Nick Crews

I don't. And thinking about it more, pluggy seems a bit overkill. What about we do what pandas does with their plotting backend? https://plotly.com/python/pandas-backend/ Eg they say `pandas.options.plotting.backend = "plotly"`...

I looked at #1121, some specific comments there. Check out my solution at #1122. Another benefit I realized of using spec-strings instead of pluggy is that it is way more...

I'll throw in data import/export. I find guidance on this topic currently pretty limited in the docs, but this is a step that every single user is going to do....

The Sink idea is good. A few thoughts: 1. why does `to_sinks()` handle multiple sinks? In case you want to write to both parquet and csv, and they could be...

I just ran into this when implementing [a PR in Ibis](https://github.com/ibis-project/ibis/pull/8666/commits/148ebb0e4d1eb6613326f6a51c3beb14b54e1826#diff-4b1d72810ad7e91258141d5a9995f7bfb42d6ad232f7975ea7a68406d5292a8aR72-R89) Not just arrow, but other backends such as duckdb, flink, and bigquery all have nullable structs. Polars is definitely...

Thoughts: - The log function should take the same severity levels as the building logging module as an optional kwarg. Default is INFO? - If we're smart, it could also...

This looks good to me!

Unfortunately I don't think that pluggy was actually a good suggestion from me, sorry :( I think the largest drawback is the boilerplate that you describe above that plugin authors...

This is much better than pluggy, awesome! I think the basic philosophical difference between this and #1122 is who has the responsibility of defining the lookup key, and whether namespacing...

> i like the simplicity of this approach, but it will be a breaking change for data models that include datetime. in this code this is only working because of...