that pdfrw cannot do.
  • MuPDF is a free top performance PDF, (Open)XPS, CBZ and EPUB rendering library that also comes with some command line tools. One of those, mutool, has big overlaps with pdftk’s - except it is up to 10 times faster.

8   Release information


0.2 – Released 21 June, 2015. Supports Python 2.6, 2.7, 3.3, and 3.4.

  • Several bugs have been fixed
  • New regression test functionally tests core with dozens of PDFs, and also tests examples.
  • Core has been ported and tested on Python3 by round-tripping several difficult files and observing binary matching results across the different Python versions.
  • Still only minimal support for compression and no support for encryption or newer PDF features. (pdftk is useful to put PDFs in a form that pdfrw can use.)

0.1 – Released to PyPI in 2012. Supports Python 2.5 - 2.7

Release History

Release History


This version

History Node

TODO: Figure out how to actually get changelog content.

Changelog content for this version goes here.

Donec et mollis dolor. Praesent et diam eget libero egestas mattis sit amet vitae augue. Nam tincidunt congue enim, ut porta lorem lacinia consectetur. Donec ut libero sed arcu vehicula ultricies a non tortor. Lorem ipsum dolor sit amet, consectetur adipiscing elit.

Show More


History Node

TODO: Figure out how to actually get changelog content.

Changelog content for this version goes here.

Donec et mollis dolor. Praesent et diam eget libero egestas mattis sit amet vitae augue. Nam tincidunt congue enim, ut porta lorem lacinia consectetur. Donec ut libero sed arcu vehicula ultricies a non tortor. Lorem ipsum dolor sit amet, consectetur adipiscing elit.

Show More


History Node

TODO: Figure out how to actually get changelog content.

Changelog content for this version goes here.

Donec et mollis dolor. Praesent et diam eget libero egestas mattis sit amet vitae augue. Nam tincidunt congue enim, ut porta lorem lacinia consectetur. Donec ut libero sed arcu vehicula ultricies a non tortor. Lorem ipsum dolor sit amet, consectetur adipiscing elit.

Show More

Download Files

Download Files

TODO: Brief introduction on what you do with files - including link to relevant help section.

File Name & Checksum SHA256 Checksum Help Version File Type Upload Date
pdfrw-0.3-py2.py3-none-any.whl (58.0 kB) Copy SHA256 Checksum SHA256 py2.py3 Wheel Oct 19, 2016
pdfrw-0.3.tar.gz (71.6 kB) Copy SHA256 Checksum SHA256 Source Oct 19, 2016

Supported By

WebFaction WebFaction Technical Writing Elastic Elastic Search Pingdom Pingdom Monitoring Dyn Dyn DNS HPE HPE Development Sentry Sentry Error Logging CloudAMQP CloudAMQP RabbitMQ Heroku Heroku PaaS Kabu Creative Kabu Creative UX & Design Fastly Fastly CDN DigiCert DigiCert EV Certificate Rackspace Rackspace Cloud Servers DreamHost DreamHost Log Hosting