In a call last week we discussed dropping the support for cythonization of the scheduler code.

The cythonization project was an attempt to improve the scheduler performance. While it indeed delivered improved performance it was not as significant as originally hoped.

The cythonization annotations introduce a few subtle problems to maintenance and development

• CI builds take longer
• Annotations are not mypy / IDE friendly
• The code became increasingly unreadable
• Perfectly fine python code is not always compilable requiring us to deal with cython whether we want to or not
• Development on the project stalled
• There is no published / packaged build available
• We do not have any benchmarking available and need to rely on few individuals to "guesstimate" whether code changes might hurt the cythonization

Considering that the project is currently stalled, the costs are ongoing and the value for users is very limited, I suggest dropping support for cythonization.
We may want to pick up cythonization at a later point in time for specific applications again, for instance as specialized c extensions.

@quasiben, @jakirkham, @jrbourbeau, @jcrist

xref Drop PyPy #5681