Remove hyper: Use methods in viz and axum.

[martinfrances107]

hyper 1.0 has been released

-hyper = "0.14.23"
+hyper = "1"

Conventionally 0.x releases are considered pre-releases.
It seems hyper is acting in this vein.

here is the relevant blogs post, which is linked in the release notes.

stability-promise

[martinfrances107]

I have found googling the solution to this stuff troublesome

So I'm just making notes for myself, that will also be useful in review.

upgrading guide.. to 1.0

Also this change log

contains this pivotal sentence

The free functions hyper::body::to_bytes and aggregate have been removed. Similar functionality is on http_body_util::BodyExt::collect.

[martinfrances107]

This is a experimental PR.

At two places in the codebase "A" is converted to "B" - where

A)

let body = body::to_bytes(body).await.unwrap_or_default();

---

B)

let body = body
        .collect::<Vec<_>>()
        .await
        .into_iter()
        .filter_map(|result| result.ok())
        .collect::<Vec<Bytes>>()
        .concat()
        .into();

The question becomes is this the best use of the new API to convert all
stream chunks into a HTTTP Body while minimising the number of copies/clones.

My current thoughts into_iter() is not free !

and

collect.cocat().into() is cumbersome.

[martinfrances107]

Just making notes for myself in public

There is good discussion here about the various ways or iterating over a Result type. The strategies to handle the error type.

https://doc.rust-lang.org/rust-by-example/error/iter_result.html#fail-the-entire-operation-with-collect

[martinfrances107]

Cool, we now have a base line -- a change that passes tests

I am still worried about the performance implications here.

Regarding clones and copies .. I am not sure this is bad, indifference or a good change

Please be aware I am still thinking about this ... I will ask on discord in the morning if there are any standard ways the gauge a suspected performance regressions.

[martinfrances107]

To justify that last change

hyper is a low level API and we are now accessing that functionality through the viz module, or through the axum module. Modules on which we directly depend.

From a maintenance perspective this is a good change, less dependencies to track.

[dgsantana]

You can check my PR #2082, it moves everything to http 1.0 and axum 0.7. It uses the http-body-util to get the bytes in a easier way. Right now all examples, except for the one which depends on axum_js_fetch is converted. This examples also requires an update on gloo-net, for which I made a PR, let's see if it gets merged.

PS: For Viz, you can't move to http 1.0, because Viz hasn't update yet to hyper 1 and http 1. We need to wait a bit. Maybe make a PR for it?

[martinfrances107]

Thanks, StreamBody and BoxBody -- leads to a much cleaner solution

PS: For Viz, you can't move to http 1.0, because Viz hasn't update yet to hyper 1 and http 1. We need to wait a bit. Maybe make a PR for it?

I will help out on that where possible.