style: fix lint issues in feat/ascend-framework

- Fix `ruff format` violations in `generate_wrappers.py` and `test_gemm.py`.
- Fix `ruff isort` violation: move `import re` into stdlib group.
- Add backticks around identifiers in comments (`numel()`, `operator()`,
  `make()`, `torch_npu`, `uint16`/`uint32`/`uint64`).
- Add missing blank line after `if` block in `skip_unsupported_dtype`.
- Remove `.worktrees/` from project `.gitignore` (belongs in global gitignore).

Author: zhangyue

.gitignore

@@ -1,7 +1,6 @@
 # Generated files
 build/
 generated/
-.worktrees/
 
 # Prerequisites
 *.d

scripts/generate_wrappers.py

@@ -1,12 +1,11 @@
 import argparse
 import json
 import pathlib
+import re
 import shutil
 import subprocess
 import textwrap
 
-import re
-
 import clang.cindex
 from clang.cindex import CursorKind
 
@@ -120,10 +119,8 @@ def _generate_params(node):
             if _is_optional_tensor(arg):
                 parts.append(f"std::optional<py::object> {arg.spelling}")
             else:
-                param = (
-                    arg.type.spelling
-                    .replace("const Tensor", "py::object")
-                    .replace("Tensor", "py::object")
+                param = arg.type.spelling.replace("const Tensor", "py::object").replace(
+                    "Tensor", "py::object"
                 )
                 parts.append(f"{param} {arg.spelling}")
 
@@ -136,9 +133,7 @@ def _generate_arguments(node):
             if arg.spelling == "stream":
                 continue
             if _is_optional_tensor(arg):
-                args.append(
-                    f"OptionalTensorFromPybind11Handle({arg.spelling})"
-                )
+                args.append(f"OptionalTensorFromPybind11Handle({arg.spelling})")
             elif "Tensor" in arg.type.spelling:
                 args.append(f"TensorFromPybind11Handle({arg.spelling})")
             else:
@@ -184,7 +179,7 @@ def _generate_call(op_name, call, method=True):
                 f"      handle.set_stream(reinterpret_cast<void*>(stream));\n"
                 f"    }}\n"
                 f"    return Self::call(handle, config, {call_args});\n"
-                f"  }}, {py_args_str}py::kw_only(), py::arg(\"implementation_index\") = 0, py::arg(\"stream\") = 0);"
+                f'  }}, {py_args_str}py::kw_only(), py::arg("implementation_index") = 0, py::arg("stream") = 0);'
             )
 
         return f"""      .def("__call__", [](const Self& self, {call_params}) {{

src/ascend/common.h

@@ -28,7 +28,7 @@ inline aclTensor* buildAclTensor(const Tensor& t,
   }
 
   // Compute the minimum physical storage needed for this strided view.
-  // For contiguous tensors this equals numel(); for non-contiguous (gapped)
+  // For contiguous tensors this equals `numel()`; for non-contiguous (gapped)
   // tensors it may be larger; for broadcast (stride-0) tensors it may be
   // smaller.  Passing the view shape as the storage shape causes
   // "ViewShape overlap" errors in ACLNN for non-contiguous inputs.

src/operator.h

@@ -177,8 +177,8 @@ class Operator : public OperatorBase {
     auto it{cache.find(key)};
 
     if (it == cache.end()) {
-      // Pass args as lvalue refs so they remain valid for the operator() call
-      // below. Forwarding rvalue temporaries into make() would leave the args
+      // Pass args as lvalue refs so they remain valid for the `operator()` call
+      // below. Forwarding rvalue temporaries into `make()` would leave the args
       // in a moved-from (empty) state before operator() can use them.
       it = cache.emplace(std::move(key), make(config, args...)).first;
     }

tests/conftest.py

@@ -46,7 +46,7 @@ def set_seed_per_test(request):
 
 _NPU_UNSUPPORTED_DTYPES = {torch.float64}
 
-# torch_npu does not implement random number generation for uint16/uint32/uint64.
+# `torch_npu` does not implement random number generation for `uint16`/`uint32`/`uint64`.
 for _bits in (16, 32, 64):
     _t = getattr(torch, f"uint{_bits}", None)
     if _t is not None:
@@ -57,6 +57,7 @@ def set_seed_per_test(request):
 def skip_unsupported_dtype(request):
     if not hasattr(request.node, "callspec"):
         return
+
     params = request.node.callspec.params
 
     if params.get("device") == "npu" and params.get("dtype") in _NPU_UNSUPPORTED_DTYPES:

tests/test_gemm.py

@@ -86,7 +86,13 @@ def test_gemm(
 def _gemm(a, b, alpha, beta, trans_a, trans_b, c, implementation_index=0):
     if a.device.type == "npu":
         infini.ops.gemm(
-            a, b, alpha, beta, trans_a, trans_b, c,
+            a,
+            b,
+            alpha,
+            beta,
+            trans_a,
+            trans_b,
+            c,
             stream=get_npu_stream(a),
         )
     else: