Repository navigation
ENH: support for PySide6 in %gui - #13864
Conversation
bd6947d to
5409de6
Compare
|
I'm not sure what to do about the docs failing to build - it doesn't seem to be an issue in main? |
aeed549 to
18cf490
Compare
|
@Carreau what's this force push about? |
Some failures were due to a problem on the main branch, I fixed the test on the main branch and rebase this one top of main, to get the test to pass. |
See ipython/ipykernel#1071 Note that it's still there if coming from `matplotlib`
|
This is ready for review. The equivalent changes in ipykernel have already been merged. |
|
There's a problem here with
I have an application where I require the if matplotlib.get_backend() != 'TkAgg':
_l.debug('IPython shell detected. Attempting to change '
'matplotlib backend to `TkAgg`...')
try:
matplotlib.use('TkAgg')
except ImportError as e:
raise RuntimeError('Cannot run. Matplotlib backend cannot be changed to `TkAgg`:\n'
f' {str(e)}')Sadly, the call to So I'm changing the error to a warning. My rationale was that I'd want to prevent people from unknowingly changing which event loop was tied in - you could stop some other application from functioning if you do so - but I suppose if you really wanted to prevent that, |
An exception here causes issues because `matplotlib.get_backend()` actually sets a backend if none was set.
ccordoba12
left a comment
ccordoba12
left a comment
There was a problem hiding this comment.
Thanks @shaperilio for your work on this! I left some comments for you, which are basically the things I changed in ipykernel to provide a better user experience, in my opinion.
Thanks for these. I should have learned my lesson with the |
shaperilio
changed the title
PySide6 in %gui`PySide6 in %gui
|
Any thoughts on when this may be merged, or if there's more feedback? I believe it's been released in ipykernel. |
|
Sorry, this fell under the radar. Let's try it. |
I started using the released version of my `PySide6`-enabling changes and noted some problems. In this PR, I fix those, and also overall improve the feedback to the user when a GUI event loop is hooked in: - Report which event loop is running when using `%gui <some GUI>`; e.g. `%gui qt` will show `Installed qt6 event loop hook.` - Report when the event loop is disabled; i.e. `%gui` will show `GUI event loop hook disabled.` if an event loop hook was installed, or `No event loop hook running.` if nothing was installed. - Requesting a second event loop will give the message `Shell is already running a gui event loop for <some GUI>. Call with no arguments to disable current loop.` - Requesting a different version of Qt, i.e. `%gui qt6` followed by `%gui` followed by `%gui qt5` will show `Cannot switch Qt versions for this session; will use qt6.` followed by `Installed qt6 event loop hook.` (Fixes / improves #13864)
Addresses #13859
Changes here parallel the changes in ipykernel, i.e. prefer
PyQtoverPySideand allow requesting explicit versions (e.g. 'qt5') with one difference (see below).I believe that, eventually, the Qt importing logic should be all here, and
ipykernelshould defer toipython. I chose not to do that at this time since it would mean the latestipykernelwould require the latestIPython; I'm happy to discuss further.The only difference between
IPythonandipykernel, after these two pull requests, is that, inIPython, it's not possible to explicitly request an event loop for Qt4. This is because an alias exists here which effectively makes "qt4" be "the latest Qt available". I did not remove the alias because I don't know the history behind it.