autogen: dump patch in CI (#14010)

* autogen: don't fast-fail, produce patch artifact on differences

All verification steps now use continue-on-error to run completely.
Each job generates a patch artifact containing all differences found.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

* add gen from header test

* fix tests

* fail if diff

* add forward decl autogen test

* remove confusing/wrong comments

* macos unittests set LIBCLANG_PATH

---------

Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>

.github/workflows/autogen.yml

@@ -40,11 +40,13 @@ jobs:
     - name: Install autogen support packages
       run: sudo apt-get install -y --no-install-recommends libclang-20-dev llvm-20-dev hip-dev libusb-1.0-0-dev
     - name: Verify OpenCL autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/opencl.py /tmp/opencl.py.bak
         python3 -c "from tinygrad.runtime.autogen import opencl"
         diff /tmp/opencl.py.bak tinygrad/runtime/autogen/opencl.py
     - name: Verify CUDA autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/cuda.py /tmp/cuda.py.bak
         mv tinygrad/runtime/autogen/nvrtc.py /tmp/nvrtc.py.bak
@@ -58,6 +60,7 @@ jobs:
         diff /tmp/nv_570.py.bak tinygrad/runtime/autogen/nv_570.py
         diff /tmp/nv.py.bak tinygrad/runtime/autogen/nv.py
     - name: Verify AMD autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/comgr.py /tmp/comgr.py.bak
         mv tinygrad/runtime/autogen/hsa.py /tmp/hsa.py.bak
@@ -89,6 +92,7 @@ jobs:
         diff /tmp/am_smu_v13_0_0.py.bak tinygrad/runtime/autogen/am/smu_v13_0_0.py
         diff /tmp/am_smu_v14_0_2.py.bak tinygrad/runtime/autogen/am/smu_v14_0_2.py
     - name: Verify Linux autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/libc.py /tmp/libc.py.bak
         mv tinygrad/runtime/autogen/kfd.py /tmp/kfd.py.bak
@@ -104,16 +108,19 @@ jobs:
         diff /tmp/pci.py.bak tinygrad/runtime/autogen/pci.py
         diff /tmp/vfio.py.bak tinygrad/runtime/autogen/vfio.py
     - name: Verify LLVM autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/llvm.py /tmp/llvm.py.bak
         python3 -c "from tinygrad.runtime.autogen import llvm"
         diff /tmp/llvm.py.bak tinygrad/runtime/autogen/llvm.py
     - name: Verify WebGPU autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/webgpu.py /tmp/webgpu.py.bak
         python3 -c "from tinygrad.runtime.autogen import webgpu"
         diff /tmp/webgpu.py.bak tinygrad/runtime/autogen/webgpu.py
     - name: Verify Qualcomm autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/kgsl.py /tmp/kgsl.py.bak
         mv tinygrad/runtime/autogen/qcom_dsp.py /tmp/qcom_dsp.py.bak
@@ -121,20 +128,36 @@ jobs:
         diff /tmp/kgsl.py.bak tinygrad/runtime/autogen/kgsl.py
         diff /tmp/qcom_dsp.py.bak tinygrad/runtime/autogen/qcom_dsp.py
     - name: Verify libusb autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/libusb.py /tmp/libusb.py.bak
         python3 -c "from tinygrad.runtime.autogen import libusb"
         diff /tmp/libusb.py.bak tinygrad/runtime/autogen/libusb.py
     - name: Verify mesa autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/mesa.py /tmp/mesa.py.bak
         python3 -c "from tinygrad.runtime.autogen import mesa"
         diff /tmp/mesa.py.bak tinygrad/runtime/autogen/mesa.py
     - name: Verify libclang autogen
+      continue-on-error: true
       run: |
         cp tinygrad/runtime/autogen/libclang.py /tmp/libclang.py.bak
         REGEN=1 python3 -c "from tinygrad.runtime.autogen import libclang"
         diff /tmp/libclang.py.bak tinygrad/runtime/autogen/libclang.py
+    - name: Generate patch for differences
+      run: |
+        if ! git diff --quiet; then
+          git diff > autogen-ubuntu.patch
+        fi
+    - name: Upload patch artifact
+      uses: actions/upload-artifact@v4
+      with:
+        name: autogen-ubuntu-patch
+        path: autogen-ubuntu.patch
+        if-no-files-found: ignore
+    - name: Fail if differences found
+      run: git diff --quiet
   autogen-mac:
     name: In-tree Autogen (macos)
     runs-on: macos-14
@@ -147,10 +170,24 @@ jobs:
       with:
         llvm: 'true'
     - name: Verify macos autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/metal.py /tmp/metal.py.bak
         LIBCLANG_PATH=/opt/homebrew/opt/llvm@20/lib/libclang.dylib python3 -c "from tinygrad.runtime.autogen import metal"
         diff /tmp/metal.py.bak tinygrad/runtime/autogen/metal.py
+    - name: Generate patch for differences
+      run: |
+        if ! git diff --quiet; then
+          git diff > autogen-macos.patch
+        fi
+    - name: Upload patch artifact
+      uses: actions/upload-artifact@v4
+      with:
+        name: autogen-macos-patch
+        path: autogen-macos.patch
+        if-no-files-found: ignore
+    - name: Fail if differences found
+      run: git diff --quiet
   autogen-comgr-3:
     name: In-tree Autogen (comgr 3)
     runs-on: ubuntu-24.04
@@ -170,7 +207,21 @@ jobs:
         sudo apt -qq update || true
         sudo apt-get install -y --no-install-recommends libclang-20-dev comgr
     - name: Verify comgr (3) autogen
+      continue-on-error: true
       run: |
         mv tinygrad/runtime/autogen/comgr_3.py /tmp/comgr_3.py.bak
         python3 -c "from tinygrad.runtime.autogen import comgr_3"
         diff /tmp/comgr_3.py.bak tinygrad/runtime/autogen/comgr_3.py
+    - name: Generate patch for differences
+      run: |
+        if ! git diff --quiet; then
+          git diff > autogen-comgr3.patch
+        fi
+    - name: Upload patch artifact
+      uses: actions/upload-artifact@v4
+      with:
+        name: autogen-comgr3-patch
+        path: autogen-comgr3.patch
+        if-no-files-found: ignore
+    - name: Fail if differences found
+      run: git diff --quiet

.github/workflows/test.yml

@@ -257,6 +257,7 @@ jobs:
         key: unittest-12
         pydeps: "pillow numpy ftfy regex"
         deps: testing_unit
+        llvm: 'true'
     - name: Check Device.DEFAULT
       run: python -c "from tinygrad import Device; assert Device.DEFAULT == 'CPU', Device.DEFAULT"
     - name: Run unit tests
@@ -309,7 +310,7 @@ jobs:
         deps: testing_unit
         python-version: '3.14'
     - name: Test SPEC=2
-      run: SPEC=2 pytest --maxfail=10 -n auto --durations=30 --ignore=test/models --ignore test/test_custom_kernel.py --ignore test/unit/test_hashing.py --timeout 60 -k "not test_setitem_big" --splits 2 --group ${{ matrix.group }}
+      run: SPEC=2 pytest --maxfail=10 -n auto --durations=30 --ignore=test/models --ignore test/test_custom_kernel.py --ignore test/unit/test_hashing.py --ignore test/unit/test_autogen.py --timeout 60 -k "not test_setitem_big" --splits 2 --group ${{ matrix.group }}
 
   fuzzing:
     name: Fuzzing
@@ -793,6 +794,8 @@ jobs:
         ocelot: 'true'
         llvm: 'true'
     - name: Run unit tests
+      env:
+        LIBCLANG_PATH: '/opt/homebrew/opt/llvm@20/lib/libclang.dylib'
       run: METAL=1 python -m pytest -n=auto test/unit/ --durations=20
     - name: Run ONNX
       run: METAL=1 python -m pytest -n=auto test/external/external_test_onnx_backend.py --durations=20

test/unit/test_autogen.py

@@ -1,6 +1,7 @@
 import ctypes, subprocess, tempfile, unittest
 from tinygrad.helpers import WIN
 from tinygrad.runtime.support.c import Struct
+from tinygrad.runtime.support.autogen import gen
 
 class TestAutogen(unittest.TestCase):
   def test_packed_struct_sizeof(self):
@@ -159,4 +160,97 @@ class TestAutogen(unittest.TestCase):
     assert ihdr.num_dies == 1
     assert ihdr.base_addr_64_bit == 1
 
+  @unittest.skipIf(WIN, "doesn't compile on windows")
+  def test_gen_from_header(self):
+    header_content = """
+    typedef struct {
+      int x;
+      int y;
+    } Point;
+
+    typedef enum {
+      RED = 0,
+      GREEN = 1,
+      BLUE = 2
+    } Color;
+
+    typedef struct {
+      Point origin;
+      int width;
+      int height;
+      Color color;
+    } Rectangle;
+
+    int add_points(Point a, Point b);
+    """
+
+    with tempfile.NamedTemporaryFile(mode='w', suffix='.h') as f:
+      f.write(header_content)
+      f.flush()
+
+      generated_code = gen(name="test_header", dll=None, files=[f.name])
+
+      namespace = {}
+      exec(generated_code, namespace)
+
+      self.assertIn('Point', namespace)
+      self.assertIn('Color', namespace)
+      self.assertIn('Rectangle', namespace)
+      self.assertIn('RED', namespace)
+      self.assertIn('GREEN', namespace)
+      self.assertIn('BLUE', namespace)
+
+      self.assertEqual(namespace['RED'], 0)
+      self.assertEqual(namespace['GREEN'], 1)
+      self.assertEqual(namespace['BLUE'], 2)
+
+      Point = namespace['Point']
+      p = Point()
+      self.assertIsInstance(p, Struct)
+      self.assertTrue(hasattr(p, 'x'))
+      self.assertTrue(hasattr(p, 'y'))
+
+      Rectangle = namespace['Rectangle']
+      rect = Rectangle()
+      self.assertTrue(hasattr(rect, 'origin'))
+      self.assertTrue(hasattr(rect, 'width'))
+      self.assertTrue(hasattr(rect, 'height'))
+      self.assertTrue(hasattr(rect, 'color'))
+
+  @unittest.skipIf(WIN, "doesn't compile on windows")
+  def test_struct_ordering(self):
+    header_content = """
+    struct A;
+    struct C;
+    typedef struct A A;
+
+    struct B {
+      struct C *c_ptr;
+    };
+
+    struct C {
+      struct A *a_ptr;
+    };
+
+    struct A {
+      int x;
+      struct B *b_ptr;
+    };
+    """
+    with tempfile.NamedTemporaryFile(mode='w', suffix='.h') as f:
+      f.write(header_content)
+      f.flush()
+      generated_code = gen(name="test_ordering", dll=None, files=[f.name])
+      namespace = {}
+      exec(generated_code, namespace)
+      self.assertIn('struct_A', namespace)
+      self.assertIn('struct_B', namespace)
+      self.assertIn('struct_C', namespace)
+      A, B, C = namespace['struct_A'], namespace['struct_B'], namespace['struct_C']
+      a, b, c = A(), B(), C()
+      self.assertTrue(hasattr(a, 'x'))
+      self.assertTrue(hasattr(a, 'b_ptr'))
+      self.assertTrue(hasattr(b, 'c_ptr'))
+      self.assertTrue(hasattr(c, 'a_ptr'))
+
 if __name__ == "__main__": unittest.main()