xargs: clarify -x and -t help text

The previous text implied -x only interacts with -L/-n and used
imprecise phrasing. The new text mirrors the POSIX spec: terminate
when a constructed command line would exceed the size limit rather
than reducing the argument count.

Clarify that -t prints each command to stderr, rather than just
"be verbose".

.github/workflows/ci.yml

@@ -65,6 +65,29 @@ jobs:
       - run: |
           cargo clippy --all-targets -- -D warnings
 
+  wasm:
+    name: cargo build (wasm32-wasip1)
+    runs-on: ubuntu-latest
+    env:
+      WASI_SDK_VERSION: "33"
+      WASI_SDK_PATH: ${{ github.workspace }}/wasi-sdk
+      # onig_sys compiles bundled C, so it needs a C compiler with a WASI
+      # sysroot to cross-compile to wasm.
+      CC_wasm32_wasip1: ${{ github.workspace }}/wasi-sdk/bin/clang
+      AR_wasm32_wasip1: ${{ github.workspace }}/wasi-sdk/bin/llvm-ar
+    steps:
+      - uses: actions/checkout@v6
+      - name: Install wasm target
+        run: rustup target add wasm32-wasip1
+      - name: Install WASI SDK
+        run: |
+          curl -sSfL "https://github.com/WebAssembly/wasi-sdk/releases/download/wasi-sdk-${WASI_SDK_VERSION}/wasi-sdk-${WASI_SDK_VERSION}.0-x86_64-linux.tar.gz" | tar xz
+          mv "wasi-sdk-${WASI_SDK_VERSION}.0-x86_64-linux" "${WASI_SDK_PATH}"
+      - name: Build
+        run: cargo build --target wasm32-wasip1
+      - name: Clippy
+        run: cargo clippy --target wasm32-wasip1 -- -D warnings
+
   grcov:
     name: Code coverage
     runs-on: ${{ matrix.os }}
@@ -92,7 +115,7 @@ jobs:
           path: lcov.info
 
       - name: Upload coverage to codecov.io
-        uses: codecov/codecov-action@v6
+        uses: codecov/codecov-action@v7
         with:
           files: lcov.info
           fail_ci_if_error: true

.github/workflows/codspeed.yml

@@ -0,0 +1,37 @@
+name: CodSpeed
+
+on:
+  push:
+    branches:
+      - "main"
+  pull_request:
+  # `workflow_dispatch` allows CodSpeed to trigger backtest
+  # performance analysis in order to generate initial data.
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  id-token: write
+
+jobs:
+  codspeed:
+    name: Run benchmarks
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+
+      - name: Setup Rust toolchain, cache and cargo-codspeed binary
+        uses: moonrepo/setup-rust@v1
+        with:
+          channel: stable
+          cache-target: release
+          bins: cargo-codspeed
+
+      - name: Build the benchmark target(s)
+        run: cargo codspeed build
+
+      - name: Run the benchmarks
+        uses: CodSpeedHQ/action@v4
+        with:
+          mode: simulation
+          run: cargo codspeed run

.github/workflows/compat.yml

@@ -46,7 +46,7 @@ jobs:
           install --strip -Dm755 target/release/{find,xargs} -t findutils-x86_64-unknown-linux-gnu
           ZSTD_CLEVEL=19 tar --zstd -caf ../findutils-x86_64-unknown-linux-gnu.tar.zst findutils-x86_64-unknown-linux-gnu
       - name: Publish latest commit
-        uses: softprops/action-gh-release@v2
+        uses: softprops/action-gh-release@v3
         if: github.event_name == 'push' && github.ref == 'refs/heads/main'
         with:
           tag_name: latest-commit

.github/workflows/release.yml

@@ -272,7 +272,7 @@ jobs:
           # Write and read notes from a file to avoid quoting breaking things
           echo "$ANNOUNCEMENT_BODY" > $RUNNER_TEMP/notes.txt
 
-          gh release create "${{ needs.plan.outputs.tag }}" --target "$RELEASE_COMMIT" $PRERELEASE_FLAG --title "$ANNOUNCEMENT_TITLE" --notes-file "$RUNNER_TEMP/notes.txt" artifacts/*
+          gh release create "${{ needs.plan.outputs.tag }}" --draft --target "$RELEASE_COMMIT" $PRERELEASE_FLAG --title "$ANNOUNCEMENT_TITLE" --notes-file "$RUNNER_TEMP/notes.txt" artifacts/*
 
   announce:
     needs:

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "findutils"
-version = "0.8.0"
+version = "0.9.0"
 homepage = "https://github.com/uutils/findutils"
 repository = "https://github.com/uutils/findutils"
 edition = "2021"
@@ -11,26 +11,55 @@ authors = ["uutils developers"]
 
 [dependencies]
 argmax = "0.4.0"
-chrono = "0.4.44"
+chrono = "0.4.45"
-clap = "4.6"
+clap = { version = "4.6", features = ["env"] }
 faccess = "0.2.4"
 nix = { version = "0.31", features = ["fs", "user"] }
 onig = { version = "6.5", default-features = false }
 regex = "1.12"
 uucore = { version = "0.9.0", features = ["entries", "fs", "fsext", "mode"] }
 walkdir = "2.5"
+itertools = "0.14.0"
+thiserror = "2.0.12"
 
 [dev-dependencies]
+criterion = { version = "4.7.0", package = "codspeed-criterion-compat" }
+assert_cmd = "2"
 ctor = "1.0"
 filetime = "0.2"
 nix = { version = "0.31", features = ["fs"] }
+rstest = "0.26.1"
 tempfile = "3"
 uutests = "0.9.0"
 
+[[bench]]
+name = "find_bench"
+harness = false
+
+[[bench]]
+name = "xargs_bench"
+harness = false
+
+[[bench]]
+name = "updatedb_bench"
+harness = false
+
+[[bench]]
+name = "locate_bench"
+harness = false
+
 [[bin]]
 name = "find"
 path = "src/find/main.rs"
 
+[[bin]]
+name = "locate"
+path = "src/locate/main.rs"
+
+[[bin]]
+name = "updatedb"
+path = "src/updatedb/main.rs"
+
 [[bin]]
 name = "xargs"
 path = "src/xargs/main.rs"
@@ -48,17 +77,43 @@ inherits = "release"
 panic = "abort"
 
 [lints.clippy]
+all = { level = "warn", priority = -1 }
+cargo = { level = "warn", priority = -1 }
+pedantic = { level = "warn", priority = -1 }
+use_self = "warn" # nursery lint
+
+# Group-level allows for lints that are too noisy or not worth enforcing for now.
 multiple_crate_versions = "allow"
 cargo_common_metadata = "allow"
 uninlined_format_args = "allow"
+missing_errors_doc = "allow"
 missing_panics_doc = "allow"
-
+must_use_candidate = "allow"
-use_self = "warn"
+match_same_arms = "allow"
-needless_pass_by_value = "warn"
+cast_possible_truncation = "allow"
-semicolon_if_nothing_returned = "warn"
+too_many_lines = "allow"
-single_char_pattern = "warn"
+cast_possible_wrap = "allow"
-explicit_iter_loop = "warn"
+cast_sign_loss = "allow"
-if_not_else = "warn"
+struct_excessive_bools = "allow"
-manual_let_else = "warn"
+cast_precision_loss = "allow"
+cast_lossless = "allow"
+ignored_unit_patterns = "allow"
+similar_names = "allow"
+float_cmp = "allow"
+return_self_not_must_use = "allow"
+inline_always = "allow"
+fn_params_excessive_bools = "allow"
+used_underscore_items = "allow"
+should_panic_without_expect = "allow"
+doc_markdown = "allow"
+unused_self = "allow"
+enum_glob_use = "allow"
+unnested_or_patterns = "allow"
+implicit_hasher = "allow"
+doc_link_with_quotes = "allow"
+format_push_string = "allow"
+flat_map_option = "allow"
+from_iter_instead_of_collect = "allow"
+large_types_passed_by_value = "allow"
 # Disable for now, we have a few occurrences
 # panic = "warn"

benches/find_bench.rs

@@ -0,0 +1,166 @@
+// Copyright 2024 the uutils developers
+//
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file or at
+// https://opensource.org/licenses/MIT.
+
+//! End-to-end benchmarks for `find`, run through the real `find_main` entry
+//! point so they exercise argument parsing, the matcher tree and the directory
+//! walk together. Output is sent to a sink so the benchmarks measure the work
+//! `find` does rather than terminal I/O.
+
+use std::cell::RefCell;
+use std::io::{self, Write};
+use std::path::{Path, PathBuf};
+use std::time::SystemTime;
+
+use criterion::{black_box, criterion_group, criterion_main, Criterion};
+use findutils::find::{find_main, Dependencies};
+
+/// `Dependencies` implementation that throws output away. A fixed `now` keeps
+/// time-based matchers (`-newer`, `-mtime`, …) deterministic across runs.
+struct SinkDependencies {
+    output: RefCell<io::Sink>,
+    now: SystemTime,
+}
+
+impl SinkDependencies {
+    fn new() -> Self {
+        Self {
+            output: RefCell::new(io::sink()),
+            now: SystemTime::now(),
+        }
+    }
+}
+
+impl Dependencies for SinkDependencies {
+    fn get_output(&self) -> &RefCell<dyn Write> {
+        &self.output
+    }
+
+    fn now(&self) -> SystemTime {
+        self.now
+    }
+
+    fn confirm(&self, _prompt: &str) -> bool {
+        false
+    }
+}
+
+/// Run `find` end-to-end. `args` are the arguments after the program name
+/// (flags, paths, expression). The exit status is ignored — we only care about
+/// the work performed.
+fn run(args: &[&str]) {
+    let mut argv: Vec<&str> = Vec::with_capacity(args.len() + 1);
+    argv.push("find");
+    argv.extend_from_slice(args);
+    let deps = SinkDependencies::new();
+    let _ = find_main(&argv, &deps);
+}
+
+/// Build a moderately deep directory tree to walk. `depth` levels, each holding
+/// `dirs` sub-directories and `files` files. File names vary so name/regex
+/// matchers have something to discriminate on, and sizes vary so `-size` does
+/// real work. Returns the root directory.
+fn build_tree(depth: u32, dirs: u32, files: u32) -> PathBuf {
+    let root = std::env::temp_dir().join(format!("uu_find_bench_{}", std::process::id()));
+    let _ = std::fs::remove_dir_all(&root);
+    std::fs::create_dir_all(&root).unwrap();
+    populate(&root, depth, dirs, files, 0);
+    root
+}
+
+fn populate(dir: &Path, depth: u32, dirs: u32, files: u32, seed: u32) {
+    for f in 0..files {
+        let n = seed.wrapping_add(f);
+        // A mix of extensions and a sprinkling of a rare marker name.
+        let name = match n % 5 {
+            0 => format!("module_{n}.rs"),
+            1 => format!("data_{n}.txt"),
+            2 => format!("image_{n}.png"),
+            3 => format!("README_{n}.md"),
+            _ if n % 500 == 0 => format!("RAREHIT_{n}.log"),
+            _ => format!("file_{n}.log"),
+        };
+        // Sizes from a few bytes up to ~8 KiB so -size buckets differ.
+        let size = (n as usize % 8192) + 1;
+        let path = dir.join(name);
+        std::fs::write(&path, vec![b'x'; size]).unwrap();
+    }
+
+    if depth == 0 {
+        return;
+    }
+
+    for d in 0..dirs {
+        let sub = dir.join(format!("dir_{d}"));
+        std::fs::create_dir_all(&sub).unwrap();
+        populate(
+            &sub,
+            depth - 1,
+            dirs,
+            files,
+            seed.wrapping_add((d + 1) * 31),
+        );
+    }
+}
+
+fn bench_e2e(c: &mut Criterion) {
+    // depth 4, 4 dirs/level, 25 files/dir → a few thousand entries.
+    let root = build_tree(4, 4, 25);
+    let dir = root.to_str().unwrap();
+
+    let mut group = c.benchmark_group("find");
+
+    // Plain full walk with the implicit -print.
+    group.bench_function("walk_all", |b| {
+        b.iter(|| run(black_box(&[dir])));
+    });
+    // Filter by file type only.
+    group.bench_function("type_f", |b| {
+        b.iter(|| run(black_box(&[dir, "-type", "f"])));
+    });
+    // Glob name match — a common invocation.
+    group.bench_function("name_glob", |b| {
+        b.iter(|| run(black_box(&[dir, "-name", "*.rs"])));
+    });
+    // Case-insensitive name match.
+    group.bench_function("iname_glob", |b| {
+        b.iter(|| run(black_box(&[dir, "-iname", "*.RS"])));
+    });
+    // Regex over the whole path.
+    group.bench_function("regex_path", |b| {
+        b.iter(|| run(black_box(&[dir, "-regex", r".*/module_[0-9]+\.rs"])));
+    });
+    // Size predicate forces a stat per entry.
+    group.bench_function("size_gt", |b| {
+        b.iter(|| run(black_box(&[dir, "-type", "f", "-size", "+4k"])));
+    });
+    // Combined predicate with AND/OR and grouping.
+    group.bench_function("combined_expr", |b| {
+        b.iter(|| {
+            run(black_box(&[
+                dir, "-type", "f", "(", "-name", "*.rs", "-o", "-name", "*.md", ")",
+            ]));
+        });
+    });
+    // Prune whole subtrees, then print the rest.
+    group.bench_function("prune", |b| {
+        b.iter(|| {
+            run(black_box(&[
+                dir, "-name", "dir_0", "-prune", "-o", "-type", "f", "-print",
+            ]));
+        });
+    });
+    // -printf with several directives exercises the formatter.
+    group.bench_function("printf", |b| {
+        b.iter(|| run(black_box(&[dir, "-type", "f", "-printf", "%p %s %y\\n"])));
+    });
+
+    group.finish();
+
+    let _ = std::fs::remove_dir_all(&root);
+}
+
+criterion_group!(benches, bench_e2e);
+criterion_main!(benches);

benches/locate_bench.rs

@@ -0,0 +1,127 @@
+// Copyright 2024 the uutils developers
+//
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file or at
+// https://opensource.org/licenses/MIT.
+
+//! End-to-end benchmarks for `locate`, run through the real `locate_main` entry
+//! point. A LOCATE02 database is built once (via `updatedb`) from a generated
+//! tree, then each case scans it in `-c`/count mode so the database decoding
+//! and pattern matching dominate rather than terminal output.
+//!
+//! `locate` is Unix-only (it relies on `OsStr`/`CStr` byte APIs), so the
+//! benchmark body is compiled only there.
+
+#[cfg(unix)]
+mod bench {
+    use std::path::{Path, PathBuf};
+
+    use criterion::{black_box, criterion_group, Criterion};
+    use findutils::locate::locate_main;
+    use findutils::updatedb::updatedb_main;
+
+    /// Run `locate` end-to-end. `args` are the arguments after the program name.
+    fn run(args: &[&str]) {
+        let mut argv: Vec<&str> = Vec::with_capacity(args.len() + 1);
+        argv.push("locate");
+        argv.extend_from_slice(args);
+        let _ = locate_main(&argv);
+    }
+
+    fn build_tree(depth: u32, dirs: u32, files: u32) -> PathBuf {
+        let root = std::env::temp_dir().join(format!("uu_locate_bench_{}", std::process::id()));
+        let _ = std::fs::remove_dir_all(&root);
+        std::fs::create_dir_all(&root).unwrap();
+        populate(&root, depth, dirs, files, 0);
+        root
+    }
+
+    fn populate(dir: &Path, depth: u32, dirs: u32, files: u32, seed: u32) {
+        for f in 0..files {
+            let n = seed.wrapping_add(f);
+            let name = match n % 4 {
+                0 => format!("module_{n}.rs"),
+                1 => format!("data_{n}.txt"),
+                2 => format!("image_{n}.png"),
+                _ => format!("file_{n}.log"),
+            };
+            std::fs::write(dir.join(name), b"x").unwrap();
+        }
+
+        if depth == 0 {
+            return;
+        }
+
+        for d in 0..dirs {
+            let sub = dir.join(format!("dir_{d}"));
+            std::fs::create_dir_all(&sub).unwrap();
+            populate(
+                &sub,
+                depth - 1,
+                dirs,
+                files,
+                seed.wrapping_add((d + 1) * 31),
+            );
+        }
+    }
+
+    pub fn bench_e2e(c: &mut Criterion) {
+        // ~8.5k paths, then front-coded into a LOCATE02 database once.
+        let root = build_tree(4, 4, 25);
+        let dir = root.to_str().unwrap();
+        let db = std::env::temp_dir().join(format!("uu_locate_bench_{}.db", std::process::id()));
+        let db_path = db.to_str().unwrap();
+
+        let _ = updatedb_main(&[
+            "updatedb",
+            &format!("--localpaths={dir}"),
+            &format!("--output={db_path}"),
+            "--prunepaths=",
+            "--prunefs=",
+        ]);
+
+        let mut group = c.benchmark_group("locate");
+
+        // Pattern matching nothing: forces a full decode + match of every entry.
+        group.bench_function("count_no_match", |b| {
+            b.iter(|| run(black_box(&["-d", db_path, "-c", "NONEXISTENT_TOKEN_XYZ"])));
+        });
+        // Substring matching many entries (count mode → no per-match output).
+        group.bench_function("count_many", |b| {
+            b.iter(|| run(black_box(&["-d", db_path, "-c", ".log"])));
+        });
+        // Basename matching (-b) restricts the comparison to the file name.
+        group.bench_function("basename", |b| {
+            b.iter(|| run(black_box(&["-d", db_path, "-c", "-b", "module"])));
+        });
+        // Case-insensitive matching (-i).
+        group.bench_function("ignore_case", |b| {
+            b.iter(|| run(black_box(&["-d", db_path, "-c", "-i", "MODULE"])));
+        });
+        // Regex matching (-r) over the whole path.
+        group.bench_function("regex", |b| {
+            b.iter(|| {
+                run(black_box(&[
+                    "-d",
+                    db_path,
+                    "-c",
+                    "-r",
+                    r"module_[0-9]+\.rs",
+                ]));
+            });
+        });
+
+        group.finish();
+
+        let _ = std::fs::remove_dir_all(&root);
+        let _ = std::fs::remove_file(&db);
+    }
+
+    criterion_group!(benches, bench_e2e);
+}
+
+#[cfg(unix)]
+criterion::criterion_main!(bench::benches);
+
+#[cfg(not(unix))]
+fn main() {}

benches/updatedb_bench.rs

@@ -0,0 +1,110 @@
+// Copyright 2024 the uutils developers
+//
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file or at
+// https://opensource.org/licenses/MIT.
+
+//! End-to-end benchmarks for `updatedb`, run through the real `updatedb_main`
+//! entry point. Each case walks a generated directory tree (via the internal
+//! `find`), encodes it into the LOCATE02 database format and writes the result
+//! to a temp file, so the walk + front-coding + write dominate the measurement.
+
+use std::path::{Path, PathBuf};
+
+use criterion::{black_box, criterion_group, criterion_main, Criterion};
+use findutils::updatedb::updatedb_main;
+
+/// Run `updatedb` end-to-end. `args` are the arguments after the program name.
+fn run(args: &[&str]) {
+    let mut argv: Vec<&str> = Vec::with_capacity(args.len() + 1);
+    argv.push("updatedb");
+    argv.extend_from_slice(args);
+    let _ = updatedb_main(&argv);
+}
+
+/// Build a directory tree to index. `depth` levels, each holding `dirs`
+/// sub-directories and `files` files with varied names. Returns the root.
+fn build_tree(depth: u32, dirs: u32, files: u32) -> PathBuf {
+    let root = std::env::temp_dir().join(format!("uu_updatedb_bench_{}", std::process::id()));
+    let _ = std::fs::remove_dir_all(&root);
+    std::fs::create_dir_all(&root).unwrap();
+    populate(&root, depth, dirs, files, 0);
+    root
+}
+
+fn populate(dir: &Path, depth: u32, dirs: u32, files: u32, seed: u32) {
+    for f in 0..files {
+        let n = seed.wrapping_add(f);
+        let name = match n % 4 {
+            0 => format!("module_{n}.rs"),
+            1 => format!("data_{n}.txt"),
+            2 => format!("image_{n}.png"),
+            _ => format!("file_{n}.log"),
+        };
+        // Tiny files: updatedb only records names, so content size is irrelevant.
+        std::fs::write(dir.join(name), b"x").unwrap();
+    }
+
+    if depth == 0 {
+        return;
+    }
+
+    for d in 0..dirs {
+        let sub = dir.join(format!("dir_{d}"));
+        std::fs::create_dir_all(&sub).unwrap();
+        populate(
+            &sub,
+            depth - 1,
+            dirs,
+            files,
+            seed.wrapping_add((d + 1) * 31),
+        );
+    }
+}
+
+fn bench_e2e(c: &mut Criterion) {
+    // depth 4, 4 dirs/level, 25 files/dir → ~8.5k indexed paths.
+    let root = build_tree(4, 4, 25);
+    let dir = root.to_str().unwrap();
+    let db = std::env::temp_dir().join(format!("uu_updatedb_bench_{}.db", std::process::id()));
+    let db_path = db.to_str().unwrap();
+
+    let localpaths = format!("--localpaths={dir}");
+    let output = format!("--output={db_path}");
+
+    let mut group = c.benchmark_group("updatedb");
+
+    // Full build: walk, front-code and write the whole tree. Empty prune
+    // options keep the run deterministic regardless of where temp_dir lives.
+    group.bench_function("build_full", |b| {
+        b.iter(|| {
+            run(black_box(&[
+                &localpaths,
+                &output,
+                "--prunepaths=",
+                "--prunefs=",
+            ]));
+        });
+    });
+
+    // Build with a prune clause that drops every `dir_0` subtree, exercising the
+    // -regex/-prune path updatedb hands to find.
+    group.bench_function("build_pruned", |b| {
+        b.iter(|| {
+            run(black_box(&[
+                &localpaths,
+                &output,
+                "--prunepaths=.*/dir_0",
+                "--prunefs=",
+            ]));
+        });
+    });
+
+    group.finish();
+
+    let _ = std::fs::remove_dir_all(&root);
+    let _ = std::fs::remove_file(&db);
+}
+
+criterion_group!(benches, bench_e2e);
+criterion_main!(benches);

benches/xargs_bench.rs

@@ -0,0 +1,77 @@
+// Copyright 2024 the uutils developers
+//
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file or at
+// https://opensource.org/licenses/MIT.
+
+//! End-to-end benchmarks for `xargs`, run through the real `xargs_main` entry
+//! point. Input is supplied via `-a <file>` (rather than stdin) so the harness
+//! can feed a fixed corpus without touching the process's stdin, and the
+//! command is `true` so the measurement is dominated by `xargs`'s own work
+//! (reading, splitting and batching arguments) rather than the child.
+
+use std::path::PathBuf;
+
+use criterion::{black_box, criterion_group, criterion_main, Criterion};
+use findutils::xargs::xargs_main;
+
+/// Run `xargs` end-to-end. `args` are the arguments after the program name.
+/// The exit status is ignored — we only care about the work performed.
+fn run(args: &[&str]) {
+    let mut argv: Vec<&str> = Vec::with_capacity(args.len() + 1);
+    argv.push("xargs");
+    argv.extend_from_slice(args);
+    let _ = xargs_main(&argv);
+}
+
+/// Write `count` short tokens to a temp file, joined by `sep`. Returns the path.
+fn build_input(count: u32, sep: &str, tag: &str) -> PathBuf {
+    let mut content = String::new();
+    for i in 0..count {
+        if i > 0 {
+            content.push_str(sep);
+        }
+        content.push_str(&format!("item_{i}"));
+    }
+    let path = std::env::temp_dir().join(format!("uu_xargs_bench_{}_{tag}", std::process::id()));
+    std::fs::write(&path, content).unwrap();
+    path
+}
+
+fn bench_e2e(c: &mut Criterion) {
+    // Whitespace- and newline-separated corpora, plus a NUL-separated one for
+    // the `-0` path. A few thousand short tokens keeps parsing dominant while
+    // `true` batches stay to a handful of (untraced) spawns.
+    let ws = build_input(4000, " ", "ws");
+    let nul = build_input(4000, "\0", "nul");
+    let ws_path = ws.to_str().unwrap();
+    let nul_path = nul.to_str().unwrap();
+
+    let mut group = c.benchmark_group("xargs");
+
+    // Default whitespace splitting, single batch (fits one command line).
+    group.bench_function("split_whitespace", |b| {
+        b.iter(|| run(black_box(&["-a", ws_path, "true"])));
+    });
+    // NUL-delimited input (`find -print0 | xargs -0` shape).
+    group.bench_function("split_null", |b| {
+        b.iter(|| run(black_box(&["-0", "-a", nul_path, "true"])));
+    });
+    // Cap arguments per command with -n, exercising the batching path across
+    // several (untraced) invocations.
+    group.bench_function("batched_n", |b| {
+        b.iter(|| run(black_box(&["-a", ws_path, "-n", "1000", "true"])));
+    });
+    // Bound each command line by size with -s, another batching trigger.
+    group.bench_function("batched_size", |b| {
+        b.iter(|| run(black_box(&["-a", ws_path, "-s", "4096", "true"])));
+    });
+
+    group.finish();
+
+    let _ = std::fs::remove_file(&ws);
+    let _ = std::fs::remove_file(&nul);
+}
+
+criterion_group!(benches, bench_e2e);
+criterion_main!(benches);

src/find/matchers/entry.rs

@@ -174,9 +174,9 @@ impl From<walkdir::Error> for WalkError {
 impl From<&walkdir::Error> for WalkError {
     fn from(e: &walkdir::Error) -> Self {
         Self {
-            path: e.path().map(|p| p.to_owned()),
+            path: e.path().map(std::borrow::ToOwned::to_owned),
             depth: Some(e.depth()),
-            raw: e.io_error().and_then(|e| e.raw_os_error()),
+            raw: e.io_error().and_then(std::io::Error::raw_os_error),
         }
     }
 }
@@ -190,8 +190,7 @@ impl From<WalkError> for io::Error {
 impl From<&WalkError> for io::Error {
     fn from(e: &WalkError) -> Self {
         e.raw
-            .map(Self::from_raw_os_error)
+            .map_or_else(|| ErrorKind::Other.into(), Self::from_raw_os_error)
-            .unwrap_or_else(|| ErrorKind::Other.into())
     }
 }
 
@@ -279,8 +278,7 @@ impl WalkEntry {
                 // Path::file_name() only works if the last component is normal
                 path.components()
                     .next_back()
-                    .map(|c| c.as_os_str())
+                    .map_or_else(|| path.as_os_str(), std::path::Component::as_os_str)
-                    .unwrap_or_else(|| path.as_os_str())
             }
             Entry::WalkDir(ent) => ent.file_name(),
         }
@@ -311,7 +309,7 @@ impl WalkEntry {
             Entry::Explicit(_, _) => Ok(self.get_metadata()?),
             Entry::WalkDir(ent) => Ok(ent.metadata()?),
         });
-        result.as_ref().map_err(|e| e.clone())
+        result.as_ref().map_err(std::clone::Clone::clone)
     }
 
     /// Get the file type of this entry.
@@ -319,8 +317,7 @@ impl WalkEntry {
         match &self.inner {
             Entry::Explicit(_, _) => self
                 .metadata()
-                .map(|m| m.file_type().into())
+                .map_or(FileType::Unknown, |m| m.file_type().into()),
-                .unwrap_or(FileType::Unknown),
             Entry::WalkDir(ent) => ent.file_type().into(),
         }
     }

src/find/matchers/exec.rs

@@ -18,8 +18,17 @@ enum Arg {
     LiteralArg(OsString),
 }
 
+fn parse_arg(s: &str) -> Arg {
+    let parts = s.split("{}").collect::<Vec<_>>();
+    if parts.len() == 1 {
+        Arg::LiteralArg(OsString::from(s))
+    } else {
+        Arg::FileArg(parts.iter().map(OsString::from).collect())
+    }
+}
+
 pub struct SingleExecMatcher {
-    executable: String,
+    executable: Arg,
     args: Vec<Arg>,
     exec_in_parent_dir: bool,
     interactive: bool,
@@ -31,7 +40,7 @@ impl SingleExecMatcher {
         args: &[&str],
         exec_in_parent_dir: bool,
     ) -> Result<Self, Box<dyn Error>> {
-        Self::new_impl(executable, args, exec_in_parent_dir, false)
+        Ok(Self::new_impl(executable, args, exec_in_parent_dir, false))
     }
 
     pub fn new_interactive(
@@ -39,7 +48,7 @@ impl SingleExecMatcher {
         args: &[&str],
         exec_in_parent_dir: bool,
     ) -> Result<Self, Box<dyn Error>> {
-        Self::new_impl(executable, args, exec_in_parent_dir, true)
+        Ok(Self::new_impl(executable, args, exec_in_parent_dir, true))
     }
 
     fn new_impl(
@@ -47,26 +56,15 @@ impl SingleExecMatcher {
         args: &[&str],
         exec_in_parent_dir: bool,
         interactive: bool,
-    ) -> Result<Self, Box<dyn Error>> {
+    ) -> Self {
-        let transformed_args = args
+        let transformed_args = args.iter().map(|&a| parse_arg(a)).collect();
-            .iter()
-            .map(|&a| {
-                let parts = a.split("{}").collect::<Vec<_>>();
-                if parts.len() == 1 {
-                    // No {} present
-                    Arg::LiteralArg(OsString::from(a))
-                } else {
-                    Arg::FileArg(parts.iter().map(OsString::from).collect())
-                }
-            })
-            .collect();
 
-        Ok(Self {
+        Self {
-            executable: executable.to_string(),
+            executable: parse_arg(executable),
             args: transformed_args,
             exec_in_parent_dir,
             interactive,
-        })
+        }
     }
 }
 
@@ -82,28 +80,30 @@ impl Matcher for SingleExecMatcher {
             file_info.path().to_path_buf()
         };
 
+        let resolved_executable = match self.executable {
+            Arg::LiteralArg(ref a) => a.clone(),
+            Arg::FileArg(ref parts) => parts.join(path_to_file.as_os_str()),
+        };
+
         if self.interactive {
-            let rendered_args: Vec<String> = self
+            // GNU find prints a fixed, abbreviated prompt of the form
-                .args
+            // "< executable ... pathname > ? ".  It does not render the
-                .iter()
+            // substituted argument list, and always shows the full path of
-                .map(|arg| match arg {
+            // the entry being processed (even for -okdir, whose command runs
-                    Arg::LiteralArg(a) => a.to_string_lossy().into_owned(),
+            // with the "./basename" form).
-                    Arg::FileArg(parts) => parts
+            let prompt = format!(
-                        .join(path_to_file.as_os_str())
+                "< {} ... {} > ? ",
-                        .to_string_lossy()
+                resolved_executable.to_string_lossy(),
-                        .into_owned(),
+                file_info.path().to_string_lossy()
-                })
+            );
-                .collect();
-            let mut prompt_parts = vec![self.executable.clone()];
-            prompt_parts.extend(rendered_args);
-            let prompt = format!("< {} >? ", prompt_parts.join(" "));
 
             if !matcher_io.confirm(&prompt) {
                 return false;
             }
         }
 
-        let mut command = Command::new(&self.executable);
+        let mut command = Command::new(&resolved_executable);
+
         for arg in &self.args {
             match *arg {
                 Arg::LiteralArg(ref a) => command.arg(a.as_os_str()),
@@ -127,7 +127,13 @@ impl Matcher for SingleExecMatcher {
         match command.status() {
             Ok(status) => status.success(),
             Err(e) => {
-                writeln!(&mut stderr(), "Failed to run {}: {}", self.executable, e).unwrap();
+                writeln!(
+                    &mut stderr(),
+                    "Failed to run {}: {}",
+                    resolved_executable.to_string_lossy(),
+                    e
+                )
+                .unwrap();
                 false
             }
         }

src/find/matchers/fs.rs

@@ -42,8 +42,18 @@ pub fn get_file_system_type(path: &Path, cache: &RefCell<Option<Cache>>) -> URes
         Ok(metadata) => metadata,
         Err(err) => Err(err)?,
     };
-    let dev_id = metadata.dev().to_string();
 
+    fs_type_for_dev(metadata.dev().to_string(), cache)
+}
+
+/// Resolve a device id to its filesystem type, consulting (and updating) `cache`.
+///
+/// Callers are expected to obtain `dev_id` from metadata they already hold (e.g. the cached
+/// [`WalkEntry::metadata`]), so that no extra `stat`/`statx` syscall is issued per entry.
+///
+/// This is only supported on Unix.
+#[cfg(unix)]
+pub fn fs_type_for_dev(dev_id: String, cache: &RefCell<Option<Cache>>) -> UResult<String> {
     if let Some(cache) = cache.borrow().as_ref() {
         if cache.dev_id == dev_id {
             return Ok(cache.fs_type.clone());
@@ -94,18 +104,34 @@ impl FileSystemMatcher {
 impl Matcher for FileSystemMatcher {
     #[cfg(unix)]
     fn matches(&self, file_info: &WalkEntry, _: &mut MatcherIO) -> bool {
-        match get_file_system_type(file_info.path(), &self.cache) {
+        use std::os::unix::fs::MetadataExt;
-            Ok(result) => result == self.fs_text,
-            Err(_) => {
-                writeln!(
-                    &mut stderr(),
-                    "Error getting filesystem type for {}",
-                    file_info.path().to_string_lossy()
-                )
-                .unwrap();
 
-                false
+        // Reuse the metadata already cached on the entry (a single shared `statx` per entry)
-            }
+        // rather than issuing a fresh `lstat`/`statx` here. With several `-fstype` clauses (as
+        // `updatedb` builds) this turns N stats per file into one.
+        let Ok(metadata) = file_info.metadata() else {
+            writeln!(
+                &mut stderr(),
+                "Error getting filesystem type for {}",
+                file_info.path().to_string_lossy()
+            )
+            .unwrap();
+
+            return false;
+        };
+        let dev_id = metadata.dev().to_string();
+
+        if let Ok(result) = fs_type_for_dev(dev_id, &self.cache) {
+            result == self.fs_text
+        } else {
+            writeln!(
+                &mut stderr(),
+                "Error getting filesystem type for {}",
+                file_info.path().to_string_lossy()
+            )
+            .unwrap();
+
+            false
         }
     }
 

src/find/matchers/glob.rs

@@ -83,7 +83,10 @@ fn extract_bracket_expr(pattern: &str) -> Option<(String, &str)> {
 
                     if matches!(delim, '.' | '=' | ':') {
                         let rest = chars.as_str();
-                        let end = rest.find([delim, ']'])? + 2;
+                        // Search for the two-byte closer `<delim>]` (e.g. `:]`);
+                        // matching either byte alone let `+ 2` overshoot a char boundary.
+                        let closer = format!("{delim}]");
+                        let end = rest.find(closer.as_str())? + 2;
                         expr.push_str(&rest[..end]);
                         chars = rest[end..].chars();
                     }
@@ -222,6 +225,16 @@ mod tests {
         assert_glob_regex(r"foo[bar[!baz", r"foo\[bar\[!baz");
     }
 
+    #[test]
+    fn malformed_posix_class_with_multibyte_char() {
+        for pat in ["[[:]é", "[[:a]é", "[[.]é", "[[:é]", "[[=]😀"] {
+            assert!(
+                glob_to_regex(pat).is_some(),
+                "panicked or rejected: {pat:?}"
+            );
+        }
+    }
+
     #[test]
     fn incomplete_escape() {
         assert_eq!(glob_to_regex(r"foo\"), None);

src/find/matchers/group.rs

@@ -11,6 +11,8 @@ use nix::unistd::Group;
 use std::os::unix::fs::MetadataExt;
 
 pub struct GroupMatcher {
+    // Only read on Unix; the non-Unix `matches` implementation is a stub.
+    #[cfg_attr(not(unix), allow(dead_code))]
     gid: ComparableValue,
 }
 
@@ -31,7 +33,7 @@ impl GroupMatcher {
         Self { gid }
     }
 
-    #[cfg(windows)]
+    #[cfg(not(unix))]
     pub fn from_group_name(_group: &str) -> Option<Self> {
         None
     }
@@ -46,10 +48,10 @@ impl Matcher for GroupMatcher {
         }
     }
 
-    #[cfg(windows)]
+    #[cfg(not(unix))]
     fn matches(&self, _file_info: &WalkEntry, _: &mut MatcherIO) -> bool {
-        // The user group acquisition function for Windows systems is not implemented in MetadataExt,
+        // The user group acquisition function for non-Unix systems is not implemented in
-        // so it is somewhat difficult to implement it. :(
+        // MetadataExt, so it is somewhat difficult to implement it. :(
         false
     }
 }
@@ -80,7 +82,7 @@ impl Matcher for NoGroupMatcher {
         false
     }
 
-    #[cfg(windows)]
+    #[cfg(not(unix))]
     fn matches(&self, _file_info: &WalkEntry, _: &mut MatcherIO) -> bool {
         false
     }

src/find/matchers/ls.rs

@@ -126,7 +126,6 @@ impl Ls {
         mut out: impl Write,
         print_error_message: bool,
     ) {
-        use nix::unistd::{Gid, Group, Uid, User};
         use std::os::unix::fs::{MetadataExt, PermissionsExt};
 
         let metadata = file_info.metadata().unwrap();
@@ -150,14 +149,13 @@ impl Ls {
         let permission =
             { format_permissions(metadata.permissions().mode() as uucore::libc::mode_t) };
         let hard_links = metadata.nlink();
-        let user = {
+        // Fall back to the numeric id when the uid/gid has no passwd/group entry
-            let uid = metadata.uid();
+        // (unmapped owner) — matching GNU find, which never crashes. uucore::entries
-            User::from_uid(Uid::from_raw(uid)).unwrap().unwrap().name
+        // serializes the non-thread-safe getpwuid/getgrgid behind a mutex.
-        };
+        let user =
-        let group = {
+            uucore::entries::uid2usr(metadata.uid()).unwrap_or_else(|_| metadata.uid().to_string());
-            let gid = metadata.gid();
+        let group =
-            Group::from_gid(Gid::from_raw(gid)).unwrap().unwrap().name
+            uucore::entries::gid2grp(metadata.gid()).unwrap_or_else(|_| metadata.gid().to_string());
-        };
         let size = metadata.size();
         let last_modified = {
             let system_time = metadata.modified().unwrap();
@@ -195,7 +193,7 @@ impl Ls {
         }
     }
 
-    #[cfg(windows)]
+    #[cfg(not(unix))]
     fn print(
         &self,
         file_info: &WalkEntry,
@@ -203,16 +201,15 @@ impl Ls {
         mut out: impl Write,
         print_error_message: bool,
     ) {
-        use std::os::windows::fs::MetadataExt;
+        // Non-Unix targets (Windows, wasm, ...) don't expose inode, owner,
-
+        // group or Unix permission bits, so those columns are left blank.
         let metadata = file_info.metadata().unwrap();
 
         let inode_number = 0;
+        let size = metadata.len();
         let number_of_blocks = {
-            let size = metadata.file_size();
             let number_of_blocks = size / 1024;
             let remainder = number_of_blocks % 4;
-
             if remainder == 0 {
                 if number_of_blocks == 0 {
                     4
@@ -220,14 +217,19 @@ impl Ls {
                     number_of_blocks
                 }
             } else {
-                number_of_blocks + (4 - (remainder))
+                number_of_blocks + (4 - remainder)
             }
         };
-        let permission = { format_permissions(metadata.file_attributes()) };
+        #[cfg(windows)]
+        let permission = {
+            use std::os::windows::fs::MetadataExt;
+            format_permissions(metadata.file_attributes())
+        };
+        #[cfg(not(windows))]
+        let permission = "?---------";
         let hard_links = 0;
         let user = 0;
         let group = 0;
-        let size = metadata.file_size();
         let last_modified = {
             let system_time = metadata.modified().unwrap();
             let now_utc: DateTime<chrono::Utc> = system_time.into();
@@ -235,9 +237,9 @@ impl Ls {
         };
         let path = file_info.path().to_string_lossy();
 
-        match write!(
+        match writeln!(
             out,
-            " {:<4} {:>6} {:<10} {:>3} {:<8} {:<8} {:>8} {} {}\n",
+            " {:<4} {:>6} {:<10} {:>3} {:<8} {:<8} {:>8} {} {}",
             inode_number,
             number_of_blocks,
             permission,
@@ -292,15 +294,15 @@ mod tests {
     fn test_format_permissions() {
         use super::format_permissions;
 
-        let mode: uucore::libc::mode_t = 0o100644;
+        let mode: uucore::libc::mode_t = 0o100_644;
         let expected = "-rw-r--r--";
         assert_eq!(format_permissions(mode), expected);
 
-        let mode: uucore::libc::mode_t = 0o040755;
+        let mode: uucore::libc::mode_t = 0o040_755;
         let expected = "drwxr-xr-x";
         assert_eq!(format_permissions(mode), expected);
 
-        let mode: uucore::libc::mode_t = 0o100777;
+        let mode: uucore::libc::mode_t = 0o100_777;
         let expected = "-rwxrwxrwx";
         assert_eq!(format_permissions(mode), expected);
     }

src/find/matchers/mod.rs

@@ -74,6 +74,7 @@ use std::{
 use super::{Config, Dependencies};
 
 pub use entry::{FileType, WalkEntry, WalkError};
+pub use regex::RegexType;
 
 /// Symlink following mode.
 #[derive(Clone, Copy, Debug, Eq, PartialEq)]
@@ -399,7 +400,7 @@ fn parse_date_str_to_timestamps(date_str: &str) -> Option<i64> {
 ///
 /// Additionally, there is support for the -anewer and -cnewer short arguments. as follows:
 /// 1. -anewer is equivalent to -neweram
-/// 2. -cnewer is equivalent to - newercm
+/// 2. -cnewer is equivalent to -newercm
 ///
 /// If -newer is used it will be resolved to -newermm.
 fn parse_str_to_newer_args(input: &str) -> Option<(String, String)> {
@@ -1031,10 +1032,10 @@ fn parse_files0_args(config: &mut Config) -> Result<(), Box<dyn Error>> {
     let mut string_segments: Vec<String> = buffer_split
         .iter()
         .filter_map(|s| std::str::from_utf8(s).ok())
-        .map(|s| s.to_string())
+        .map(std::string::ToString::to_string)
         .collect();
     // empty starting point checker
-    if string_segments.iter().any(|s| s.is_empty()) {
+    if string_segments.iter().any(std::string::String::is_empty) {
         eprintln!("find: invalid zero-length file name");
         // remove the empty ones so as to avoid file not found error
         string_segments.retain(|s| !s.is_empty());

src/find/matchers/name.rs

@@ -31,7 +31,7 @@ impl Matcher for NameMatcher {
             self.pattern.matches(&name)
         }
 
-        #[cfg(windows)]
+        #[cfg(not(unix))]
         self.pattern.matches(&name)
     }
 }

src/find/matchers/perm.rs

@@ -91,7 +91,7 @@ impl PermMatcher {
     }
 
     #[cfg(not(unix))]
-    pub fn new(_dummy_pattern: &str) -> Result<PermMatcher, Box<dyn Error>> {
+    pub fn new(_dummy_pattern: &str) -> Result<Self, Box<dyn Error>> {
         Err(From::from(
             "Permission matching is not available on this platform",
         ))
@@ -132,7 +132,7 @@ impl Matcher for PermMatcher {
             "Permission matching not available on this platform!"
         )
         .unwrap();
-        return false;
+        false
     }
 }
 

src/find/matchers/printf.rs

@@ -17,6 +17,7 @@ use super::{FileType, Matcher, MatcherIO, WalkEntry, WalkError};
 #[cfg(unix)]
 use std::os::unix::prelude::MetadataExt;
 
+#[cfg(unix)]
 const STANDARD_BLOCK_SIZE: u64 = 512;
 
 #[derive(Debug, PartialEq, Eq)]
@@ -167,16 +168,21 @@ impl FormatStringParser<'_> {
         // Try parsing an octal sequence first.
         let first = self.front()?;
         if first.is_digit(OCTAL_RADIX) {
-            if let Ok(code) = self.peek(OCTAL_LEN).and_then(|octal| {
+            // A GNU octal escape is 1 to 3 octal digits. Consume only the leading
-                u32::from_str_radix(octal, OCTAL_RADIX).map_err(std::convert::Into::into)
+            // octal digits (which are ASCII), rather than slicing a fixed 3 bytes
-            }) {
+            // that can land inside a following multibyte char.
-                // safe to unwrap: .peek() already succeeded above.
+            let octal: String = self
-                let octal = self.advance_by(OCTAL_LEN).unwrap();
+                .string
-                return match char::from_u32(code) {
+                .chars()
-                    Some(c) => Ok(FormatComponent::Literal(c.to_string())),
+                .take(OCTAL_LEN)
-                    None => Err(format!("Invalid character value: \\{octal}").into()),
+                .take_while(|c| c.is_digit(OCTAL_RADIX))
-                };
+                .collect();
-            }
+            let code = u32::from_str_radix(&octal, OCTAL_RADIX)?;
+            self.advance_by(octal.len())?;
+            return match char::from_u32(code) {
+                Some(c) => Ok(FormatComponent::Literal(c.to_string())),
+                None => Err(format!("Invalid character value: \\{octal}").into()),
+            };
         }
 
         self.advance_one()?;
@@ -205,7 +211,7 @@ impl FormatStringParser<'_> {
         let start = self.string;
         let mut digits = 0;
 
-        while self.front().map(|c| c.is_ascii_digit()).unwrap_or(false) {
+        while self.front().is_ok_and(|c| c.is_ascii_digit()) {
             digits += 1;
             // safe to unwrap: the front() check already succeeded above.
             self.advance_one().unwrap();
@@ -688,6 +694,31 @@ mod tests {
         assert!(FormatString::parse("\\").is_err());
     }
 
+    #[test]
+    fn test_parse_octal_escape_before_multibyte_char() {
+        assert_eq!(
+            FormatString::parse("\\0€").unwrap().components,
+            vec![
+                FormatComponent::Literal("\0".to_owned()),
+                FormatComponent::Literal("€".to_owned()),
+            ]
+        );
+        assert_eq!(
+            FormatString::parse("\\1😀").unwrap().components,
+            vec![
+                FormatComponent::Literal("\u{1}".to_owned()),
+                FormatComponent::Literal("😀".to_owned()),
+            ]
+        );
+        assert_eq!(
+            FormatString::parse("\\00é").unwrap().components,
+            vec![
+                FormatComponent::Literal("\0".to_owned()),
+                FormatComponent::Literal("é".to_owned()),
+            ]
+        );
+    }
+
     #[test]
     fn test_parse_formatting() {
         fn unaligned_directive(directive: FormatDirective) -> FormatComponent {