[C++][Compute] Fix null sorting of multiple sort keys

[Light-City]

Describe the bug, including details regarding any error messages, version, and platform.

Currently, NullPlacement is a global control, and this parameter is used for all sortkeys. However, for databases like pg, it should be one for each sortkey, for example:

order by i nulls first, j, k nulls first;

code is

 /// Column key(s) to order by and how to order by these sort keys.
 std::vector<SortKey> sort_keys;
 /// Whether nulls and NaNs are placed at the start or at the end
 NullPlacement null_placement;

Component(s)

C++

[mapleFU]

So the NullPlacement should support being std::vector<NullPlacement> ?

[kou]

How about moving NullPlacement to SortKey?

[Light-City]

So the NullPlacement should support being std::vector<NullPlacement> ?

yes.

[Light-City]

How about moving NullPlacement to SortKey?
In this case, it's better

[Light-City]

If no one does it, I can try to implement it

[mapleFU]

Also cc @bkietz @ianmcook for some info here

[mapleFU]

https://github.com/apache/arrow-rs/blob/829708d4911bc0798ca2a388cd7df20478a4cc08/arrow-schema/src/lib.rs#L37

Seems that maybe this is better. But waiting for others idea for compability

[bkietz]

when sorting multiple columns, rust also allows per-column null placement https://github.com/apache/arrow-rs/blob/1807abace4153cbfd287ee830f86ebaeeae50234/arrow-ord/src/sort.rs#L611

Seems reasonable to support in c++

[Light-City]

It has been implemented, I will send a PR later, haha

[llama90]

I saw a comment about difficulties with sorting for multiple batches (ChunkedArray, Table), but it looks like you've solved it. lol

I'm interested in this issue because I'm interested in sorting for DictionaryType in multiple batches.