Coverage increased (+0.05%) to 99.422% when pulling 1492c8a on eriknw:descriptor_opts into 70a0d17 on python-graphblas:main.

1. axb_method= or AxB_method=?

I decided to normalize everything to lowercase and raise if this causes duplicates. Since descriptor options are keyword arguments, they can potentially overlap with other argument names, and a workaround is to use a capitalized version of the keyword for the descriptor.

5. There are a couple spots where descriptors are ignored b/c they can't be used, because a backend could always have an extended API that does. Should we try to create the descriptor anyway? If we do and it's non-NULL, should we raise (b/c options aren't used)?

I decided to create the descriptor, then ignore it. This at least ensures that the descriptor options were valid, which will help us transition to when descriptors can be used.

@jim22k, auto-compute of expressions leads to an interesting edge case when it comes to custom descriptors. Suppose an expression is used causing it to auto-compute (and the result is stored as e.g. expr._value). Now suppose you call .new(**opts) on the expression giving it custom descriptors. Currently, we ignore the custom descriptors and return the already-computed expr._value. I think this is okay. If a workflow with a backend requires fine control with custom descriptors, then auto-compute should probably be disabled to avoid this.

I'm going to merge soon--gonna do one final scan and check coverage.

This is in!

This is a significant addition, and it feels great to close an issue that is 2.5 years old 😃