Question

当我尝试运行此代码时：

from numba import njit, int64
from numba.types import List

@njit(int64(List(int64, True)))
def f(a):
    a[0] = 0
    return a[0]

@njit(List(int64)())
def g(): return [f([0])]

我得到以下内容。为什么是这样？如何正确识别类型？

Traceback (most recent call last):
  File "Temp.py", line 9, in <module>
    @njit(List(int64)())
  File "numba\decorators.py", line 172, in wrapper
    disp.compile(sig)
  File "numba\dispatcher.py", line 350, in compile
    flags=flags, locals=self.locals)
  File "numba\compiler.py", line 644, in compile_extra
    return pipeline.compile_extra(func)
  File "numba\compiler.py", line 361, in compile_extra
    return self.compile_bytecode(bc, func_attr=self.func_attr)
  File "numba\compiler.py", line 370, in compile_bytecode
    return self._compile_bytecode()
  File "numba\compiler.py", line 631, in _compile_bytecode
    return pm.run(self.status)
  File "numba\compiler.py", line 251, in run
    raise patched_exception
numba.errors.TypingError: Caused By:
Traceback (most recent call last):
  File "numba\compiler.py", line 243, in run
    res = stage()
  File "numba\compiler.py", line 458, in stage_nopython_frontend
    self.locals)
  File "numba\compiler.py", line 759, in type_inference_stage
    infer.propagate()
  File "numba\typeinfer.py", line 510, in propagate
    raise errors[0]
TypingError: Internal error at <numba.typeinfer.CallConstraint object at 0x00000000071C82E8>:
--%<-----------------------------------------------------------------
Traceback (most recent call last):
  File "numba\typeinfer.py", line 111, in propagate
    constraint(typeinfer)
  File "numba\typeinfer.py", line 270, in __call__
    self.resolve(typeinfer, typevars, fnty)
  File "numba\typeinfer.py", line 297, in resolve
    sig = context.resolve_function_type(fnty, pos_args, kw_args)
  File "numba\typing\context.py", line 113, in resolve_function_type
    return func.get_call_type(self, args, kws)
  File "numba\types.py", line 334, in get_call_type
    sig.pysig = self.pysig
AttributeError: 'NoneType' object has no attribute 'pysig'
--%<-----------------------------------------------------------------
File "Temp.py", line 10

Answer 1

将f的装饰器签名从@njit(int64(List(int64, True)))更改为@njit(int64(List(int64)))会让Numba 0.31.0上的内容更有效。

对于f，a不是反映列表，因此将该标记设置为True似乎会导致问题。对于Numba 0.31.0，我收到了不同的错误消息。

总的来说，我会允许Numba对输入和输出进行自己的类型推断。我不确定您使用的是哪个版本，但在最近的版本中，很少有理由在装饰器签名中定义类型。如果您对类型推断器的作用感到好奇，只需使用nb.njit或nb.jit(nopython=True)，使用所需的输入调用该函数，然后检查f.inspect_types()（如果您的jitted函数为{ {1}}）。

是什么导致了Numba的AttributeError：'NoneType'对象没有属性'pysig'？

1 个答案: