报错信息处理

    在动态图代码转换成静态图代码的过程中，如果 ProgramTranslator 无法转换一个函数时，将会显示警告信息，并尝试直接运行该函数。 如下代码中，函数 在调用前被转换成静态图代码，当 x = inner_func(data) 调用该函数时，不能重复转换，会给出警告信息：

    ProgramTranslator打印的警告信息如下：

    1. 2020-01-01 00:00:00,104-WARNING: <function inner_func at 0x125b3a550> doesn't have to be transformed to static function because it has been transformed before, it will be run as-is.

    运行转换后的代码报错

    如果在动转静后的静态图代码中发生异常，ProgramTranslator 会捕获该异常，增强异常报错信息，将静态图代码报错行映射到转换前的动态图代码，并重新抛出该异常。 重新抛出的异常具有以下特点：

    • 隐藏了部分对用户无用的动转静过程调用栈；

    • 转换前的代码会给出提示：”In User Code:”；

    • 报错信息中包含了转换前的原始动态图代码；

    运行结果：

    1. Traceback (most recent call last):
    2.   <ipython-input-13-f9c3ea702e3a> in <module>()
    3.      func(np.ones([3, 2]))
    4.   File "paddle/fluid/dygraph/dygraph_to_static/program_translator.py", line 332, in __call__
    5.     raise new_exception
    6. AssertionError: In user code:
    8.     File "<ipython-input-13-f9c3ea702e3a>", line 7, in func
    9.         x = fluid.layers.reshape(x, shape=[-1, -1])
    10.     File "paddle/fluid/layers/nn.py", line 6193, in reshape
    11.         attrs["shape"] = get_attr_shape(shape)
    12.     File "paddle/fluid/layers/nn.py", line 6169, in get_attr_shape
    13.         "be -1. But received shape[%d] is also -1." % dim_idx)
    14.     AssertionError: Only one dimension value of 'shape' in reshape can be -1. But received shape[1] is also -1.

    上述报错信息可以分为3点：

    1. 报错栈中，涉及代码转换过程的信息栈默认会被隐藏，不进行展示，以减少干扰信息。

    2. ProgramTranslator 处理后的报错信息中，会包含提示 “In user code:”，表示之后的报错栈中，包含动转静前的动态图代码，即用户写的代码：

      其中，File "<ipython-input-13-f9c3ea702e3a>", line 7, in func 是转换前的代码位置信息，x = fluid.layers.reshape(x, shape=[-1, -1]) 是转换前的代码。

    运行以下代码，在静态图运行期会抛出异常：

    运行结果：

    1. Traceback (most recent call last):
    3.      func(np.ones([3]).astype("int32"))
    4.   File "paddle/fluid/dygraph/dygraph_to_static/program_translator.py", line 332, in __call__
    5.     raise new_exception
    7. EnforceNotMet: In user code:
    9.     File "<ipython-input-57-c63d6a351262>", line 7, in func
    10.       x = paddle.reshape(x, shape=[1, two])
    11.     File "paddle/tensor/manipulation.py", line 1347, in reshape
    12.       return paddle.fluid.layers.reshape(x=x, shape=shape, name=name)
    13.     File "paddle/fluid/layers/nn.py", line 6209, in reshape
    14.       "XShape": x_shape})
    15.     File "paddle/fluid/layer_helper.py", line 43, in append_op
    16.       return self.main_program.current_block().append_op(*args, **kwargs)
    17.     File "paddle/fluid/framework.py", line 2880, in append_op
    18.       attrs=kwargs.get("attrs", None))
    19.     File "paddle/fluid/framework.py", line 1977, in __init__
    20.       for frame in traceback.extract_stack():
    22. C++ Traceback (most recent call last):
    24. 0   paddle::imperative::Tracer::TraceOp(std::string const&, paddle::imperative::NameVarBaseMap const&, paddle::imperative::NameVarBaseMap const&, paddle::framework::AttributeMap, paddle::platform::Place const&, bool)
    25. 1   paddle::imperative::OpBase::Run(paddle::framework::OperatorBase const&, paddle::imperative::NameVarBaseMap const&, paddle::imperative::NameVarBaseMap const&, paddle::framework::AttributeMap const&, paddle::platform::Place const&)
    26. 2   paddle::imperative::PreparedOp::Run(paddle::imperative::NameVarBaseMap const&, paddle::imperative::NameVarBaseMap const&, paddle::framework::AttributeMap const&)
    27. 3   std::_Function_handler<void (paddle::framework::ExecutionContext const&), paddle::framework::OpKernelRegistrarFunctor<paddle::platform::CPUPlace, false, 0ul, paddle::operators::RunProgramOpKernel<paddle::platform::CPUDeviceContext, float> >::operator()(char const*, char const*, int) const::{lambda(paddle::framework::ExecutionContext const&)#1}>::_M_invoke(std::_Any_data const&, paddle::framework::ExecutionContext const&)
    28. 4   paddle::operators::RunProgramOpKernel<paddle::platform::CPUDeviceContext, float>::Compute(paddle::framework::ExecutionContext const&) const
    29. 5   paddle::framework::Executor::RunPartialPreparedContext(paddle::framework::ExecutorPrepareContext*, paddle::framework::Scope*, long, long, bool, bool, bool)
    30. 6   paddle::framework::OperatorBase::Run(paddle::framework::Scope const&, paddle::platform::Place const&)
    31. 7   paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, paddle::platform::Place const&) const
    32. 8   paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, paddle::platform::Place const&, paddle::framework::RuntimeContext*) const
    33. 9   paddle::operators::ReshapeKernel::operator()(paddle::framework::ExecutionContext const&) const
    34. 10  paddle::operators::ReshapeOp::ValidateShape(std::vector<int, std::allocator<int> >, paddle::framework::DDim const&)
    35. 11  paddle::platform::EnforceNotMet::EnforceNotMet(std::string const&, char const*, int)
    36. 12  paddle::platform::GetCurrentTraceBackString()
    38. ----------------------
    39. Error Message Summary:
    40. ----------------------
    41. InvalidArgumentError: The 'shape' in ReshapeOp is invalid. The input tensor X'size must be equal to the capacity of 'shape'. But received X's shape = [3], X's size = 3, 'shape' is [1, 2], the capacity of 'shape' is 2.
    42.   [Hint: Expected capacity == in_size, but received capacity:2 != in_size:3.] (at /paddle/paddle/fluid/operators/reshape_op.cc:206)
    43.   [operator < reshape2 > error]  [operator < run_program > error]

    上述异常中，除了隐藏部分报错栈、报错定位到转换前的动态图代码外，报错信息中包含了C++报错栈 C++ Traceback 和 ，这是 Paddle 的 C++ 端异常信息，经处理后在 Python 的异常信息中显示。

    注解:

    如果您想查看被隐藏的信息栈，可以设置环境变量 TRANSLATOR_SIMPLIFY_NEW_ERROR=0。该环境变量默认值为1，表示隐藏冗余的报错信息栈。