Next Release

Cargo.toml

@@ -15,6 +15,14 @@ categories = ["command-line-utilities", "development-tools"]
 [dependencies]
 clap = { version = "4", features = ["derive"] }
 anyhow = "1.0"
+# Console code page decoding for child-process output. Kept off the
+# cfg(windows) target so the mapping and the incremental UTF-8 walk stay
+# compiled — and unit-tested — on every platform; only the code page *lookup*
+# is Windows-specific. codepage covers the ANSI/DBCS pages encoding_rs
+# implements, oem_cp the legacy OEM/DOS pages (437, 850, …) it does not.
+encoding_rs = "0.8"
+codepage = "0.1"
+oem_cp = "2"
 ignore = "0.4"
 walkdir = "2"
 regex = "1"
@@ -37,6 +45,12 @@ automod = "1"
 [target.'cfg(unix)'.dependencies]
 libc = "0.2"
 
+[target.'cfg(windows)'.dependencies]
+windows-sys = { version = "0.59", features = [
+    "Win32_System_Console",
+    "Win32_Globalization",
+] }
+
 [build-dependencies]
 toml = "0.8"
 

scripts/benchmark.sh

@@ -13,9 +13,15 @@ fi
 BENCH_DIR="$(pwd)/scripts/benchmark"
 RTK_ROOT="$(pwd)"
 
+# Only the gitignored output subdirectories get wiped between local runs.
+# `$BENCH_DIR` itself is tracked — it also holds the TypeScript VM-benchmark
+# harness (run.ts, cleanup.ts, lib/) — so `rm -rf "$BENCH_DIR"` deleted that
+# harness from the working tree on every local run.
 if [ -z "$CI" ]; then
-  rm -rf "$BENCH_DIR"
-  mkdir -p "$BENCH_DIR/unix" "$BENCH_DIR/rtk" "$BENCH_DIR/diff"
+  for bench_output_dir in unix rtk diff; do
+    rm -rf "${BENCH_DIR:?}/$bench_output_dir"
+    mkdir -p "$BENCH_DIR/$bench_output_dir"
+  done
 fi
 
 safe_name() {
@@ -342,21 +348,98 @@ section "wc"
 bench "wc" "wc Cargo.toml src/main.rs" "$RTK wc Cargo.toml src/main.rs"
 
 # ===================
-# curl
-# ===================
+# curl / wget — fully offline. mockhttp.org was both a network dependency and
+# non-deterministic (its /json output is random), which made the benchmark flaky.
+# Serve fixed fixtures locally instead. The JSON is pretty-printed on purpose so
+# `rtk` actually exercises JSON minification (a pre-minified body leaves nothing
+# to compact).
+# ===================
+NET_FIXTURE_DIR="$(mktemp -d)"
+# Server log lives outside the served directory so it is never itself served.
+NET_HTTP_LOG="$(mktemp)"
+NET_HTTP_PID=""
+# Every step is failure-tolerant: this runs from an EXIT trap under `set -e`, so
+# a single failing command (e.g. `kill` on a server that already died) would
+# otherwise abort the trap and leak the fixture dir and downloaded files.
+cleanup_net_fixtures() {
+  if [ -n "$NET_HTTP_PID" ]; then
+    kill "$NET_HTTP_PID" 2>/dev/null || true
+  fi
+  rm -rf "$NET_FIXTURE_DIR" "$NET_HTTP_LOG" || true
+  # `rtk wget <url>/data.json` saves to ./data.json (default basename); remove
+  # that download and any numbered duplicates from repeated runs.
+  rm -f data.json data.json.* 2>/dev/null || true
+}
+trap cleanup_net_fixtures EXIT
+
+cat > "$NET_FIXTURE_DIR/data.json" << 'JSONEOF'
+{
+    "message": "Hello from RTK benchmark",
+    "status": "success",
+    "code": 200,
+    "items": [
+        { "id": 1, "name": "first" },
+        { "id": 2, "name": "second" }
+    ]
+}
+JSONEOF
+cat > "$NET_FIXTURE_DIR/robots.txt" << 'TXTEOF'
+User-agent: *
+Disallow: /private/
+Allow: /
+Sitemap: https://example.com/sitemap.xml
+TXTEOF
+
+# Bring up a loopback HTTP server once so both curl and wget get real response
+# headers (`Content-Type: application/json`) — that's what lets rtk detect JSON
+# and minify it. file:// carries no headers, so it can't demonstrate that path.
+#
+# Port 0 asks the kernel for a free port, so a busy fixed port can never make the
+# benchmark fail. `python3 -u` keeps stdout unbuffered, so the "Serving HTTP on
+# 127.0.0.1 port NNNNN" line — printed only once the socket is bound and
+# listening — reaches the log as soon as the server is ready.
+readonly NET_HTTP_EPHEMERAL_PORT=0
+readonly NET_HTTP_READY_ATTEMPTS=25
+readonly NET_HTTP_READY_DELAY=0.2
+NET_HTTP_URL=""
+if command -v python3 &> /dev/null; then
+  ( cd "$NET_FIXTURE_DIR" \
+      && exec python3 -u -m http.server "$NET_HTTP_EPHEMERAL_PORT" --bind 127.0.0.1 ) \
+    > "$NET_HTTP_LOG" 2>&1 &
+  NET_HTTP_PID=$!
+  for _ in $(seq 1 "$NET_HTTP_READY_ATTEMPTS"); do
+    net_http_port="$(sed -n 's/.*port \([0-9][0-9]*\).*/\1/p' "$NET_HTTP_LOG" | head -1)"
+    if [ -n "$net_http_port" ]; then
+      NET_HTTP_URL="http://127.0.0.1:$net_http_port"
+      break
+    fi
+    sleep "$NET_HTTP_READY_DELAY"
+  done
+fi
+
 section "curl"
 if command -v curl &> /dev/null; then
-  bench "curl json" "curl -s https://mockhttp.org/json/1" "$RTK curl https://mockhttp.org/json/1"
-  bench "curl text" "curl -s https://mockhttp.org/robots.txt" "$RTK curl https://mockhttp.org/robots.txt"
+  if [ -n "$NET_HTTP_URL" ]; then
+    bench "curl json" "curl -s $NET_HTTP_URL/data.json" "$RTK curl $NET_HTTP_URL/data.json"
+    bench "curl text" "curl -s $NET_HTTP_URL/robots.txt" "$RTK curl $NET_HTTP_URL/robots.txt"
+  else
+    # No python3 to host a local server — fall back to file:// (no headers, so
+    # no JSON minification, but still fully offline and deterministic).
+    bench "curl json" "curl -s file://$NET_FIXTURE_DIR/data.json" "$RTK curl file://$NET_FIXTURE_DIR/data.json"
+    bench "curl text" "curl -s file://$NET_FIXTURE_DIR/robots.txt" "$RTK curl file://$NET_FIXTURE_DIR/robots.txt"
+  fi
 fi
 
-# ===================
-# wget
-# ===================
+# wget has no offline fallback: it rejects file:// outright ("Unsupported
+# scheme"), so without the loopback server there is nothing local to fetch. Say
+# so explicitly instead of letting the case disappear from the report.
 if command -v wget &> /dev/null; then
   section "wget"
-  bench "wget" "wget -qO- https://mockhttp.org/json/1" "$RTK wget https://mockhttp.org/json/1"
-  rm -f 1 2>/dev/null
+  if [ -n "$NET_HTTP_URL" ]; then
+    bench "wget" "wget -qO- $NET_HTTP_URL/data.json" "$RTK wget $NET_HTTP_URL/data.json"
+  else
+    echo "⏭️  wget (no local HTTP server available, skipped)"
+  fi
 fi
 
 # ===================