Rework of dfs functions to recursively

[Jolanrensen]

#333

[Jolanrensen]

While the current tactic works in all current tests, I found a major breaking new case:

public fun <C> ColumnSet<C>.first(condition: ColumnFilter<C> = { true }): SingleColumn<C> =
    transform { listOf(it.first(condition)) }.singleImpl()
    
df.select { something.first { ... }.recursively() }

Not only does recursively() currently not work well on SingleColumns (it puts an all() in between, to flatten a potential ColumnGroup and then runs on that), it should also pass the "recursively" call through singleImpl() and deliver it to transform {}...

Maybe my original idea of creating a different return-type for transform calls that allows them to be re-executed recursively was not such a bad idea...

[Jolanrensen]

df.select { first { ... }.recursively() } now works, but I 'm not sure why. Needs more tests and maybe a clearer way to indicate a ColumnSet passes on its recursive behavior to its parent or not.
For first {} which uses singleImp(), singleImpl() was modified to pass on its recursive behavior. This might however be needed in more places.

[Jolanrensen]

df.select { it[0, 3].recursively() }
what does this even mean? Take index 0 and 3 of every column group maybe. But this would then break when it hits a non-columnGroup.
Allowing .recursively() on any columnSet might open a can of worms we don't want. I think I'm gonna have another attempt at changing the return type of functions that allow a .recursively() call in favor of this generic but messy approach.

@nikitinas what do you think?

Also, I'm curious about how you divided up the use of SingleColumn<*> and ColumnSet<*>, because many functions were defined like ColumnSet<*>.something() (because SingleColumn inherits from ColumnSet too) but then assume the receiver to be a SingleColumn ColumnGroup taking its children instead of just looping over the columns in the column set. This also made dfs {} behave unexpectedly combined with all().

How I see it there are two options:

• SingleColumn<*>.cols(): ColumnSet<*> operates on the children of the single column.
  • df.select { cols(1, 2, 3) }
  • df.select { someGroup.cols(1, 2, 3) }
  • df.select { first { it.isColumnGroup() }.cols(1, 2, 3) }
• ColumnSet<C>.cols(): ColumnSet<C> operates on the columns in the column set.
  • df.select { colsOf<String>().cols(1, 2, 3) }
  • df.select { all().cols(1, 2, 3) }

Also, what's the difference between all(), cols(), and children()?

This distinction also impacts how .recursively() works

[Jolanrensen]

This will be closed in favor of #363

[koperagen]

Why? I think SingleColumn<AnyRow> is a ColumnGroup and these operations make sense, but not for SingleColumn<*>

[Jolanrensen]

Well yes, but then it breaks when you

df.select { it["name"][0..5] }

since it doesn't know the type of the single column you select that way.

Also, I'm gonna close this PR in favor of dfs-rename3, just waiting for Tolya to answer.