Docs are slow to load on slow internet

[mb64]

The generated documentation currently loads a 10-megabyte data.js file with all the data on startup. On my home internet, this typically takes ≥20 seconds to load. (Firefox says it took 22 seconds just now.) While the browser cache definitely helps, it would be good to not require transferring a 10-megabyte file when documentation is loaded.

Short of restructuring how generated documentation works, compression helps a lot. (gzip reduces size 10x.) This might be as easy as tweaking some settings on the ziglang.org server.

[zigazeljko]

This might be as easy as tweaking some settings on the ziglang.org server.

ziglang.org is hosted behind CloudFront, which has automatic gzip support, but only for files under 10MB (which the current data.js exceeds), so we would need a workaround for that.

[mrakh]

This might be as easy as tweaking some settings on the ziglang.org server.

ziglang.org is hosted behind CloudFront, which has automatic gzip support, but only for files under 10MB (which the current data.js exceeds), so we would need a workaround for that.

Why not just compress data.js on the server side? CloudFront should pass the compressed payload along as-is as long as the server sets its own Content-Encoding header in the response.

As a side note, compressing data.js with brotli instead of gzip brings the compressed size down from ~1 MB to ~800 KB.

But even that is a temporary measure. Eventually, the docs are going to need detailed descriptions and maybe even example code snippets, which will blow up the size of data.js beyond what is acceptable. Some kind of pagination needs to be set up to make the docs load fast for the less-endowed Internet users, ideally before the 1.0 release.

[pascaldekloe]

All data in data.js is on a single line (var zigAnalysis={…). Less(1) and more(1) from a Mac choke on it. 😄

Also, documentation builds change this mega-line every time due to the random ordering of JSON fields. This is the full diff between two runs:

<                 "zuri": 0,
<                 "std": 1
---
>                 "std": 1,
>                 "zuri": 0

[tau-dev]

Loading all of the docs at once was a conscious design decision to facilitate snappy search and navigation at the cost of up-front load time. Given that the total download size is now down to 2.6 MB, I don't think a secondary site with an incrementally-loading design is currently worth the substantial additional effort.
Also note the useful zig std command (#22360).