feat: add write_bytes for GenericBinaryBuilder

[tisonkun]

Similar to impl Write for GenericStringBuilder.

We found this helps in reducing one allocation when append bytes in many calls (compared with gather those bytes into a Vec first).

Not sure if there is some trait suitable for this method but just add it first.

cc @alamb @waynexia

The doc tests cover the basic usage.

[tisonkun]

Try impl io::Write interface. No sure if a good fit.

[tisonkun]

@andylokandy Thanks for your review! Addressed.

[waynexia]

From the API level: this looks kind of misleading to me. The proposed Write::write or write_bytes makes BinaryBuilder closer to an opaque byte container, shadowing the concept of item and offset. I.e., array built from this API would contain only one element. Plus we don't have an API to "cut" the bytes buffer, but need to call builder.append_value(""); which looks a bit hack.

How about considering something like Vec::from_raw_parts, which accepts different buffers obviously and hence avoids the potential misunderstand?

[andylokandy]

I'm the member of databend, which heavily adopts arrow-rs and I'm sure it is a very common pattern to construct an item in a StringArrayBuilder/BinaryArrayBuilder by several parts of &str or &[u8].

For example, once this PR is accepted, to add a space padding before each item of BinaryArray and generate a new BinaryArray, we can do the following:

for x in array.iter() {
    write!(&mut builder, b" ").unwrap();
    write!(&mut builder, x).unwrap();
    builder.append_value("");
}

Moreover, StringBuilder has already supported this pattern. We use it in databend smoothly and we hope BinaryBuilder can have the same ability.

[waynexia]

I'm the member of databend, which heavily adopts arrow-rs and I'm sure it is a very common pattern to construct an item in a StringArrayBuilder/BinaryArrayBuilder by several parts of &str or &[u8].

For example, to add a space padding before each item of BinaryArray and generate a new BinaryArray, we can do the following:

for x in array.iter() {
    write!(&mut builder, b" ").unwrap();
    write!(&mut builder, x).unwrap();
    builder.append_value("");
}

Moreover, StringBuilder has already supported this pattern. We use it in databend smoothly and we hope BinaryBuilder can have the same ability.

Thanks for your insight, this case makes sense to me 🙌 BTW, do you think it's necessary to encapsulate this pattern builder.append_value(""); into a dedicated method?

[andylokandy]

I'll suggest builder.commit_row();. What do you think?

[alamb]

Thank you @tisonkun and @tustvold