fix: remove Windows from benchmark, fix measure() timing bugs

- Remove windows-latest from all 3 benchmark jobs (ubuntu-only)
- Replace broken TIMEFORMAT/time subshell with portable python3 timing
- Fix ((failures++)) returning exit code 1 under set -e
- Fix measure() return 1 killing script under set -e
- Update run.sh with same measure/avg_runs fixes
- Update snippet count glob to include .yaml/.yml

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: brunoborges

.github/workflows/benchmark.yml

@@ -9,10 +9,7 @@ permissions:
 
 jobs:
   benchmark:
-    strategy:
-      matrix:
-        os: [ubuntu-latest, windows-latest]
-    runs-on: ${{ matrix.os }}
+    runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v6
 
@@ -28,8 +25,7 @@ jobs:
         with:
           python-version: '3.13'
 
-      - name: Install system dependencies (Ubuntu)
-        if: runner.os == 'Linux'
+      - name: Install system dependencies
         run: sudo apt-get update && sudo apt-get install -y libcairo2-dev
 
       - name: Install Python dependencies
@@ -46,41 +42,39 @@ jobs:
 
           snippet_count=$(find content \( -name '*.json' -o -name '*.yaml' -o -name '*.yml' \) -not -name 'template.*' | wc -l | tr -d ' ')
           java_ver=$(java -version 2>&1 | head -1 | sed 's/.*"\(.*\)".*/\1/')
-          os_name="${{ matrix.os }}"
+          os_name="ubuntu-latest"
 
-          # Timing helper using bash TIMEFORMAT
+          # Nanosecond timestamp helper
+          _now() { python3 -c "import time; print(int(time.time()*1e9))"; }
+
+          # Timing helper — returns seconds or "FAIL"
           measure() {
-            TIMEFORMAT='%R'
-            local t ec
-            t=$( { time "$@" > /dev/null 2>&1; ec=$?; } 2>&1 )
-            if [[ ${ec:-1} -ne 0 ]]; then
-              # Re-check: the subshell exit code isn't always captured
-              if ! "$@" > /dev/null 2>&1; then
-                echo "FAIL"
-                return 1
-              fi
+            local start end
+            start=$(_now)
+            if "$@" > /dev/null 2>&1; then
+              end=$(_now)
+              awk "BEGIN {printf \"%.2f\", ($end - $start) / 1000000000}"
+            else
+              echo "FAIL"
             fi
-            echo "$t"
           }
 
           avg_runs() {
             local n="$1"; shift
-            local sum=0 failures=0
+            local sum=0 success=0
             for ((i = 1; i <= n; i++)); do
               local t
-              t=$(measure "$@") || true
-              if [[ "$t" == "FAIL" ]]; then
-                ((failures++)) || true
-                continue
+              t=$(measure "$@")
+              if [[ "$t" != "FAIL" ]]; then
+                sum=$(awk "BEGIN {print $sum + $t}")
+                success=$((success + 1))
               fi
-              sum=$(awk "BEGIN {print $sum + $t}")
             done
-            local success=$((n - failures))
             if [[ $success -eq 0 ]]; then
               echo "FAIL"
-              return 0
+            else
+              awk "BEGIN {printf \"%.2f\", $sum / $success}"
             fi
-            awk "BEGIN {printf \"%.2f\", $sum / $success}"
           }
 
           echo "Running benchmark on $os_name (Java $java_ver, $snippet_count snippets)..."
@@ -164,10 +158,7 @@ jobs:
   # Python and JBang start with zero caches, just like real CI.
   # ---------------------------------------------------------------------------
   build-jar:
-    strategy:
-      matrix:
-        os: [ubuntu-latest, windows-latest]
-    runs-on: ${{ matrix.os }}
+    runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v6
 
@@ -188,7 +179,7 @@ jobs:
       - name: Upload JAR and AOT
         uses: actions/upload-artifact@v4
         with:
-          name: generator-${{ matrix.os }}
+          name: generator
           path: |
             html-generators/generate.jar
             html-generators/generate.aot
@@ -197,10 +188,7 @@ jobs:
 
   ci-cold-start:
     needs: build-jar
-    strategy:
-      matrix:
-        os: [ubuntu-latest, windows-latest]
-    runs-on: ${{ matrix.os }}
+    runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v6
 
@@ -214,8 +202,7 @@ jobs:
         with:
           python-version: '3.13'
 
-      - name: Install system dependencies (Ubuntu)
-        if: runner.os == 'Linux'
+      - name: Install system dependencies
         run: sudo apt-get update && sudo apt-get install -y libcairo2-dev
 
       - name: Install Python dependencies
@@ -224,27 +211,27 @@ jobs:
       - name: Download JAR and AOT
         uses: actions/download-artifact@v4
         with:
-          name: generator-${{ matrix.os }}
+          name: generator
           path: html-generators
 
       - name: CI cold-start benchmark
         shell: bash
         run: |
-          os_name="${{ matrix.os }}"
+          os_name="ubuntu-latest"
           java_ver=$(java -version 2>&1 | head -1 | sed 's/.*"\(.*\)".*/\1/')
           snippet_count=$(find content \( -name '*.json' -o -name '*.yaml' -o -name '*.yml' \) -not -name 'template.*' | wc -l | tr -d ' ')
 
+          _now() { python3 -c "import time; print(int(time.time()*1e9))"; }
+
           measure() {
-            TIMEFORMAT='%R'
-            local t ec
-            t=$( { time "$@" > /dev/null 2>&1; ec=$?; } 2>&1 )
-            if [[ ${ec:-1} -ne 0 ]]; then
-              if ! "$@" > /dev/null 2>&1; then
-                echo "FAIL"
-                return 1
-              fi
+            local start end
+            start=$(_now)
+            if "$@" > /dev/null 2>&1; then
+              end=$(_now)
+              awk "BEGIN {printf \"%.2f\", ($end - $start) / 1000000000}"
+            else
+              echo "FAIL"
             fi
-            echo "$t"
           }
 
           # Everything is cold: no __pycache__, no JBang cache, fresh JVM

html-generators/benchmark/run.sh

@@ -31,34 +31,32 @@ UPDATE_MD=false
 # Helpers
 # ---------------------------------------------------------------------------
 measure() {
-  local t
-  t=$( { /usr/bin/time -p "$@" > /dev/null; } 2>&1 | awk '/^real/ {print $2}' )
-  # Verify the command actually succeeded
-  if ! "$@" > /dev/null 2>&1; then
+  local start end
+  start=$(python3 -c "import time; print(time.time())")
+  if "$@" > /dev/null 2>&1; then
+    end=$(python3 -c "import time; print(time.time())")
+    python3 -c "print(f'{$end - $start:.2f}')"
+  else
     echo "FAIL"
-    return 1
   fi
-  echo "$t"
 }
 
 avg_runs() {
   local n="$1"; shift
-  local sum=0 failures=0
+  local sum=0 success=0
   for ((i = 1; i <= n; i++)); do
     local t
-    t=$(measure "$@") || true
-    if [[ "$t" == "FAIL" ]]; then
-      ((failures++)) || true
-      continue
+    t=$(measure "$@")
+    if [[ "$t" != "FAIL" ]]; then
+      sum=$(echo "$sum + $t" | bc)
+      success=$((success + 1))
     fi
-    sum=$(echo "$sum + $t" | bc)
   done
-  local success=$((n - failures))
   if [[ $success -eq 0 ]]; then
     echo "FAIL"
-    return 0
+  else
+    echo "scale=2; $sum / $success" | bc | sed 's/^\./0./'
   fi
-  echo "scale=2; $sum / $success" | bc | sed 's/^\./0./'
 }
 
 # ---------------------------------------------------------------------------
@@ -70,7 +68,7 @@ JAVA_VER=$(java -version 2>&1 | head -1 | sed 's/.*"\(.*\)".*/\1/')
 JBANG_VER=$(jbang version 2>/dev/null || echo "n/a")
 PYTHON_VER=$(python3 --version 2>/dev/null | awk '{print $2}' || echo "n/a")
 OS=$(uname -s)
-SNIPPET_COUNT=$(find content -name '*.json' | wc -l | tr -d ' ')
+SNIPPET_COUNT=$(find content \( -name '*.json' -o -name '*.yaml' -o -name '*.yml' \) -not -name 'template.*' | wc -l | tr -d ' ')
 
 echo ""
 echo "Environment: $CPU · $RAM · Java $JAVA_VER · $OS"