Use web_sys::Url for IDN conversion on wasm32-unknown-unknown

micolous commented 7 months ago

This pull request is based on #886. Merging this PR will also merge #886.

This removes the dependency on idna on wasm32-unknown-unknown, replacing it with a small shim which calls out to web_sys::Url.

With a small yew demo app (which calls Url::parse()), this reduces binary size by 228 KiB on a release build and by 425 KiB on a debug build. These savings should go a lot of the way towards addressing #557, while retaining IDN support.

API changes

The deprecations could be split out from this, I've included them as related work.

Deprecate quirks::domain_to_unicode(): https://github.com/whatwg/url/issues/63

Removed on wasm32-unknown-unknown because browsers never implemented it.
Deprecate Origin::unicode_serialization(): #881

Removed on wasm32-unknown-unknown because it requires domain_to_unicode.
Deprecate quirks::domain_to_ascii(): https://github.com/whatwg/url/issues/63

Kept on wasm32-unknown-unknown, because its implementation does not depend on browser functionality.

Alternatives considered

Downstream libraries could adopt web_sys::Url on wasm32-unknown-unknown.

This is API-breaking, and complicated to handle, for reasons I've covered in #557 and https://github.com/seanmonstar/reqwest/issues/2050.

There are similar pains discussed in https://github.com/sagebind/isahc/issues/382 for a http::Uri to url::Url migration.
Shimming out every single piece of functionality to JavaScript's Url type (via web_sys::Url) on wasm32-unknown-unknown.

This would probably make things smaller again, but there are enough subtle API differences to make this a problem. For example, url::Url::host() returns a url::Host, but web_sys::Url::hostname() returns a String.
Shimming in idna.

Not all the functionality required is supported in web APIs (domain_to_unicode), and so it needs changes to Url anyway.

This also allows one to continue using the idna crate directly on wasm32-unknown-unknown with its normalisation database.

Known issues

There are a few extra test failures (see expected_failures_{chromium,firefox,safari}.txt) – I'm not certain which of these are browser bugs or url::Host bugs.

In any case, I don't think I can fix them, and I don't know how much one should care.
I haven't tried this on redox or wasi targets – they don't appear to be covered by CI, and so I've left them as-is.

It may be possible to get similar binary size savings on those targets, but I'm keeping them out of scope for now.

micolous commented 3 months ago

While this is based on #886, I've marked it as "ready for review", as this isn't really a work in progress – it's just blocked.

codecov[bot] commented 2 months ago

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

:exclamation: No coverage uploaded for pull request base (main@a4dd58b). Click here to learn what that means.

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #887 +/- ## ======================================= Coverage ? 81.88% ======================================= Files ? 20 Lines ? 3545 Branches ? 0 ======================================= Hits ? 2903 Misses ? 642 Partials ? 0 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

servo / rust-url