One principle for rastr development that I think we should document is that NaNs are treated as usual instead of unusual
Numpy tends to view NaNs as suspicious, and needing an escape hatch
Rastr tends to treat them as expected and first-class values (in a sense)
As a result, bugs like #395 are quite easy to creep in. It would be worth adding this information to the copilot instructions file, as well as CONTRIBUTING.md
Instructions file might be something like
Something like "always consider how NaNs are handled. We can expect that most rasters will have NaNs. Don't let them propogate disproportionately causing unexpected behaviour for operations and methods."
One principle for
rastrdevelopment that I think we should document is that NaNs are treated as usual instead of unusualNumpy tends to view NaNs as suspicious, and needing an escape hatch
Rastr tends to treat them as expected and first-class values (in a sense)
As a result, bugs like #395 are quite easy to creep in. It would be worth adding this information to the copilot instructions file, as well as
CONTRIBUTING.mdInstructions file might be something like