[lhorie]

I find the text a bit hard to read (and it seems at least one other commenter mentioned the compressed line-height).

What I do notice is that the site loads very fast (sub-1s w/ cache disabled via devtools to simulate first visit, and sub-50ms otherwise). That alone deserves big props.

[ddevault]

I am very proud of our performance :)

https://forgeperf.org/

[mappu]

I think Codeberg.org is not doing the (normally very fast) Gitea any favors here, because it's hosted on Hetzner in Germany. Lighthouse may only throttle the connection further.

Where are the tests being run from?

Is there any way to control for network latency (maybe subtract the ICMP rtt from the tests?)

You might consider comparing it to other Gitea instances like https://try.gitea.io/ (Digital Ocean / USA), https://mirror.git.trinitydesktop.org/ (Czech Republic) or https://gitea.com/ (China), but I guess Codeberg is probably the most representative instance for public use.

[ddevault]

This has been brought up many times, but latency and bandwidth is controlled for with Lighthouse. The same tests have also been run from Germany without appreciably different results. It's also easy to run the test suite yourself if you have an hour of compute time to spare:

https://git.sr.ht/~sircmpwn/forgeperf

The goal here isn't to compare forge software (that would require a completely different approach to level the playing field), but to compare hosted options. The test suite is easily extended with other hosts or pages, and a few people have tested against their own Gitea instance - usually it ranks somewhere just above the middle of the pack.

[kick]

Line-height on Sourcehut itself is reasonable. It's just on the blog that it's off.

[Scarbutt]

some heavy caching is happening somewhere.

[ddevault]

No caching (except for the CSS file, which your browser is probably caching), just simple pages and lightweight backends.

https://forgeperf.org

All of our pages are done in 2 requests, or 1 with a warm cache.

[Hello71]

I was going to ask how you speed up pygments so much, but apparently the answer is "actually, some stuff is cached".

[ddevault]

Ah, yes, some stuff is cached :) but it doesn't do much, we kind of cache it on principle. An example that makes more of a difference is the git SSH pipeline.