[Build][SM70] Fix the Docker source build and keep all SM70 extensions (#1207)

Two gaps in the Docker source build (docker/Dockerfile --target build):

- setup.py preloads vllm/envs_metadata.py before vllm/envs.py and installs
  scripts/serve_flash_next_nvfp4_v100.sh, but the native csrc-build stage
  copies only envs.py, __init__.py and the 27B serve script, so the stage
  stops with FileNotFoundError before compiling anything. Copy both files.
- The build stage reinstalls vLLM from the native-stage wheel with
  VLLM_USE_PRECOMPILED=1, and setup.py copies only the native libraries its
  extraction patterns name. The SM70 pattern lists sampler, exact_reduce,
  sparse_attention and qsa_device, so _sm70_qsa_indexer_C (#1127) and
  _sm70_router_C (#1147) are built but left out of the final wheel. Without
  the indexer, the QSA shared-key path reports native_extension_unavailable
  and falls back to the paged Triton indexer; the router operator tests
  skip. Add both names.

The extraction test now derives the expected libraries from the SM70/H3
CMakeExtension declarations in setup.py, so a newly declared extension
cannot be dropped silently again.

Signed-off-by: Leonccaa <166551845+Leonccaa@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

docker/Dockerfile

@@ -342,13 +342,14 @@ RUN --mount=type=cache,target=/opt/uv/cache \
 WORKDIR /workspace
 
 COPY pyproject.toml setup.py CMakeLists.txt ./
-COPY scripts/serve_qwen38_27b_nvfp4_v100.sh scripts/
+COPY scripts/serve_qwen38_27b_nvfp4_v100.sh scripts/serve_flash_next_nvfp4_v100.sh scripts/
 COPY cmake cmake/
 COPY csrc csrc/
 COPY flash_qla flash_qla/
 COPY flashinfer-sm70 flashinfer-sm70/
 COPY flash-attention-v100 flash-attention-v100/
 COPY vllm/envs.py vllm/envs.py
+COPY vllm/envs_metadata.py vllm/envs_metadata.py
 COPY vllm/__init__.py vllm/__init__.py
 
 # max jobs used by Ninja to build extensions

setup.py

@@ -949,8 +949,8 @@ class precompiled_wheel_utils:
                     r"flash_qla/ops/gated_delta_rule/chunk/sm70/[^/]+\.so"
                 )
                 sm70_ext_regex = re.compile(
-                    r"vllm/_sm70_(?:sampler|exact_reduce|sparse_attention|qsa_device)_C"
-                    r"(?:\.[^/]+)?\.so$"
+                    r"vllm/_sm70_(?:sampler|exact_reduce|sparse_attention|qsa_device"
+                    r"|qsa_indexer|router)_C(?:\.[^/]+)?\.so$"
                 )
                 h3_ext_regex = re.compile(
                     r"vllm/_h3_(?:w8a16|flashinfer|flashattn)_C(?:\.[^/]+)?\.so$"

tests/video/test_h3_host_memory.py

@@ -301,15 +301,34 @@ def test_source_wheel_reuse_keeps_sm70_native_operators(
         compile(ast.Module(body=[function], type_ignores=[]), str(source), "exec"),
         scope,
     )
-    wheel = tmp_path / "native.whl"
-    names = [
-        f"_h3_{kind}_C.cpython-312-x86_64-linux-gnu.so"
-        for kind in ("w8a16", "flashinfer", "flashattn")
-    ] + [
+    # The Docker build stage reinstalls from the native-stage wheel, so every
+    # SM70/H3 extension declared in setup.py must survive the extraction.
+    names: list[str] = []
+    for node in ast.walk(tree):
+        if not (
+            isinstance(node, ast.Call)
+            and getattr(node.func, "id", None) == "CMakeExtension"
+        ):
+            continue
+        keywords = {k.arg: k.value for k in node.keywords}
+        declared = keywords.get("name")
+        module = declared.value if isinstance(declared, ast.Constant) else None
+        if not (
+            isinstance(module, str)
+            and re.fullmatch(r"vllm\._(?:sm70|h3)_\w+_C", module)
+        ):
+            continue
+        limited = keywords.get("py_limited_api")
+        abi3 = not (isinstance(limited, ast.Constant) and limited.value is False)
+        suffix = ".abi3.so" if abi3 else ".cpython-312-x86_64-linux-gnu.so"
+        names.append(module.removeprefix("vllm.") + suffix)
+    assert {
         "_sm70_sampler_C.abi3.so",
-        "_sm70_exact_reduce_C.cpython-312-x86_64-linux-gnu.so",
-        "_sm70_sparse_attention_C.cpython-312-x86_64-linux-gnu.so",
-    ]
+        "_sm70_qsa_indexer_C.cpython-312-x86_64-linux-gnu.so",
+        "_sm70_router_C.cpython-312-x86_64-linux-gnu.so",
+        "_h3_w8a16_C.cpython-312-x86_64-linux-gnu.so",
+    } <= set(names)
+    wheel = tmp_path / "native.whl"
     with zipfile.ZipFile(wheel, "w") as archive:
         for name in names:
             archive.writestr("vllm/" + name, b"native-test-bytes")