More trivial implementation of `Box<dyn AsyncArrowWriter>` and `Box<dyn AsyncArrowReader>`

[ethe]

In the current,

impl AsyncFileReader for Box<dyn AsyncFileReader> {
    ...
}

equals to

impl AsyncFileReader for Box<dyn AsyncFileReader + 'static> {
    ...
}

which 'static bound is not necessary, for all 'a, Box<dyn AsyncFileReader + 'a> should be an AsyncFileReader. I removed this bound to make it more trivial

[jayzhan211]

I think adding read and write is not the same as static

[ethe]

It is not the same, but a superset of it. 'read and 'write are able to be 'static, and also make non-static for<'a>: Box<dyn AsyncArrowWriter + 'a> be an AsyncArrowWriter

[jayzhan211]

I'm still not convinced why we need to add lifetime for it. 👀 In my understanding, we should start from the simple one without lifetime until we want to implement with reference that compiler starts to complain about it.

It would be clear if there is an example that compilation failed without lifetime

[ethe]

Considering this case:

pub struct ArrowReader<'reader, T> {
    reader: &'reader T,
}

impl<T: AsyncFileReader + Send + Sync> AsyncFileReader for ArrowReader<'_, T> {
    fn get_bytes(
        &mut self,
        range: std::ops::Range<usize>,
    ) -> std::pin::Pin<
        Box<dyn std::future::Future<Output = parquet::errors::Result<bytes::Bytes>> + Send + '_>,
    > {
        todo!()
    }

    fn get_metadata(
        &mut self,
    ) -> std::pin::Pin<
        Box<
            dyn std::future::Future<
                    Output = parquet::errors::Result<Arc<parquet::file::metadata::ParquetMetaData>>,
                > + Send
                + '_,
        >,
    > {
        todo!()
    }
}

fn test<'r, T: AsyncFileReader + Send + Sync>(
    file: ArrowReader<'r, T>,
) -> impl AsyncFileReader + 'r {
    Box::new(file) as Box<dyn AsyncFileReader>
}

test function would not be passed using current implementers, after merging this pr, it will

[ethe]

In my understanding, we should start from the simple one without lifetime

Actually it does have one lifetime in the current: 'static, lifetime is just elided rather than non-existent.

If you are still confused, there is a basic explanation about it: https://quinedot.github.io/rust-learning/dyn-elision-basic.html#impl-headers

Consider using implementations like the following if possible, as they are more general:

// Implemented for all lifetimes
impl Two for Box<dyn Trait + '_> {}

// Implemented for all lifetimes such that the inner lifetime is // at least as long as the outer lifetime impl Two for &(dyn Trait + '_) {}