group by

[ggrothendieck]

For increased expressiveness and consistency it would be nice if every subcommand for which it makes sense had a group by flag specifying which columns to group by. Conceptually this amounts to splitting into disjoint sets of rows, applying the subcommand to each set and putting it back together. 

Another possibility would be to have a group by subcommand which takes another qsv command or even non-qsv command as arguments. This would be more powerful but likely slower.

These are the subcommands that seem relevant:

enum
count
frequency
luau
pseudo
reverse
sample
slice
sort
sortcheck
transpose

[jqnatividad]

Thanks for this detailed request @ggrothendieck .

I agree that this would be a powerful addition to qsv.

For the initial implementation, I'd go with the second option by creating a new groupby command.

In later releases, we can build on the groupby command and use it as a common option in a manner akin to the select command and the --select common option used throughout the qsv suite.