IPython 0.12 contains several major new features, as well as a large amount of bug and regression fixes. The 0.11 release brought with it a lot of new functionality and major refactorings of the codebase; by and large this has proven to be a success as the number of contributions to the project has increased dramatically, proving that the code is now much more approachable. But in the refactoring inevitably some bugs were introduced, and we have also squashed many of those as well as recovered some functionality that had been temporarily disabled due to the API changes.
The following major new features appear in this version.
A powerful new interface puts IPython in your browser. You can start it with the command ipython notebook:
The new IPython notebook showing text, mathematical expressions in LaTeX, code, results and embedded figures created with Matplotlib.
This new interface maintains all the features of IPython you are used to, as it is a new client that communicates with the same IPython kernels used by the terminal and Qt console. But the web notebook provides for a different workflow where you can integrate, along with code execution, also text, mathematical expressions, graphics, video, and virtually any content that a modern browser is capable of displaying.
You can save your work sessions as documents that retain all these elements and which can be version controlled, emailed to colleagues or saved as HTML or PDF files for printing or publishing statically on the web. The internal storage format is a JSON file that can be easily manipulated for manual exporting to other formats.
This Notebook is a major milestone for IPython, as for years we have tried to build this kind of system. We were inspired originally by the excellent implementation in Mathematica, we made a number of attempts using older technologies in earlier Summer of Code projects in 2005 (both students and Robert Kern developed early prototypes), and in recent years we have seen the excellent implementation offered by the Sage <http://sagemath.org> system. But we continued to work on something that would be consistent with the rest of IPython’s design, and it is clear now that the effort was worth it: based on the ZeroMQ communications architecture introduced in version 0.11, the notebook can now retain 100% of the features of the real IPython. But it can also provide the rich media support and high quality Javascript libraries that were not available in browsers even one or two years ago (such as high-quality mathematical rendering or built-in video).
The notebook has too many useful and important features to describe in these release notes; our documentation now contains a directory called examples/notebooks with several notebooks that illustrate various aspects of the system. You should start by reading those named 00_notebook_tour.ipynb and 01_notebook_introduction.ipynb first, and then can proceed to read the others in any order you want.
To start the notebook server, go to a directory containing the notebooks you want to open (or where you want to create new ones) and type:
ipython notebook
You can see all the relevant options with:
ipython notebook --help
ipython notebook --help-all # even more
and just like the Qt console, you can start the notebook server with pylab support by using:
ipython notebook --pylab
for floating matplotlib windows or:
ipython notebook --pylab inline
for plotting support with automatically inlined figures. Note that it is now possible also to activate pylab support at runtime via %pylab, so you do not need to make this decision when starting the server.
See the Notebook docs for technical details.
Based on the same architecture as the notebook and the Qt console, we also have now a terminal-based console that can connect to an external IPython kernel (the same kernels used by the Qt console or the notebook, in fact). While this client behaves almost identically to the usual IPython terminal application, this capability can be very useful to attach an interactive console to an existing kernel that was started externally. It lets you use the interactive %debug facilities in a notebook, for example (the web browser can’t interact directly with the debugger) or debug a third-party code where you may have embedded an IPython kernel.
This is also something that we have wanted for a long time, and which is a culmination (as a team effort) of the work started last year during the 2010 Google Summer of Code project.
The QtConsole now supports starting multiple kernels in tabs, and has a menubar, so it looks and behaves more like a real application. Keyboard enthusiasts can disable the menubar with ctrl-shift-M (PR #887).
The improved Qt console for IPython, now with tabs to control multiple kernels and full menu support.
IPython can now be installed from a single codebase on Python 2 and Python 3. The installation process for Python 3 automatically runs 2to3. The same ‘default’ profile is now used for Python 2 and 3 (the previous version had a separate ‘python3’ profile).
The ipython kernel subcommand has been added, to allow starting a standalone kernel, that can be used with various frontends. You can then later connect a Qt console or a terminal console to this kernel by typing e.g.:
ipython qtconsole --existing
if it’s the only one running, or by passing explicitly the connection parameters (printed by the kernel at startup).
The terminal interface to IPython now runs under PyPy. We will continue to monitor PyPy’s progress, and hopefully before long at least we’ll be able to also run the notebook. The Qt console may take longer, as Qt is a very complex set of bindings to a huge C++ library, and that is currently the area where PyPy still lags most behind. But for everyday interactive use at the terminal, with this release and PyPy 1.7, things seem to work quite well from our admittedly limited testing.
In this cycle, we have closed over 500 issues, but a few major ones merit special mention:
PR #648). Note that pickling saves a reference to the class definition, so unpickling the instances will only work where the class has been defined.
PromptManager, which has traits for in_template, in2_template (the ...: continuation prompt), out_template and rewrite_template. This uses Python’s string formatting system, so you can use {time} and {cwd}, although we have preserved the abbreviations from previous versions, e.g. \# (prompt number) and \w (working directory). For the list of available fields, refer to the source of IPython/core/prompts.py.
The class inheritance of the Launchers in IPython.parallel.apps.launcher used by ipcluster has changed, so that trait names are more consistent across batch systems. This may require a few renames in your config files, if you customized the command-line args for launching controllers and engines. The configurable names have also been changed to be clearer that they point to class names, and can now be specified by name only, rather than requiring the full import path of each class, e.g.:
IPClusterEngines.engine_launcher = 'IPython.parallel.apps.launcher.MPIExecEngineSetLauncher'
IPClusterStart.controller_launcher = 'IPython.parallel.apps.launcher.SSHControllerLauncher'
would now be specified as:
IPClusterEngines.engine_launcher_class = 'MPI'
IPClusterStart.controller_launcher_class = 'SSH'
The full path will still work, and is necessary for using custom launchers not in IPython’s launcher module.
Further, MPIExec launcher names are now prefixed with just MPI, to better match other batch launchers, and be generally more intuitive. The MPIExec names are deprecated, but continue to work.
For embedding a shell, note that the parameters user_global_ns and global_ns have been deprectated in favour of user_module and module respsectively. The new parameters expect a module-like object, rather than a namespace dict. The old parameters remain for backwards compatibility, although user_global_ns is now ignored. The user_ns parameter works the same way as before, and calling embed() with no arguments still works as before.
The previous version (IPython 0.11) was released on July 31 2011, so this release cycle was roughly 4 1/2 months long, we closed a total of 515 issues, 257 pull requests and 258 regular issues (a :ref:`detailed list <issues_list_012>`_ is available).
Many users and developers contributed code, features, bug reports and ideas to this release. Please do not hesitate in contacting us if we’ve failed to acknowledge your contribution here. In particular, for this release we have had commits from the following 45 contributors, a mix of new and regular names (in alphabetical order by first name):
Note
This list was generated with the output of git log rel-0.11..HEAD --format='* %aN <%aE>' | sed 's/@/\-at\-/' | sed 's/<>//' | sort -u after some cleanup. If you should be on this list, please add yourself.