What's the tradeoff?

[mxstbr]

The readme and website make a bunch of extraordinary claims ("same api, much faster, better browser compat, smaller bundle size"), but what are the tradeoffs of choosing Nerv vs vanilla React? What do I miss when I choose Nerv?

Note: Not trying to be a pain in the butt here, genuinely curious why React wouldn't just take this code base and make it their own

[benjamn]

Just to underscore @mxstbr's point, you lose nothing by documenting the tradeoffs honestly. If your users have to discover those tradeoffs themselves, they won't trust the project anymore, and they will have a very easy time switching back to React, given the API compatibility.

[petermikitsh]

I'd like to see this project runs against the official React test suite, to have a quantifiable measure of knowing how closely this implementation matches official API behavior. If it passed 100%, and offered superior performance / reduced file size, I would be excited to use this.

[dzek69]

I think no 10kb-or-so-react-alternative would have async per frame rendering like React 16 have. Splitting rendering into job chunks that fits 16ms frame isn't that easy. This probably the biggest trade off of every React alternative now. Until R16 it wouldn't be much problem.

[yuche]

Sorry for the late response.

I would love to add tradeoffs to README. In fact, I have been make a public talk in China to explain How and Why we build a another React-like library, there is also a Chinese article in this repo's wiki posted weeks ago: 《为什么我们还需要一个 React like 框架？》 (In English: Why we need another React-like library?), which compares several React-like library(preact, Inferno), React itself and Nerv.

But as you can see, I'm not good at English, translations takes time. After I finish README.md and home page translation last night, just feel exciting (it was not easy for me to do this) and submit to Hacker news and reddit.

I apologize for the misunderstanding for some words in README, later there might be some change after we consult with some colleagues who is good at English. For now, there is some background about React, React-like library and Nerv:

React

The bottom line is: You should use React while you can use React.

Just forget about all React-like libraries (including Nerv) advertising about how fast they are, the key of performance is about the guy who write the code, not about the library or framework. If performance really that matters to you, you should consider manipulate DOM directly like VSCode did.

If you like the idea of React, but can’t use it for some reason (like Apache, or IE8, size), you should consider choose a React alternative:

preact

Preact is the most popular React-like library, as they describe: got everything your need. The author Jason Miller is very active in GitHub, maintaining the repo. It looks perfect for a React replacement. But there still some problem we don’t satisfy.

First, preact just so obsessed about the size, every PR must be small, some (we think) important feature/fix would’t merge because it takes too many code. We fork a preact in the beginning for internal project because preact would’t allow the size of it over 3Kb gzipped.

And there is performance issue. If just use preact alone that’s fine, even faster than React in several benchmark. But in our scenario, we use third-party React library a lot, so preact-compat is required. preact-compat will transform a preact element to a React compatible Element (such as change prototype descriptor ), there is huge impact in performance, you can test by aliaing preact to preact-compat in any benchmark.

Inferno

Inferno is my favorite React-like library, when the company decide make our own React-like library (at the time React is not MIT license), I was thinking: “Inferno is the way how I implementing a React”. To be honest, Nerv learn a lot from Inferno, diff algorithm, event delegation, etc. Thanks to @trueadm and his great work, if wasn’t for Inferno, Nerv couldn’t make so far today.

As you can imagine, Inferno also can not be satisfied with our requirements. It just because Inferno doesn’t wanna be a React replacement, they has their own ecosystem, such as inferno-router . As I assume, they just want a extremely fast library, React-like is a bonus point. To achieve the target, there will be some compromise, such as you can’t use react-router in Inferno, because React will continue diff when the type, props, context and the reference of old vnode and new vnode are the same, instead, Inferno just quit diff when old vnode and new vnode shared a same reference.

Nerv

In my humble opinion, The biggest difference between preact, Inferno, and Nerv, is not some technical issue like what’s the right way to implementing a React feature. It is about the what does the library want to achieve. In preact, maybe they just want a lite library, Inferno want to be fast as they can, React compatibility just a bonus, they are both need a compat module to do that. But for Nerv, compatible with React is our main goal, by doing that, we can sacrifice performance and size.

Peter Mikitsh’s criticism is right. Nerv can’t pass React fiber(16) 100% unit test, which is predictable —— the React team takes whole year to achieve the target, how can two guys come from nowhere(says by some guy in Hacker news) be able to do that?

So, What’s the tradeoffs? Obviously, some third-party React component/library couldn’t work properly in Nerv. But which one? To be honest I don’t know. I’m glad to be heard that Next.js next-news worked on first try., but in the meantime, Benchmark tabs don't work in Firefox (fixed) —— All of this, We wouldn’t know if we don’t go public.

Rome wasn't built in a day. Nerv is not perfect, no library is, particularly in early stage, maybe there still tons of bug we don’t know. Therefore we decided to open source, to go public, we need the community’s help, we need your help. Of course you can sit there and do nothing but complain about this or that(like me doing it now), or you can submit a issue even a PR, tell us how can we make Nerv better. With community’s help, such as [feature missing] (#17) , [typo fixed] (NervJS/nerv-website#3), I believe after couple month later, Nerv will be much stronger and smarter —— Perhaps this is what open source about: we help each other, we do things together.

Hope this reply answers your question.

[mxstbr]

That is an incredible response, thank you so much @yuche for taking the time to type that out! Again, no hard feelings at all, I totally understand that language can be a big barrier. ❤️

Have a great day!

[dzek69]

Thank you for your reply. So to sum this up the main goals of Nerv are 100% compatibility, free as in freedom license and IE8 support. The trade off is not being well-tested yet and it may not pass all unit tests from React, but are you going to to achieve that as well?

Personally I think this is very interesting project and I'm soon going to publish small isomorphic webpage and I'm obsessed with file sizes as well, but preact forced me to use compatibility package even when I thought it shouldn't be necessary and I'm not satisfied with that. I'll try to use Nerv instead.

[sospedra]

Can we maybe have some numbers? Benchmarking?

[popey456963]

@sospedra There are a few benchmarks on the website main page - https://nerv.aotu.io/

[yuche]

@mxstbr
Thanks for point it out that we do really need a tradeoff description. A good question like this could be very useful to sort things out.
I going to reopen this issue. Any question or criticism is welcomed.

[luckyadam]

@sospedra Yeah, we will show the benchmarks ASAP.
You can see the results at https://nerv.aotu.io/ first.

[ColinEberhardt]

@yuche thanks for the detailed response

If you like the idea of React, but can’t use it for some reason (like Apache, or IE8, size), you should consider choose a React alternative:

On the 'like Apache' note specifically, that refers to React's previous BSD licence. I believe it has moved to MIT.