Introduce `StringViewArrayBuilder::map` to avoid duplication

[alamb]

I think you could avoid a lot of this copy / paste by using a StringViewBuilder with the append_block and append_view_unchecked methods

https://docs.rs/arrow/latest/arrow/array/type.StringViewBuilder.html#method.append_block

That being said, I do think it could be slightly slower than this implementation because it would have to re-check the length

It almost seems like what we want is some sort of API on StringViewArray itself, similar to https://docs.rs/arrow/latest/arrow/array/struct.PrimitiveArray.html#method.unary

So this code could be written something like

let new_array = orig_array.map_values(convert)

That would also let us do potentially crazy things like reuse the buffer allocations and modify the values in place if they weren't shared 🤔

If that makes sense to you I can file a ticket in arrow-rs perhaps.

Originally posted by @alamb in #21980 (comment)

[alamb]

As @neilconway says #21980 (comment)

I meant something similar to what you suggested in the PR -- for example, a function like:

  impl StringViewArrayBuilder {
      /// Reserve `len` bytes of output storage and let the caller fill them.
      pub fn append_with<F: FnOnce(&mut [u8])>(&mut self, len: usize, fill: F);
  }

We could use that in reverse, translate, initcap, and lower/upper. I want to check whether it regresses performance against the hand-optimized version though.

[alamb]

I think the idea of reusing the allocations for StringView is an interesting one (e.g. mutating the bytes in place when possible), especially for our string functions which will often be creating a temporary intermediate (the input is often not used again)

[buraksenn]

@neilconway will you take a look into this? If not I can take a look into this quickly

[neilconway]

@buraksenn I'm already working on a PR for the append_with idea I mentioned.

[neilconway]

take