chore: add background binary wrapper and standalone adapter-cycle probe to CI integration tests

Author: Shreemanarjun

.github/workflows/integration_test.yml

@@ -107,19 +107,36 @@ jobs:
           ulimit -c unlimited
           # Wrap the app binary so its stderr (Rust panic messages — the
           # flutter tool discards child stderr) lands in a file we can print.
+          # `flutter test` reinstalls the bundle at the start of every
+          # invocation, so a background watcher re-wraps the fresh binary in
+          # the build→launch gap.
           BUNDLE=build/linux/x64/debug/bundle
-          if [ -f "$BUNDLE/nitro_webgpu_example" ] && [ ! -f "$BUNDLE/nitro_webgpu_example.real" ]; then
-            mv "$BUNDLE/nitro_webgpu_example" "$BUNDLE/nitro_webgpu_example.real"
-            cat > "$BUNDLE/nitro_webgpu_example" <<'WRAP'
+          wrap_binary() {
+            if [ -f "$BUNDLE/nitro_webgpu_example" ] \
+              && ! head -c2 "$BUNDLE/nitro_webgpu_example" | grep -q '#!'; then
+              mv "$BUNDLE/nitro_webgpu_example" "$BUNDLE/nitro_webgpu_example.real"
+              cat > "$BUNDLE/nitro_webgpu_example" <<'WRAP'
           #!/bin/bash
           exec "$(dirname "$0")/nitro_webgpu_example.real" "$@" 2>>/tmp/app_stderr.log
           WRAP
-            chmod +x "$BUNDLE/nitro_webgpu_example"
-          fi
+              chmod +x "$BUNDLE/nitro_webgpu_example"
+              echo "(re)wrapped app binary"
+            fi
+          }
+          wrap_binary
+          ( while true; do wrap_binary; sleep 0.3; done ) &
+          WATCHER=$!
           xvfb-run -a -s "-screen 0 1280x800x24" flutter test integration_test/wgpu_instance_test.dart -d linux \
             --dart-define=WGPU_FORCE_FALLBACK=true --reporter expanded || true
+          kill $WATCHER 2>/dev/null || true
           echo "=== app stderr ==="
           tail -80 /tmp/app_stderr.log 2>/dev/null || echo "(no stderr captured)"
+          echo "=== standalone adapter-cycle probe (wgpu panic goes to visible stderr) ==="
+          cc ../scripts/ci/adapter_cycle_probe.c \
+            ../src/third_party/wgpu_native/linux-x86_64/lib/libwgpu_native.a \
+            -I../src/third_party/wgpu_native/include -lm -ldl -lpthread \
+            -o /tmp/adapter_cycle_probe
+          /tmp/adapter_cycle_probe 2>&1 | tail -8 || echo "PROBE CRASHED (exit $?)"
           shopt -s nullglob
           cores=(/tmp/core.*)
           if [ ${#cores[@]} -eq 0 ]; then echo "NO CORE PRODUCED"; exit 0; fi

scripts/ci/adapter_cycle_probe.c

@@ -0,0 +1,85 @@
+// Standalone CI probe: repeated fallback-adapter + device acquire/release
+// cycles on one WGPUInstance — the exact pattern the integration suite runs
+// before wgpuInstanceRequestAdapter starts panicking on lavapipe (CI Linux,
+// core-dump verified). Running it outside Flutter keeps wgpu's panic message
+// on a visible stderr.
+//
+// Build (from the repo root, after fetch_wgpu_native.sh):
+//   cc scripts/ci/adapter_cycle_probe.c \
+//      src/third_party/wgpu_native/linux-x86_64/lib/libwgpu_native.a \
+//      -Isrc/third_party/wgpu_native/include -lm -ldl -lpthread \
+//      -o /tmp/adapter_cycle_probe
+#include <stdio.h>
+#include <webgpu/webgpu.h>
+
+static WGPUAdapter g_adapter;
+static WGPUDevice g_device;
+
+static void onAdapter(WGPURequestAdapterStatus status, WGPUAdapter adapter,
+                      WGPUStringView message, void* u1, void* u2) {
+    (void)u1; (void)u2;
+    if (status != WGPURequestAdapterStatus_Success) {
+        fprintf(stderr, "adapter error: %.*s\n", (int)message.length,
+                message.data ? message.data : "");
+    }
+    g_adapter = adapter;
+}
+
+static void onDevice(WGPURequestDeviceStatus status, WGPUDevice device,
+                     WGPUStringView message, void* u1, void* u2) {
+    (void)u1; (void)u2;
+    if (status != WGPURequestDeviceStatus_Success) {
+        fprintf(stderr, "device error: %.*s\n", (int)message.length,
+                message.data ? message.data : "");
+    }
+    g_device = device;
+}
+
+int main(void) {
+    WGPUInstance instance = wgpuCreateInstance(NULL);
+    if (!instance) {
+        fprintf(stderr, "wgpuCreateInstance failed\n");
+        return 1;
+    }
+    for (int i = 0; i < 40; i++) {
+        g_adapter = NULL;
+        g_device = NULL;
+
+        WGPURequestAdapterOptions opts = WGPU_REQUEST_ADAPTER_OPTIONS_INIT;
+        opts.forceFallbackAdapter = 1;
+        WGPURequestAdapterCallbackInfo acb =
+            WGPU_REQUEST_ADAPTER_CALLBACK_INFO_INIT;
+        acb.mode = WGPUCallbackMode_AllowProcessEvents;
+        acb.callback = onAdapter;
+        wgpuInstanceRequestAdapter(instance, &opts, acb);
+        for (int t = 0; t < 1000 && !g_adapter; t++) {
+            wgpuInstanceProcessEvents(instance);
+        }
+        if (!g_adapter) {
+            fprintf(stderr, "iter %d: no adapter\n", i);
+            return 1;
+        }
+
+        WGPUDeviceDescriptor dd = WGPU_DEVICE_DESCRIPTOR_INIT;
+        WGPURequestDeviceCallbackInfo dcb =
+            WGPU_REQUEST_DEVICE_CALLBACK_INFO_INIT;
+        dcb.mode = WGPUCallbackMode_AllowProcessEvents;
+        dcb.callback = onDevice;
+        wgpuAdapterRequestDevice(g_adapter, &dd, dcb);
+        for (int t = 0; t < 1000 && !g_device; t++) {
+            wgpuInstanceProcessEvents(instance);
+        }
+        if (!g_device) {
+            fprintf(stderr, "iter %d: no device\n", i);
+            return 1;
+        }
+
+        WGPUQueue queue = wgpuDeviceGetQueue(g_device);
+        wgpuQueueRelease(queue);
+        wgpuDeviceRelease(g_device);
+        wgpuAdapterRelease(g_adapter);
+        fprintf(stderr, "iter %d ok\n", i);
+    }
+    fprintf(stderr, "PROBE PASSED: 40 adapter+device cycles\n");
+    return 0;
+}