Ian Thomas

@mrocklin There is no particular problem that I am trying to solve, so I guess we are in "just for fun" territory! I was looking around dask for issues I...

> Oh I see, great. So we're not polling on each Comm every 5ms, we're polling on all comms every 5ms. @mrocklin No, you were correct first time, each Comm...

@dalcinl Thanks for your suggestions on the improved `iprobe/recv` combination and on `mpi4py.futures`. I will take a look at both.

@mrocklin Your example computation ```python x = da.random.random((20000, 20000)).persist() start = time.time() y = (x + x.T).transpose().sum().compute() print('Seconds:', time.time() - start) ``` using 5 MPI processes (3 workers) on my...

This is as polished as I am going to get it. Now just a single object (per Scheduler/Client/Worker) polls for incoming messages and passes them to the appropriate receiver. On...

This is still a problem on an M3 running Sonoma 14.4.1 and `Python 3.12.3, IPython 8.23.0, ipykernel 6.29.4` and `Matplotlib 3.8.4`. Using `%matplotlib osx` in IPython works fine, but the...

@ivanlen Note there is a fix for this in #1237.

This should be fairly easy to implement, the question is whether it will break anything because of the previously valid assumption that a `GeoDataFrame` always contains data. I'll take a...

I think we have to stick with consistency with posix. I know we have made many backwards-incompatible changes in the last 6 months or so but they have all been...

> On our side we should consider if Bokeh could consider providing a wheel that doesn't depend on tornado and contourpy which aren't available in pyodide. ContourPy is available in...