feat: replace C++ FFI with native Rust parser by countradooku · Pull Request #119 · ada-url/rust

countradooku · Jul 29, 2026

Summary

replace the bundled C++ implementation and C ABI wrapper with a memory-safe native Rust parser
retain the existing ada-url public API, no_std support, serde behavior, and the former bundled / libcpp feature names as compatibility no-ops
use a single normalized buffer with compact component offsets and conservative optimized fast paths
add URLPattern behind an optional feature and add Radu Diaconu as an author
pin and exercise the current ada-url/ada fixtures at 30f3f3020c5a979b62f90dc9c37fd45de3cc84d7

Impact

This removes build.rs, the C/C++ FFI layer, and the vendored Ada sources. Most of the apparent diff size is the deletion of roughly 27,000 lines from ada.cpp and ada.h; the native implementation is roughly 5,000 lines of Rust.

The packaged crate verifies successfully at 132.7 KiB compressed. Existing benchmark targets remain available, and smoke runs cover both the focused parser corpus and all 891 WPT URL entries.

Conformance

The test suite now directly covers:

URL parsing and base resolution
transactional setters
the latest IDNA/UTS Test script fails on macOS #46 data
percent encoding
DNS-length validation
URLSearchParams
URLPattern WPT fixtures

The current fixtures are recorded in tests/wpt/UPSTREAM.

Validation

rustup run 1.96.1 cargo test --all-features --locked
rustup run 1.96.1 cargo test --no-default-features --locked
cargo clippy --all-targets --all-features --locked -- -D warnings
RUSTDOCFLAGS='-D warnings' cargo doc --all-features --no-deps --locked
cargo fmt --all -- --check
cargo package --allow-dirty --locked
cargo bench --bench parse -- --quick
cargo bench --bench wpt -- --quick

anonrig · Jul 29, 2026

-      # instrumented. LeakSanitizer is enabled by default under ASan on Linux;
-      # this catches regressions like https://github.com/ada-url/rust/issues/101
-      # where the failed-parse path leaked the allocation from `ada_parse`.
-      - name: Test under AddressSanitizer + LeakSanitizer


Can we keep the workflows?

anonrig · Jul 29, 2026

      if: matrix.env.CARGO_BUILD_TARGET == 'wasm32-wasip1'
      run: |
        rustup target add wasm32-wasip1
-        curl -LO https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-33/wasi-sdk-33.0-x86_64-linux.deb


We should keep this and make sure it works with wasm etc...

codspeed-hq · Jul 29, 2026

Merging this PR will regress 1 benchmark

⚠️

Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 6 improved benchmarks
❌ 1 regressed benchmark
✅ 9 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	`BASE`	`HEAD`	Efficiency
❌	Simulation	`ada_url`	3.7 µs	4.9 µs	-24.91%
⚡	Simulation	`ada_url`	4.8 ms	2.3 ms	×2.1
⚡	Simulation	`ada_url`	14.4 µs	7.5 µs	+91.06%
⚡	Simulation	`ada_url_parse`	165.6 ms	98.5 ms	+68.13%
⚡	Simulation	`ada_url`	830.7 µs	515.3 µs	+61.22%
⚡	Simulation	`ada_url`	3.4 ms	2.6 ms	+33.02%
⚡	Simulation	`ada_url`	24.8 µs	22.2 µs	+11.75%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

_{Comparing countradooku:feat/native-rust-port (6269893) with main (af35063)}

countradooku · Jul 29, 2026

i think I need to tae a look at CodSpeed Performance Analysis

countradooku · Jul 29, 2026

Performance follow-up is pushed in 6269893.

Side-by-side release benchmarks against af35063 on the same machine (lower is better) now show the native Rust implementation ahead across every measured group:

basic href: 804 ns vs 955 ns (~16% faster)
basic can_parse: 270 ns vs 438 ns (~38% faster)
BBC href: 706 ns vs 759 ns (~7% faster)
BBC can_parse: 256 ns vs 279 ns (~8% faster)
IPv4 decimal: 348 µs vs 369 µs (~6% faster)
IPv4 non-decimal: 313 µs vs 553 µs (~43% faster)
DNS: 76.2 µs vs 91.7 µs (~17% faster)
URLSearchParams: 11.2 µs vs 14.4 µs (~22% faster)

Validation is green locally: all features, no-default-features, full WPT/IDNA/URLPattern corpora, Clippy, docs, Miri, package verification, and WASI compilation. I also fixed the pinned-toolchain Miri invocation.

The new GitHub Actions/CodSpeed runs are currently marked action_required with no jobs; they need maintainer approval because the fork PR changes a workflow file.

anonrig · Jul 29, 2026

+      - run: cargo clippy --all-targets --all-features --locked -- -D warnings
+
+  miri:
+    name: Miri


what is Miri?

anonrig · Jul 29, 2026

        - os: ubuntu-latest
          env:
            CARGO_BUILD_TARGET: wasm32-wasip1
-            CARGO_TARGET_WASM32_WASI_RUNNER: /home/runner/.wasmtime/bin/wasmtime --dir=.


can we keep this

anonrig · Jul 29, 2026

   "LongYinan <github@lyn.one>",
-   "Boshen <boshenc@gmail.com>"
+   "Boshen <boshenc@gmail.com>",
+   "Diaconu Radu-Mihai <52667211+countradooku@users.noreply.github.com>"


is this the email you want to keep?

anonrig · Jul 29, 2026

 repository = "https://github.com/ada-url/rust"
 homepage = "https://ada-url.com"
 license = "MIT OR Apache-2.0"
-links = "ada-url"


why did we remove this

anonrig · Jul 29, 2026

+url-pattern = ["dep:urlpattern", "std"]

 [dependencies]
+idna = "1.1.0"


we shouldn't have any dependency at all

anonrig · Jul 29, 2026

 [dependencies]
+idna = "1.1.0"
+memchr = "2.8.0"
+percent-encoding = "2.3.2"


please don't use a dependency for this

percent encoding?

ada doesn't have any dependencies. we should keep the same.

Can i keep idna?

anonrig · Jul 29, 2026


 [dev-dependencies]
 criterion = { package = "codspeed-criterion-compat", version = "4", default-features = false, features = ["cargo_bench_support"] }
-url = "2" # Used by benchmarks


please keep this

countradooku · Jul 29, 2026

The final CodSpeed regression is addressed in 7f829ba.

BBC_BasicBench_AdaURL_CanParse now validates simple HTTPS authorities in one bounded pass instead of rescanning for backslashes, credentials, ports, numeric-host rules, and domain bytes. Same-machine release results:

native Rust before: ~256 ns
native Rust after: ~98 ns
upstream baseline: ~296 ns

That is ~62% lower than the previous native result and ~67% lower than upstream. The mixed BasicBench_AdaURL_CanParse result is also ~59% below upstream. Full tests, no-default-features, WPT plus DNS-length can_parse parity, Clippy, formatting, and Miri pass.

The new CI and CodSpeed runs are awaiting maintainer approval again because this is a fork PR.

feat: replace C++ FFI with native Rust parser

40937ed

countradooku marked this pull request as ready for review July 29, 2026 00:49

anonrig reviewed Jul 29, 2026

View reviewed changes

perf: eliminate native parser regressions

6269893

anonrig reviewed Jul 29, 2026

View reviewed changes

perf: streamline simple URL validation

7f829ba

Search code, repositories, users, issues, pull requests...

Uh oh!

Conversation

countradooku commented Jul 29, 2026

Summary

Impact

Conformance

Validation

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

codspeed-hq Bot commented Jul 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will regress 1 benchmark

Performance Changes

Uh oh!

countradooku commented Jul 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

countradooku commented Jul 29, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

countradooku commented Jul 29, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

codspeed-hq Bot commented Jul 29, 2026 •

edited

Loading

countradooku commented Jul 29, 2026 •

edited

Loading