Polled SSR Stream

[futursolo]

Description

This pull request introduces a polled SSR stream instead of using channels.

Compared to channels, Polled Streams:

• yield content before a suspended component in a single chunk and does not wait until reaching a certain capacity.
• have better performance than channel-based streams with much smaller reserved capacity (1024 vs 8192).
• resolving future is polled alongside the stream.

Checklist

• I have reviewed my own code
• I have added tests

[github-actions]

Visit the preview URL for this PR (updated for commit a78f0b3):

https://yew-rs-api--pr2824-polled-stream-x9xginht.web.app

_{(expires Sun, 11 Sep 2022 02:17:22 GMT)}

_{🔥 via Firebase Hosting GitHub Action 🌎}

[github-actions]

Benchmark - SSR

Yew Master

Benchmark	Round	Min (ms)	Max (ms)	Mean (ms)	Standard Deviation
Baseline	10	419.804	428.535	423.388	2.875
Hello World	10	1258.366	1308.084	1289.160	15.715
Function Router	10	3026.638	3176.004	3118.468	61.431
Concurrent Task	10	1011.819	1015.229	1013.623	1.109

Pull Request

Benchmark	Round	Min (ms)	Max (ms)	Mean (ms)	Standard Deviation
Baseline	10	417.519	433.946	424.418	4.352
Hello World	10	748.980	793.166	769.875	11.488
Function Router	10	2648.450	2834.495	2763.374	71.040
Concurrent Task	10	1008.841	1012.347	1011.034	1.133

[github-actions]

Size Comparison

examples	master (KB)	pull request (KB)	diff (KB)	diff (%)
boids	169.308	169.311	+0.003	+0.002%
contexts	107.335	107.335	0	0.000%
counter	84.829	84.831	+0.002	+0.002%
counter_functional	85.453	85.453	0	0.000%
dyn_create_destroy_apps	87.789	87.790	+0.001	+0.001%
file_upload	99.493	99.494	+0.001	+0.001%
function_memory_game	162.753	162.754	+0.001	+0.001%
function_router	345.724	345.643	-0.081	-0.023%
function_todomvc	157.712	157.715	+0.003	+0.002%
futures	222.151	222.155	+0.004	+0.002%
game_of_life	105.023	105.023	0	0.000%
immutable	180.612	180.610	-0.002	-0.001%
inner_html	81.929	81.929	0	0.000%
js_callback	110.685	110.688	+0.003	+0.003%
keyed_list	193.486	193.489	+0.003	+0.002%
mount_point	84.614	84.615	+0.001	+0.001%
nested_list	111.951	111.955	+0.004	+0.003%
node_refs	91.812	91.808	-0.004	-0.004%
password_strength	1547.278	1547.283	+0.005	+0.000%
portals	95.384	95.385	+0.001	+0.001%
router	315.201	315.084	-0.117	-0.037%
simple_ssr	151.356	151.347	-0.010	-0.006%
ssr_router	391.231	391.232	+0.001	+0.000%
suspense	108.385	108.389	+0.004	+0.004%
timer	87.625	87.625	0	0.000%
todomvc	139.212	139.213	+0.001	+0.001%
two_apps	85.473	85.474	+0.001	+0.001%
web_worker_fib	150.535	150.536	+0.001	+0.001%
webgl	84.398	84.395	-0.004	-0.005%

✅ None of the examples has changed their size significantly.

[ranile]

Looks good, sorry it took me a while to get to it

[ranile]

pin-project-lite maybe?

[futursolo]

However, if you already have proc-macro related dependencies in your crate’s dependency graph, there is no benefit from using this crate.

See: https://docs.rs/pin-project-lite/0.2.8/pin_project_lite/#different-no-proc-macro-related-dependencies

The primary benefit of #[pin_project] is that rustfmt will continue to work for the struct.

[ranile]

Why not do:

impl !Send for Inner {}
impl !Sync for Inner {}

[futursolo]

This is due to that negative impls is not stable.

See: https://doc.rust-lang.org/unstable-book/language-features/negative-impls.html

[ranile]

Perhaps remove the bound from here? That would mean that everywhere this type is named doesn't require the bound - only the usages do

[futursolo]

We cannot remove this as this bound is declared on the MaybeDone type itself.

[ranile]

It seems that this file was moved, rather than deleted. git mv would be better in that case since that preserves the git history

[futursolo]

fmt is a complete rewrite of the buffer and not connected to the previous implementation.

I tried to interactively rebase the changes and is having conflicts with every commits after the commit that removed io.rs.

[ranile]

Sorry, I didn't see the review notifications. Looks good!