Ian Rose

I think we should be able to follow the same model for notebooks, and have a "trust file" option. I'll take a crack at that over at the other PR....

Thanks for the report @ncclementi! This is probably due to the change to the `sort` arg in #9428 . It is now overriding the defaults for some of the aca...

> I was under the impression that passing None would be the same as passing nothing in this case I was also under that impression -- but the `None` overrides...

A bit of supporting information: shown are boxplots for duration and average memory. The first is the commit before #9302, the second is #9302, and the third is the same,...

From my perspective, going forward with #9386 seems like it would be a fine choice. Over there you said: > [it] does not improve performance when the aggregation fits comfortably...

Hmm, I'm still seeing `sort=False` have some worse behavior than just not passing it in -- I guess I'll need to dig a bit deeper.

Hmm, seems like `factors` is used in `distributed`, so perhaps we need to keep it around for the time being. Change looks good otherwise.

Another use-case for something like this would be reading in hive-partitioned parquet datasets. That partitioning scheme doesn't fit in well with dask's `divisions` tracking, and it makes certain very obvious...

Hi @aloysius-lim , thanks for the update. I was able to reproduce your issue on using numpy==1.21.4, so it seems the underlying problem in #7170 is not fully resolved. Of...

Update here: @pavithraes and I took another look at it, and it still reproduces with `distributed==2022.8.1` and `numpy==1.23.1`