Tom White

There was another race condition failure here: https://github.com/pystatgen/sgkit/runs/2157617424?check_suite_focus=true. The failure occurred in a different place though - between writing the zarr array, and reading it back in the test. So...

No particular rush, just wanted to open it for discussion. Installing Python 3.8 envs using conda is easy enough on most platforms so I suspect that shouldn't be a blocker.

Well that was quick. Our [upstream tests just failed](https://github.com/pystatgen/sgkit/runs/4843974784?check_suite_focus=true) because xarray has already [dropped Python 3.7](https://github.com/pydata/xarray/issues/6138).

[Pandas 1.4.0](https://pandas.pydata.org/docs/dev/whatsnew/v1.4.0.html#increased-minimum-version-for-python) does not support Python 3.7.

The build is now failing on Python 3.7 (https://github.com/pystatgen/sgkit/actions/runs/3186604485), so I'm thinking it's probably time to remove support for it.

One pattern that may be useful is to compute new variables in a new dataset (via `merge=False`), save only the new variables, then merge the datasets. This avoids having to...

I've been trying out a checkpoint function that I think is a better way of doing what I was trying to achieve in my previous [comment](https://github.com/pystatgen/sgkit/issues/347#issuecomment-717400639). ```python ds = ......

I used a conditional form of this pattern in a [notebook](https://nbviewer.jupyter.org/github/tomwhite/shiny-train/blob/sgkit/notebooks/gwss/sgkit_pbs.ipynb), so that it can be rerun without having to recompute variables: ```python ds = ... # original dataset if...

Some more ideas from @eric-czech and @alimanfoo: * Have option for checkpoint to force an overwrite (Hail does this) * Rather than checking for the existence of the variable in...

> Is there a way we can have the example notebooks run less frequently than most of the tests? The main build [runs Sphinx to look for doc warnings](https://github.com/pystatgen/sgkit/blob/main/.github/workflows/build.yml#L33-L36) for...