perf: avoid unnecessary re-renders with the suspense mode

[koba04]

Fixes #454
Currently, SWR doesn't apply the performance optimization mentioned in https://swr.vercel.app/advanced/performance with the suspense option, and it leads to unnecessary re-renders.

I've applied the same optimization using stateDependencies for the suspense mode.
Is there any reason not to apply the optimization for the suspense mode?

This is a fork of reproduction in #454, which is based on this PR.

• https://codesandbox.io/s/swr-basic-forked-ftjio

[codesandbox-ci]

This pull request is automatically built and testable in CodeSandbox.

To see build info of the built libraries, click here or the icon next to each commit SHA.

Latest deployment of this branch, based on commit 0771899:

Sandbox	Source
SWR-Basic	Configuration
SWR-States	Configuration
SWR-Infinite	Configuration
great-lumiere-441zk	Issue #454

[shuding]

Sorry for the delay and thank you for this PR!

Is there any reason not to apply the optimization for the suspense mode?

Mainly because that Suspense is still an experimental feature, and we are not implementing it 100% correctly (mostly with Concurrent mode). Also according to facebook/react#17526 (comment), I feel that it's better to not optimize it too early, and get everyone relying on it too much.

[koba04]

@shuding Thank you! I understand your concern.
I think rendering 6 times with a normal case (const { data } = useSWR(url, fetcher, { suspense: true })) is a bit surprising and might be a problem in complex applications. IMHO, this optimization doesn't rely on Suspense, which is just avoiding unnecessary rendering, so it would be nice if the suspense option has this optimization even this is an experimental option.

[shuding]

Yeah I think I agree with applying those perf optimizations for now since they are not specifically for Suspense. Also the code is much cleaner 👍

[koba04]

Thank you!