Our cargo publish sequence is currently quite complicated because we have circular dev dependencies. It's possible to work around this but it is annoying and not great practice!
I think we should:
- Add some check for circular dependencies in the dependency graph. The only culprits I know about are
sedona and sedona-testing in dev-dependencies.
- I don't know how sedona-raster became a dependency of sedona-functions but we should probably move that
- sedona depending on sedona-testing (for the random geometry generator) needs to get fixed.
- Separate out a crate for high-level benchmarks and/or high level tests. DataFusion has most of its tests in the
datafusion crate's tests directory and a completely separate benchmark setup, perhaps for this reason. I do like keeping most of the tests file-local and I'm sure we can find a way to separate out the logic we need (maybe a high- and low- level testing crate) to keep that (even if we have to move some higher level integration tests or benchmarks).
- Many of our benchmarks could be dynamic (e.g., we could have a CLI that passes some ArgSpecs as strings).
Our cargo publish sequence is currently quite complicated because we have circular dev dependencies. It's possible to work around this but it is annoying and not great practice!
I think we should:
sedonaandsedona-testingin dev-dependencies.datafusioncrate'stestsdirectory and a completely separate benchmark setup, perhaps for this reason. I do like keeping most of the tests file-local and I'm sure we can find a way to separate out the logic we need (maybe a high- and low- level testing crate) to keep that (even if we have to move some higher level integration tests or benchmarks).