Changes in IPython kernel¶
- Fix eventloop-integration bug preventing Qt windows/widgets from displaying with ipykernel 4.6.0 and IPython ≥ 5.2.
- Avoid deprecation warnings about naive datetimes when working with jupyter_client ≥ 5.0.
Add to API
DisplayPublisher.publishtwo new fully backward-compatible keyword-args:
display_datamessages spec for
publish. For a display data message,
transientcontains data that shouldn’t be persisted to files or documents. Add a
ipykernel_launchermodule which removes the current working directory from
sys.pathbefore launching the kernel. This helps to reduce the cases where the kernel won’t start because there’s a
random.py(or similar) module in the current working directory.
Add busy/idle messages on IOPub during processing of aborted requests
Add active event loop setting to GUI, which enables the correct response to IPython’s
Include IPython kernelspec in wheels to reduce reliance on “native kernel spec” in jupyter_client
OutStreamto inherit from
TextIOBaseinstead of object to improve API support and error reporting
Fix IPython kernel death messages at start, such as “Kernel Restarting...” and “Kernel appears to have died”, when parent-poller handles PID 1
- Fix bug when instantiating Comms outside of the IPython kernel (introduced in 4.5.1).
- Add missing
streamparameter to overridden
- Remove locks from iopub thread, which could cause deadlocks during debugging
- Fix regression where KeyboardInterrupt was treated as an aborted request, rather than an error
- Allow instantiating Comms outside of the IPython kernel
- Use figure.dpi instead of savefig.dpi to set DPI for inline figures
- Support ipympl matplotlib backend (requires IPython update as well to fully work)
- Various bugfixes, including fixes for output coming from threads,
input()when called with non-string prompts, which stdlib allows.
- Fix circular import of matplotlib on Python 2 caused by the inline backend changes in 4.4.0.
- Use MPLBACKEND environment variable to tell matplotlib >= 1.5 use use the inline backend by default.
This is only done if MPLBACKEND is not already set and no backend has been explicitly loaded,
matplotlib.use('Agg')will take precedence.
- Fixes for logging problems caused by 4.3, where logging could go to the terminal instead of the notebook.
--profilearguments to ipython kernel install
- Allow Comm (Widget) messages to be sent from background threads.
- Select inline matplotlib backend by default if
matplotlib.use()are not called explicitly (for matplotlib >= 1.5).
- Fix some longstanding minor deviations from the message protocol (missing status: ok in a few replies, connect_reply format).
- Remove calls to NoOpContext from IPython, deprecated in 5.0.
- Use a nonempty dummy session key for inprocess kernels to avoid security warnings.
- Fix Windows Python 3.5 incompatibility caused by faulthandler patch in 4.3
- Publish all IO in a thread, via
IOPubThread. This solves the problem of requiring
sys.stdout.flush()to be called in the notebook to produce output promptly during long-running cells.
- Remove refrences to outdated IPython guiref in kernel banner.
- Patch faulthandler to use
sys.__stderr__instead of forwarded
sys.stderr, which has no fileno when forwarded.
- Deprecate some vestiges of the Big Split:
ipykernel.find_connection_file()is deprecated. Use
jupyter_client.find_connection_file()instead. - Various pieces of code specific to IPython parallel are deprecated in ipykernel and moved to ipyparallel.
- Don’t show interactive debugging info when kernel crashes
- Fix handling of numerical types in json_clean
- Testing fixes for output capturing
- Fix missing
ipykernel.__version__on Python 2.
- Fix missing
target_namewhen opening comms from the frontend.