Add finish_preserve_values to ArrayBuilder trait

[adamreichold]

Which issue does this PR close?

• Closes How to handle delta dictionaries and finish_preserve_values with nested builders #9600.

Rationale for this change

Delta dictionaries require user code to call finish_preserve_values instead of finish on the dictionary builders. But this is generally not possible if those builders are nested within composite builders like lists, maps or structs. Using the new trait method, user code can call this generically which has no effect on e.g. primitive builders due to the default implementation but forwards to this method for composite ones.

What changes are included in this PR?

Adds a new method finish_preserve_values to ArrayBuilder with a default implementation that forwards to finish.

Implements this method for dictionary and composite builders.

Are these changes tested?

Adds tests where new finish_preserve_values inherent methods were created to check that the correct method on the constituent builders is called using a shared mock builder PreserveValuesMock.

Are there any user-facing changes?

This adds a new method finish_preserve_values to the central ArrayBuilder trait, but it has a default implementation so this should be backwards compatible.

[alamb]

@albertlockett I wonder if you might have some time to review this PR and offer an opinion?

[albertlockett]

LGTM, this look the right approach. Thanks for the contribution @adamreichold !

[adamreichold]

Friendly ping. With the positive feedback on the basic approach, I would be happy to fix any remaining technicalities so our project can make use of this via the planned 58.2.0 or 59.0.0 release. Thanks!

[alamb]

Thank you @adamreichold and @albertlockett

I took a look and this PR makes sense to me and seems well tested

It is a little unfortnuate that now we end up with three versions of build for all arrays, but I don't have any good suggestion for this other than to make a more generic version that can pass an options struct Array::finish_opts(&self, opts: FinishOptions)

If we find we wand to pass more options to finish in the future, perhaps then we can deprecate finish_preserve_values and add something more like finsh_opt

[adamreichold]

make a more generic version that can pass an options struct Array::finish_opts(&self, opts: FinishOptions)

The main issue I see here is that the receiver also depends on the options, i.e. &mut self for the consuming modes and &self for the cloning ones.

Will fix the intra doc links and push if the CI finds no other issues.

[adamreichold]

Will fix the intra doc links and push if the CI finds no other issues.

Pushed a rebase and the intra doc link fix.

[alamb]

The main issue I see here is that the receiver also depends on the options, i.e. &mut self for the consuming modes and &self for the cloning ones.

Yeah, it would be trickey

[alamb]

🚀