Repository navigation
Conversation
|
I haven't worked with this code recently and I don't know why we have a regression, but as a general rule, threads at the heart of our event loop system have been historically nothing but a source of pain. A long time ago, we were able to move away from threads (the original solution for GUI integration dating back to ~2003/4), and that was a huge gain in stability. We should really think long and hard about how to figure out a non-thread solution. For reference, this is an attempt to fix #9652. |
|
That sounds a bit more palatable than a threads solution (plus, it's what we've used for years, so it has the benefit of being battle-tested). |
I'll try. |
|
I like #9702 a lot more, and it's working for me. We've had headaches with eventloops in threads for years, from the old |
@jorisvandenbossche can you try that ?
I honestly have close to no clue what I am doing, as eventloop and threading is far from where I am comfortable.
I guess the other alternative would be to mimic the other inputhook are doing on the IPykernel side and timeshare. I'll see if I can borrow a windows box tomorrow from one of my officemates.
I'm also unsure if I can use things like
msvcrt.kbhit()withprompt_toolkit.CC @fperez , @minrk @takluyver .