Use Brotli instead of gzip

[fzzle]

Let's optimize for Brotli size:

• It's supported by most modern browsers (and not IE 11).
• It performs better than gzip.

[rschristian]

The key statistic there is not how many browsers support Brotli, but how many sites are distributing it. Browsers only consume what they're given, so we'd need to take a look at what's actually being given.

According to this W3Techs report, as of the 17th of May, 2021, only 23% of the world's top 10 million sites use Brotli. I'm not sure how reliable that report is, so take it with a pretty big grain of salt, but that sounds about right with my expectations.

Not that I imagine optimizing for Brotli will hurt Gzip size in any great way, but ignoring Gzip for Brotli, which is a small-ish minority at the moment, seems odd. Switching from Gzip here would mean size regressions for 77% of the world might go unnoticed.

[fzzle]

The key statistic there is not how many browsers support Brotli, but how many sites are distributing it. Browsers only consume what they're given, so we'd need to take a look at what's actually being given.

According to this W3Techs report, as of the 17th of May, 2021, only 23% of the world's top 10 million sites use Brotli. I'm not sure how reliable that report is, so take it with a pretty big grain of salt, but that sounds about right with my expectations.

Not that I imagine optimizing for Brotli will hurt Gzip size in any great way, but ignoring Gzip for Brotli, which is a small-ish minority at the moment, seems odd. Switching from Gzip here would mean size regressions for 77% of the world might go unnoticed.

I'm not sure that I would consider that a key statistic. Brotli is new - most websites aren't. A key statistic would be the % of websites created in 2021 that use Brotli or the % of Preact websites that use Brotli. I think that'd be a very different number.

A key selling point of Preact is its Brotli size ("Fast 3kB alternative..."). It'll be easier to reach 2kB Brotli size with a focus on Brotli size. 🙂

[developit]

This is a tricky one: while Brotli is certainly the more logical choice from a purely technical standpoint, changing the compression algorithm we use to track size invalidates all prior measurements.

I think the only way we can do this would be to report both brotli and gzipped sizes for a period of time, eventually moving to Brotli-only when there are fewer long-term PRs (such as when the current rewrite lands and has had a minute to settle).