Search box on nested 404 page `abc/123` goes to 404 `abc/std/index.html?...`

[huonw]

For instance, on https://doc.rust-lang.org/nightly/abc/123, the search box offers to search the standard library for 123. Clicking the search button goes to https://doc.rust-lang.org/nightly/abc/std/index.html?search=123 (note the abc/std) which is invalid. Probably related to
#54609 (cc @GuillaumeGomez).

[kzys]

Ah, my bad. I was assuming that the 404 page wouldn't contain /...

[GuillaumeGomez]

Problem is a bit more complicated than that: we need to keep the version we're looking at (stable, beta or nightly). Not a very complicated fix but still need to be careful. I'll open the PR soon.

[QuietMisdreavus]

This got brought up again in rust-lang/www.rust-lang.org#864, so it's worth bringing @rust-lang/rustdoc and @rust-lang/docs into this thread as well.

The primary concern is which libstd search to send the user to. doc.rust-lang.org hosts documentation for every version of Rust, so it's possible someone hit a 404 while trying to browse docs for version 1.30.0or the like. Should we redirect them to stable anyway? (Or nightly, since the standard library itself makes other crates link to nightly thanks to itshtml_root_url`?) Should we try to detect which version of Rust they're browsing, and introduce a JavaScript semver parser into our 404 page? (Or have some other means of injecting the version number into the page?) There's not really a clean way to approach this, at least that i can see.

[huonw]

I think the version thing can be maintained from the current page as a "black box" slug without having to parse it? The difficulty would seem to then be deciding whether a URL is versioned (like /nightly/std/... or /1.30.0/std/...) or not (like /std/...), which can potentially be done via checking a list of the known fixed slugs or whether the first slug starts with a number. That might look something like the following (which is untested, and doesn't have appropriate error handling):

const path = new URL(window.location.href).pathname
const firstComponent = path.split('/')[0]
const fixedVersions = ["stable", "beta", "nightly"]
const includeFirst = fixedVersions.includes(firstComponent) || /^[0-9]/.test(firstComponent)

// inside the search submit:

document.location.href = (includeFirst ? "/" + firstComponent : "") + "/std/index.html?search=" ...

[Dylan-DPC]

This now redirects to this link which works as it's supposed to so closing this

[huonw]

That link https://doc.rust-lang.org/std/index.html?search=123 has lost the version (/nightly/ in the original), which is probably okay, but isn't exactly 'correct'.