fix memory leak on invalid urls

.github/workflows/ci.yml

@@ -65,6 +65,40 @@ jobs:
           RUSTDOCFLAGS: '-D warnings'
         run: cargo hack doc --feature-powerset
 
+  sanitizers:
+    name: AddressSanitizer + LeakSanitizer
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+
+      # AddressSanitizer (which bundles LeakSanitizer) requires nightly. We use
+      # clang for the bundled C++ so it shares the LLVM sanitizer runtime that
+      # rustc links into the test binaries. Note: `cargo +nightly` is used
+      # explicitly because rust-toolchain.toml pins a stable channel that would
+      # otherwise override the default toolchain in this directory.
+      - name: Install nightly toolchain
+        run: rustup toolchain install nightly --profile minimal
+
+      - uses: Swatinem/rust-cache@v2
+        with:
+          shared-key: sanitizers
+          save-if: ${{ github.ref_name == 'main' }}
+
+      - run: rustup +nightly show
+
+      # `--target` is required so build scripts / proc-macros (host) are not
+      # 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
+        env:
+          CC: clang
+          CXX: clang++
+          ADA_SANITIZE: address
+          RUSTFLAGS: '-Zsanitizer=address'
+          ASAN_OPTIONS: 'detect_leaks=1'
+        run: cargo +nightly test --tests --target x86_64-unknown-linux-gnu
+
   format:
     name: Format
     runs-on: ubuntu-latest

Cargo.toml

@@ -6,7 +6,7 @@ authors = [
    "LongYinan <github@lyn.one>",
    "Boshen <boshenc@gmail.com>"
  ]
-version = "3.4.4"
+version = "3.4.5"
 edition = "2024"
 description = "Fast WHATWG Compliant URL parser"
 documentation = "https://docs.rs/ada-url"

build.rs

@@ -194,5 +194,19 @@ fn main() {
         }
     }
 
+    // Optionally compile the bundled C++ (ada) with a sanitizer. This is used
+    // by the AddressSanitizer CI job so that FFI allocations (e.g. the result
+    // `ada_parse` always heap-allocates) are instrumented and leaks/errors
+    // inside ada itself are caught. Enable with `ADA_SANITIZE=address` and a
+    // matching `RUSTFLAGS=-Zsanitizer=address`; use a clang toolchain (CC/CXX)
+    // so the C++ shares the LLVM sanitizer runtime that rustc links in.
+    println!("cargo:rerun-if-env-changed=ADA_SANITIZE");
+    if let Ok(sanitizer) = env::var("ADA_SANITIZE")
+        && !sanitizer.is_empty()
+    {
+        build.flag(format!("-fsanitize={sanitizer}"));
+        build.flag("-fno-omit-frame-pointer");
+    }
+
     build.compile("ada");
 }

src/lib.rs

@@ -230,6 +230,11 @@ impl Url {
         if unsafe { ffi::ada_is_valid(url_aggregator) } {
             Ok(url_aggregator.into())
         } else {
+            // `ada_parse`/`ada_parse_with_base` always allocate a result on the
+            // heap, even when parsing fails. On the success path the pointer is
+            // owned by `Url` and freed via its `Drop` impl, but on the error path
+            // we must free it here to avoid leaking the allocation.
+            unsafe { ffi::ada_free(url_aggregator) };
             Err(ParseUrlError { input })
         }
     }

tests/basic_tests.rs

@@ -242,6 +242,22 @@ fn confusing_mess() {
     assert_eq!(url.origin(), "http://d:2");
 }
 
+/// Regression test for <https://github.com/ada-url/rust/issues/101>.
+///
+/// `ada_parse`/`ada_parse_with_base` always allocate a result on the heap,
+/// even when the input fails to parse. The failed-parse path used to drop the
+/// returned pointer without calling `ada_free`, leaking one allocation per
+/// call. This test repeatedly parses invalid input on both the base-less and
+/// with-base code paths; under LeakSanitizer/Valgrind a regression here shows
+/// up as leaked allocations originating in `ada_parse`.
+#[test]
+fn parse_invalid_url_does_not_leak() {
+    for _ in 0..1_000 {
+        assert!(Url::parse("http://[invalid", None).is_err());
+        assert!(Url::parse("http://[invalid", Some("http://example.com")).is_err());
+    }
+}
+
 #[test]
 fn standard_file() {
     let url = parse("file:///tmp/mock/path");