Hacker News Reader

Comment by doodlesdev | Hacker News Reader

I must say the proposal for Hologram is extremely interesting. The website is made using Hologram, and it speaks loudly to how good it is: every page navigation is (practically) instantaneous. Compared to other approaches such as LiveView, it's pretty good.

The initial page load isn't impressive, though: Google's PageSpeed Insights indicates a 100+kb runtime with lots of unused JavaScript initially, resulting in a LCP of 1.5s (results will vary, of course). I wonder how much of the JavaScript is simply code that stores the website pages, haven't had the time to look at this in detail yet.

For a docs website, that's excessive and bloated; it'd be much better to just deliver no JS and provide HTML with prefetching rules and cache headers (which would also provide instant navigation and offline support).

I'm happy they made the docs website with it, though, to dogfeed and showcase it.

View on HN

That's odd - I've tested the site multiple times previously and consistently got LCP results between 0.12-0.40s on large pages, which I consider very good results. I just tested it again on a large page using the Google Chrome Performance tab and got 0.14s LCP. I'm testing on the production site, connecting from Warsaw.

The 1.5s LCP you're seeing is quite different from what I'm observing. Performance can vary significantly based on location, network conditions, and device capabilities, so perhaps that's contributing to the difference in our results?

It's also possible that different pages have different performance characteristics depending on their content complexity and the navigation structure. Could you share which specific page you tested? That would help me understand the discrepancy better.

Regarding the runtime size - it hasn't been optimized at all yet, so I expect it will be much smaller in the future. I wouldn't be surprised if we can reduce it by a few times through proper optimization.

You're absolutely right that for a pure documentation site, a no-JS approach with prefetching would be lighter. The Hologram docs site is indeed dogfooding - I wanted to showcase the framework's capabilities and stress-test it with real-world usage, even if it means some overhead for this particular use case.

The goal was to demonstrate the instant navigation experience you mentioned, plus features like client-side search and interactive examples in the future that benefit from the stateful client-side architecture.

Thanks for taking the time to test the site and provide this feedback!

The LCP I see is from Google PageSpeed Insights, meaning it's not benchmarked on my machine! In fact, in my setup I'll get much faster load times, but that performance level likely won't be representative of my users (I have a pretty good fiber connection till the last meter, I use Ethernet to connect, and I have a pretty beefy development workstation). The numbers you will see there aren't exactly comparable to the Lighthouse benchmarks you will perform locally: 1.5s LCP is still pretty decent, considering the average website speed nowadays, just not what I would expect from a static docs website, if that makes sense.

The page I tested is the one linked here in HN (the announcement)! I guess I should've referenced the PageSpeed Insights URL [0]. I hope you can get your bundle sizes smaller in the next versions, because honestly, from my first look (not very in-depth, of course), it's the only thing that's not very attractive about it. Likewise, I keep in mind, though, that most web frameworks nowadays ship very large bundles to achieve hydration and client-side routing.

I'm excited to see future updates to this project!

[0]: https://pagespeed.web.dev/analysis/https-hologram-page-blog-...

Thanks! I looked at the PageSpeed Insights report you linked. Apart from the bundle size (which will definitely be optimized) and bundles caching, other performance issues seem to be related to the page content itself - specifically some analytics scripts and a large animation GIF that's contributing to the load time.

I do notice that the report shows 104 KiB transfer size for the runtime bundle, which I think is actually a relatively reasonable result for a modern web framework, though there's certainly room for improvement. The bundle size will definitely decrease substantially in future versions as I focus on optimization.

Also worth noting that the bundles aren't being cached properly yet, which is another thing the framework should handle automatically - that alone would make a significant difference for repeat visits. This one's an easy fix though.

Everything you've mentioned is noted and will be addressed! Thanks again for taking the time to run the analysis and share the specific results - having concrete data like this is really valuable for prioritizing improvements.

Reply on news.ycombinator.com