Enable wee-alloc by default for web-sys feature

[jstarry]

Is your feature request related to a problem? Please describe.

wee_alloc helps decrease the wasm binary size of yew apps but is not compatible with stdweb. When yew supports web-sys, it would be nice to enable it by default as most users would presumably want it.

[davidkna]

Has anyone tested how this affects performance and memory usage? I assume it'll both be worse since wee_alloc aims to be very simple.

[daxpedda]

Has anyone tested how this affects performance and memory usage? I assume it'll both be worse since wee_alloc aims to be very simple.

I did not test it, but from the description I think it will, but the default wasm allocator doesn't seem to be very performant either according to their readme.

I just tested it and it's not possible to overwrite a global allocator once it's defined, so the user would have to turn on no-default-features.

Personally I am alright with this.

EDIT: Actually forgot to add the links.

[kellytk]

@davidkna IIRC when creating https://github.com/yewstack/yew-wasm-pack-minimal I evaluated wee_alloc and didn't observe sufficiently improved performance to use it. Having been several months that may be obsolete.

I would be interested to see a comparison for each of Yew's examples.

I propose that performance optimization issues be tagged as such and deferred until #555 is implemented. Consistent and automated benchmarking infrastructure is very helpful to optimization. (early regression detection, false positive avoidance, developer experience)

[daxpedda]

The reason to use wee_alloc is not performance, but binary size.
I have tested this in multiple projects already and the difference is significant.

[kellytk]

@daxpedda I include that with it and I still believe optimizations shouldn't occur outside of more rigorous infrastructure.

[daxpedda]

I don't disagree, just pointed out that it's not about run-time performance.
On a side note, are we going to test for binary size in our CI too? 😮

[kellytk]

@daxpedda It would be difficult to accurately gauge the performance impact of changes otherwise.

[jstarry]

On a side note, are we going to test for binary size in our CI too?
Yeah, let's track binary size in CI as well.

I agree with @davidkna, it's a bit presumptuous to assume that devs care more about binary size than performance. Let's not enable it by default, we can instead rely on our docs to inform users: https://yew.rs/docs/advanced-topics/optimizations#wee_alloc

[jstarry]

I added an issue for binary size tracking here: #858

[kellytk]

@jstarry

it's a bit presumptuous to assume that devs care more about binary size than performance

For a scenario where software's assets are transferred across a network to the client at runtime, I classify binary size as an aspect of performance. For example, it may be worth a larger binary size, and associated additional initial load delay, for an ongoing reduction in CPU overhead of a complex and data-intensive SPA.