Optimize `to_char` for array conversions

[alamb]

Is your feature request related to a problem or challenge?

• While reviewing Format Date32 to string given timestamp specifiers #15361 I noticed that the to_char implementation for arrays could likely be improved substntially

Specifically the code that

The use of Vec<Option<String>> to build a string array is non ideal as it requires additional allocations for each output row along with a second copy of the data

fn to_char_array(args: &[ColumnarValue]) -> Result<ColumnarValue> {
    let arrays = ColumnarValue::values_to_arrays(args)?;
    let mut results: Vec<Option<String>> = vec![]; // <--- this is non ideal

Also, the code that falls back to date formatting calls cast for a single row every time the formatting fails -- since the failure is likely due to a date specifier that will get called for all rows, it would probably be faster to convert the entire input array to Date once the first time the specifier is retried

Describe the solution you'd like

Make to_char faster

You can evaluate your changes with the existing benchmark

cargo bench --bench to_char

Describe alternatives you've considered

No response

Additional context

No response

[Omega359]

Also, the code that falls back to date formatting calls cast for a single row every time the formatting fails -- since the failure is likely due to a date specifier that will get called for all rows, it would probably be faster to convert the entire input array to Date once the first time the specifier is retried

Maybe. but that wasn't @friendlymatthew's experience as noted in this comment

[alamb]

I was thinking of something even simpler -- convert the entire array (not just the elements that failed) -- the conversion will be super fast (it is some vectorized arithmetic)

[Omega359]

Good first issue perhaps?

[alamb]

Good first issue perhaps?

I think this would be a good first issue for someone who was already an expert at arrow and rust. I tried for like 30 minutes and I found it pretty tricky