fix(pr): resolve pre-commit failures for #1955

- pyright: assert external kernel source(s) non-None in the specializer's
  _render_external so `str | None` narrows to `str`
- pyright: assert pl.parse(...) returns ir.Program before run_passes in the
  JIT extern test
- ruff PLC0415: hoist in-function imports to top level in the JIT extern test
- ruff PLR0913: mark build_specialize_context (pass-through assembler) noqa
- cpplint build/include_subdir: NOLINT the runtime headers in the ST kernel
  fixture (they resolve via the runtime include path)
- ruff-format / clang-format: apply formatting

Author: lyfne123

docs/zh-cn/dev/language/01-external-kernels.md

@@ -95,7 +95,7 @@ Python stub 未变，kernel 变更也会触发重新编译。
   仅豁免 `ReturnParamsExplicit` 属性 —— 该属性要求存在 `ReturnStmt`，而函数头本就没有。
 - **Orchestration 代码生成**：为该 kernel 分配 `func_id`，并像 DSL kernel 一样生成派发
   —— 单核 AIC/AIV 的 `rt_submit_*_task`，或 group 的 `MixedKernels{aic_id, aiv_id, ...}`
-  + `rt_submit_task`。
+  - `rt_submit_task`。
 - **Backend**：对外部 kernel 跳过 ptoas，并把其 `.cpp` 复制到
   `kernels/<aic|aiv>/<name>.cpp`，因此生成的 `kernel_config.py` manifest 会像列出生成
   kernel 一样列出它（`func_id`、`source`、`core_type`、`signature`、`arg_index`）。

python/pypto/jit/specializer.py

@@ -1679,14 +1679,19 @@ def _member(member_name: str, core_upper: str, source: str) -> list[str]:
                 "    ...",
             ]
 
+        # The @pl.jit.extern decorator guarantees the source(s) for the selected
+        # core_type are set; assert to narrow str | None -> str for the type checker.
         core = ctx.external_core_type
         if core == "aic":
+            assert ctx.external_aic_source is not None
             return _member(name, "AIC", ctx.external_aic_source)
         if core == "aiv":
+            assert ctx.external_aiv_source is not None
             return _member(name, "AIV", ctx.external_aiv_source)
 
         # Mixed: two members + a Group wrapper named after the extern so the
         # entry's ``self.<name>(...)`` call resolves to the group.
+        assert ctx.external_aic_source is not None and ctx.external_aiv_source is not None
         lines = _member(f"{name}_aic", "AIC", ctx.external_aic_source)
         lines += _member(f"{name}_aiv", "AIV", ctx.external_aiv_source)
         lines.append("@pl.function(type=pl.FunctionType.Group)")
@@ -1825,7 +1830,7 @@ def specialize(class_name: str, contexts: list[SpecializeContext]) -> str:
     return Specializer(class_name, contexts).specialize()
 
 
-def build_specialize_context(
+def build_specialize_context(  # noqa: PLR0913 — pass-through assembler; each arg maps to a SpecializeContext field
     func: Any,
     func_name: str,
     func_type: str | None,

python/pypto/language/parser/ast_parser.py

@@ -92,11 +92,7 @@ def _is_empty_body(body: list[ast.stmt]) -> bool:
     only = non_doc[0]
     if isinstance(only, ast.Pass):
         return True
-    return (
-        isinstance(only, ast.Expr)
-        and isinstance(only.value, ast.Constant)
-        and only.value.value is ...
-    )
+    return isinstance(only, ast.Expr) and isinstance(only.value, ast.Constant) and only.value.value is ...
 
 
 def _is_pld_call(node: object, attr_name: str) -> TypeGuard[ast.Call]:

runtime

@@ -26,7 +26,7 @@
 #include <cstdint>
 #include <pto/pto-inst.hpp>
 
-#include "tensor.h"
+#include "tensor.h"  // NOLINT(build/include_subdir)
 
 #ifndef __gm__
 #define __gm__
@@ -36,14 +36,14 @@
 #define __aicore__ [aicore]  // NOLINT(whitespace/braces)
 #endif
 
-#include "intrinsic.h"
+#include "intrinsic.h"  // NOLINT(build/include_subdir)
 
 static constexpr int32_t FLOATS_PER_CACHE_LINE = 16;
 
 #ifdef PTO_CPUSTUB_HPP
 #define dcci(...) \
-    do {          \
-    } while (0)
+  do {            \
+  } while (0)
 #endif
 #ifndef SINGLE_CACHE_LINE
 #define SINGLE_CACHE_LINE 0
@@ -52,15 +52,15 @@ static constexpr int32_t FLOATS_PER_CACHE_LINE = 16;
 #define CACHELINE_OUT 0
 #endif
 
-extern "C" __aicore__ void kernel_entry(__gm__ int64_t *args) {
-    __gm__ Tensor *out_tensor = reinterpret_cast<__gm__ Tensor *>(args[0]);
-    __gm__ float *out = reinterpret_cast<__gm__ float *>(out_tensor->buffer.addr) + out_tensor->start_offset;
+extern "C" __aicore__ void kernel_entry(__gm__ int64_t* args) {
+  __gm__ Tensor* out_tensor = reinterpret_cast<__gm__ Tensor*>(args[0]);
+  __gm__ float* out = reinterpret_cast<__gm__ float*>(out_tensor->buffer.addr) + out_tensor->start_offset;
 
-    int32_t base_cl = static_cast<int32_t>(args[1]);
-    int32_t block_idx = get_block_idx(args);
-    int32_t offset = (base_cl + block_idx) * FLOATS_PER_CACHE_LINE;
+  int32_t base_cl = static_cast<int32_t>(args[1]);
+  int32_t block_idx = get_block_idx(args);
+  int32_t offset = (base_cl + block_idx) * FLOATS_PER_CACHE_LINE;
 
-    out[offset] = static_cast<float>(block_idx);
+  out[offset] = static_cast<float>(block_idx);
 
-    dcci(&out[offset], SINGLE_CACHE_LINE, CACHELINE_OUT);
+  dcci(&out[offset], SINGLE_CACHE_LINE, CACHELINE_OUT);
 }

tests/st/runtime/external_kernel/kernels/aiv/spmd_write.cpp

@@ -47,8 +47,7 @@ def SPMD_WRITE_AIV(
         self,
         out: pl.InOut[pl.Tensor[[_TOTAL], pl.FP32]],
         base_cl: pl.Scalar[pl.INDEX],
-    ) -> pl.Tensor[[_TOTAL], pl.FP32]:
-        ...  # implementation lives in kernels/aiv/spmd_write.cpp
+    ) -> pl.Tensor[[_TOTAL], pl.FP32]: ...  # implementation lives in kernels/aiv/spmd_write.cpp
 
     @pl.function(type=pl.FunctionType.Orchestration)
     def main(self, out: pl.Out[pl.Tensor[[_TOTAL], pl.FP32]]) -> pl.Tensor[[_TOTAL], pl.FP32]:

tests/st/runtime/external_kernel/test_external_kernel.py

@@ -23,10 +23,7 @@
 import pytest
 from pypto.ir.compile import compile as ir_compile
 
-_KERNEL_SRC = (
-    "#include <cstdint>\n"
-    'extern "C" void kernel_entry(int64_t* args) { (void)args; }\n'
-)
+_KERNEL_SRC = '#include <cstdint>\nextern "C" void kernel_entry(int64_t* args) { (void)args; }\n'
 
 
 def _write_kernel(tmp_path: Path) -> Path:

tests/ut/codegen/test_external_kernel.py

@@ -21,12 +21,11 @@
 import pypto.language as pl
 import pytest
 import torch
+from pypto.ir.pass_manager import OptimizationStrategy, PassManager
 from pypto.jit.specializer import Specializer
+from pypto.pypto_core import ir
 
-_KERNEL_SRC = (
-    "#include <cstdint>\n"
-    'extern "C" void kernel_entry(int64_t* args) { (void)args; }\n'
-)
+_KERNEL_SRC = '#include <cstdint>\nextern "C" void kernel_entry(int64_t* args) { (void)args; }\n'
 
 
 def _write_kernel(tmp_path: Path, name: str = "ext.cpp") -> Path:
@@ -71,9 +70,8 @@ def entry(a: pl.Tensor, out: pl.Out[pl.Tensor]) -> pl.Tensor:
     assert "self.pa(a, out)" in src
 
     # The generated program parses and survives the full pass pipeline.
-    from pypto.ir.pass_manager import OptimizationStrategy, PassManager
-
     program = pl.parse(src)
+    assert isinstance(program, ir.Program)
     after = PassManager.get_strategy(OptimizationStrategy.Default).run_passes(program)
     names = {f.name for f in after.functions.values()}
     assert {"entry", "pa", "pa_aic", "pa_aiv"} <= names