Hacker News Reader

Comment by d4mi3n | Hacker News Reader

While I agree NPM bloat is a problem, this doesn't seem like too unreasonable an approach to publishing a la cart libs like these. I haven't looked closely at the repo itself, but code generation through build/release scripts is commonly used to manage the issues you describe.

EDIT: Unless I read the source code wrongly, it seems like they actually used the umbrella repo to collect docs and links to all the other components which is the opposite of what I thought was going on here.

View on HN

> this doesn't seem like too unreasonable an approach to publishing a la cart libs like these

Yes! Yes it is unreasonable. Each component is less than 10KB is source code, the majority of the repository is scaffolding and configuration, which is duplicated for each repository/component. Instead of just having one repository with all the components.

I'd understand the approach if you were dealing with MB big components. But these are not that...

> code generation through build/release scripts is commonly used to manage the issues you describe

Yeah, of course you can always fight bloat/complexity by adding more bloat/complexity, but normally we (developers) only do that when we're lacking time. If we do have time, we try to get rid of it instead, something the GitHub developers definitely should have time for by now.

Each of these components stands on their own. You can use one without having to link to or consume any of the others. Shouldn't each one have documentation that describes how they work, a license, etc? I don't want to have to go hunt down some separate documentation project, which is almost certainly years out of date, to figure out how to use these components. If someone is consuming or even contributing to one of these components everything they need is in front of them in one repo.

I'm really scratching my head to understand your concern here. Apparently there is some magical minimum bar of file size that a project needs to pass before it's worth having a repository on github with a readme, license, etc? Even though the git repo and a few text files only take a few bytes to store?

I think it's done to decouple versions. Imagine component A updating its major, causing the whole suite to update its major. But that higher major might also have newer versions of components B and C, but you don't want to upgrade your usage of these.

There are cohesive component kits versioned as whole package, but I never liked that to be frank. I think this is good evolution, just messy because it's new. I'm sure better tooling will eventually emerge.

You know what you could also do? Don't break backwards compatibility. I know it sounds strange (if you're coming from modern JS development) but many ecosystems actually pull that off, like the entirety of Windows, or 99% of the ecosystem in Clojure/Script.

Then if you have over the years collected a bunch of incompatible changes that you would like to do, you create a new library with a new name, and do all the changes there.

Now you have the first library absolutely never breaking compatibility, and the second one for your "I absolutely must change the API interface here" changes.

There are some good ideas around versioning here as well (by Rich Hickey): https://www.youtube.com/watch?v=oyLBGkS5ICk

> just messy because it's new. I'm sure better tooling will eventually emerge.

I don't understand this. Either a new solution is better and we try to use that, or the solution is worse and we should avoid it. If it's worse, why should we build better tooling when we can just use a good solution from the get-go?

> I don't understand this. Either a new solution is better and we try to use that

From my perspective coming from my experience both as a component kit developer and an application developer using that kit, having one component per versioned package is good. That we live in a world where larger packages are prevalent and thus tooling forces a developer to duplicate configuration everywhere is an unfortunate coincidence that doesn't take away the benefits of separating the project into one versioned package per component.

Also consider that it might not even be because of the development tooling, but rather VCS or CI/CD related.

> I think it's done to decouple versions.

From the repo's README:

> We have 16 open source custom elements:

It's feasible that GitHub also have some components that are temporarily or permanently private. Having one repo per component may help to remove the need for that distinction among GitHub developers, who don't have to first check whether to import a component from a public or internal repo.

Having a consistent version number gives you some reassurance that your components are compatible with each other.

I think this should have really been a monorepo myself. Would allow them to do some code sharing in that respect as appropriate.

Though, monorepos aren't everyone's 'taste' and I can understand why standalone repositories make some sense in the mental model where you want everything de-coupled and easy to distinguish

I suppose it makes sense if you're issuing semantically versioned releases of each module and you want the GitHub releases to match, but the trend of turning everything into its own module until there's nothing left can be very frustrating. This seems very common in JavaScript and Rust.

Rust at least compensates for it in its generated docs. PureScript OTOH... Sometimes modules are literally just one type or one function, and the docs don't consolidate them. Whyyyyyyyy.

Reply on news.ycombinator.com