Implement WHATWG URL spec change for Windows drive letter paths (whatwg/url#874)

[renezander030]

Summary

Implements whatwg/url#874: in the URL parser's scheme start state, <ASCII alpha> : \ is recognized as a Windows drive letter path and parsed into a file:///<drive>:/... URL.

Url::parse(r"C:\path\file.txt")?;        // file:///C:/path/file.txt
Url::parse(r"D:\foo\bar.exe")?;          // file:///D:/foo/bar.exe
Url::parse(r"c:\folder\file.txt")?;      // file:///c:/folder/file.txt  (case preserved)

Conservative scope — only the backslash shape triggers the conversion. Forward-slash drive paths (c:/foo) are intentionally not rewritten because <alpha>:/ is ambiguous with single-letter scheme URLs (c: scheme, a://example.net, h://., w://x:0). Tests guard against regressing those.

Status: draft — pending whatwg/url#874 merge

The spec PR is still open (last activity 2025-11-28). Opening this as a draft so the implementation can be reviewed in parallel and land quickly once the spec merges. The spec PR cites multi-implementer interest (Ladybird, jsdom/Node, Chromium, Gecko, WebKit, Deno).

Context: a Deno-side port (denoland/deno#33097) was redirected upstream to rust-url by @crowlKats — "we would rather have this land in rust-url even if we have to wait a bit." This PR is that upstream landing.

Implementation

url/src/parser.rs:

• New helper starts_with_windows_drive_letter_path peeks 3 chars for <alpha>:\ (placed alongside the existing starts_with_windows_drive_letter_segment family).
• New method parse_windows_drive_letter_path runs the spec's path state directly: pushes file://, sets empty host, dispatches to parse_path with the original input. parse_path handles \ → / normalization for special schemes, producing /C:/path/file.txt.
• parse_url checks the new helper before parse_scheme, short-circuiting only when the pattern matches.

The Windows-drive shortcut deliberately drops any base URL — the spec's scheme start state ignores base when scheme is set, and <alpha>:\ is unambiguously absolute.

Tests (url/tests/unit.rs)

8 new unit tests:

• windows_drive_path_basic — C:\path\file.txt → file:///C:/path/file.txt
• windows_drive_path_different_drives — D:, Z:
• windows_drive_path_preserves_drive_case — lowercase + uppercase preserved
• windows_drive_path_mixed_separators — \ and / interchangeable in body
• windows_drive_path_percent_encodes_spaces — path encoding still applies
• windows_drive_path_drops_base — base URL ignored when shortcut fires
• windows_drive_path_with_query_and_fragment — ?q=1#frag flows correctly
• windows_drive_path_does_not_rewrite_scheme_urls — regression guard for c:/foo, a://example.net, h://., w://x:0

Full suite (cargo test -p url) and WPT (cargo test --test url_wpt -p url) pass with no regressions.

Test plan

• All existing unit tests pass (74/74)
• All existing doctests pass (67/67)
• WPT suite passes (urltestdata.json)
• cargo fmt --check clean
• cargo clippy --all-targets clean
• WPT test vectors will be added to urltestdata.json once web-platform-tests/wpt#53459 lands and the spec PR merges

[mrobinson]

@renezander030 Did you use generative AI to produce the code in this pull request or the PR description?

[renezander030]

Yes. I used Claude (an LLM) to help draft both the code and the PR description, working from my earlier Deno-side PR (denoland/deno#33097) that @crowlKats redirected here.

The substance is mine: the conservative scope (backslash-only, not forward-slash drive paths) and the regression guards for c:/foo, a://example.net, h://., w://x:0 came directly from issues I hit during the Deno review. The full test suite, WPT suite, cargo fmt --check, and cargo clippy --all-targets all pass locally.

Happy to comply with whatever policy Servo / rust-url has on AI-assisted contributions — if you'd prefer I rewrite the PR description by hand, or if there's a contributor attestation you'd like me to add, please let me know.

[mrobinson]

Here's Servo's AI use policy: https://book.servo.org/contributing/getting-started.html#ai-contributions

It's fine to use LLMs to help understand an issue, find bugs, or to plan / architect a change, but all code, issue text, and PR text must be your own (apart from translation and transcription).

[renezander030]

Here's Servo's AI use policy: https://book.servo.org/contributing/getting-started.html#ai-contributions

It's fine to use LLMs to help understand an issue, find bugs, or to plan / architect a change, but all code, issue text, and PR text must be your own (apart from translation and transcription).

I decline. Closing this PR now.