Implemented native inputhooks for MacOSX backend #156
Conversation
Pulled in IPython (7.10.1) code for MacOSX backend. This code contains MacOSX-native inputhooks. It cannot directly be incorporated IPython uses prompt_tooklit for AsyncIO. In order to avoid dependency on prompt_toolkit, wrote a custom event-loop trigger. See details below. Benefits: - The MacOSX event loop is interrupted only on user input. This greatly improves responsiveness of the backend. The backend now feels more fluid and mac-like. - In PyCharm, the figure does not remain in the forefront even after Alt-Tab. Impelementation Details: IPython uses prompt_toolkit trigger mechanism to initiate callback on user input. To avoid the prompt_toolkit dependency, a similar trigger mechanism using pipes is impelemented along with dedicated thread to check for user input (by checking stdin_read()). When user input is available, this thread terminates the event loop using the pipe functionality similar to IPython. Updates
|
For reference, commented on a long-pending request on PyCharm forum asking for a fix to this problem: https://youtrack.jetbrains.com/issue/PY-28781#focus=streamItem-27-3868844.0-0 |
Implemented a singleton resource manager to avoid creating pipe resources for every event loop. This also plays nicely with the native hooks. In a previous implementation, closing the pipes lead the native callbacks getting triggered immediately.
pydev_ipython/inputhookmac.py
Outdated
| import os | ||
| from pydev_ipython.inputhook import stdin_ready | ||
| import time | ||
| from threading import Thread, Event |
There was a problem hiding this comment.
Can you change this import for: from _pydev_imps._pydev_saved_modules.threading import Thread, Event ?
The main issue is that gevent monkey-patches threading and that import will make sure that the original (non monkey-patched version) of the module is used.
pydev_ipython/inputhookmac.py
Outdated
|
|
||
| def inputhook_mac(): | ||
|
|
||
| import datetime |
There was a problem hiding this comment.
Can you make this import top-level? The reason for this is that Python 2 has an import lock, so, if a thread is stopped in a breakpoint inside of some import no other import will run (so, in the debugger imports must be top-level).
gevent monkey-patches threading and this import will make sure that the original (non monkey-patched version) of the module is used.
|
@fabioz I did some more testing of the commit and the commit itself seems to work well. But I noticed that on a MacOSX/Pycharm and using either the MacOSX or WXAgg backends, the responsiveness on new code entry starts out well but after a time, there is a 10second delay for any new input in the pydev_code_executor/add_exec loop. This happens only when the matplotlib window is not in focus. When it is in focus, the responsiveness on new code entry is immediate. This problem is not in IPython for either of these backends. In IPython, even after a long delay, the 10 second delay does not kick-in with or without the matplotlib window being in focus and the responsiveness is consistently good. Do you know the source of this difference? Once difference could be because IPython uses prompt_toolkit/asyncio and PyDev uses XML-RPC. But don't know enough to understand this further. |
|
I figured out the issue. PyCharm was under App Nap and the terminal for ipython was not. So the differences between the two. I was not able to disable App Nap only for PyCharm so I disabled App Nap across all applications. After that, I am no longer facing any issues.
|
For those interested, I disabled Nap App on macos using the following - |
|
@skrisna, it looks like this hack doesn't work in PyCharm with |
Things also work with |
I had the same issue and so did people over at ipython, so I applied their fixes and this seems to have done the trick (see #209). |
Awesome, thanks @Jofkos. I'll give the code over in #209 a shot. |
Pulled in IPython (7.10.1) code for MacOSX backend. This code contains
MacOSX-native inputhooks. It cannot directly be incorporated IPython uses
prompt_tooklit for AsyncIO. In order to avoid dependency on prompt_toolkit,
wrote a custom event-loop trigger. See details below.
Benefits:
improves responsiveness of the backend. The backend now feels more fluid and
mac-like.
Impelementation Details:
IPython uses prompt_toolkit trigger mechanism to initiate callback on user
input. To avoid the prompt_toolkit dependency, a similar trigger mechanism
using pipes is impelemented along with dedicated thread to check for user
input (by checking stdin_read()). When user input is available, this thread
terminates the event loop using the pipe functionality similar to IPython.
Updates