Python默认错误信息栈：执行Python程序默认记录的执行路径，对定位报错位置很有帮助。这是Python本身特性，此处不展开介绍。
C++错误信息栈：程序在Paddle C++ core中的错误路径，即为模块paddle.fluid.core中的程序执行路径，这部分信息对开发者帮助有限。但当开发者通过Issue向飞桨开发人员提问时，提供C++报错栈的信息将有助于开发人员快速定位问题。（目前C++错误信息栈仅支持Unix平台，Windows平台暂不支持）
Paddle Python错误信息栈：为什么这里还有一个Paddle Python错误信息栈呢？因为在声明式编程模式（静态图）下，模型编译和执行是分离的。执行时报错的路径由Python默认程序栈记录，但这并不能告知用户具体出错的程序位置，因此对于算子类型的API，飞桨额外记录了编译时的执行路径，帮助开发者定位具体代码出错的位置，该部分信息对于调试具有较大意义。
核心错误概要：信息包含错误类型、错误特征、概要提示、出错文件名与行号、出错算子名等，这些信息不仅有助于开发者理解错误，也有助于迅速定位错误。

3步快速定位问题

当使用飞桨遇到报错提示时，定位流程是啥样子的呢？请对应上文提到的飞桨报错信息结构图，按如下流程逐步分析。

报错信息分析流程

下面结合示例，向大家讲解飞桨的报错信息的分析过程（示例使用飞桨2020年7月1日的develop版本）。飞桨支持两种编程模式，声明式编程模式（静态图）和命令式编程模式（动态图），我们将逐一介绍。

执行如下静态图示例代码：

import paddle.fluid as fluid
import numpy

# 1. 网络结构定义
x = fluid.layers.data(name='X', shape=[-1, 13], dtype='float32')
y = fluid.layers.data(name='Y', shape=[-1, 1], dtype='float32')
predict = fluid.layers.fc(input=x, size=1, act=None)
loss = fluid.layers.square_error_cost(input=predict, label=y)
avg_loss = fluid.layers.mean(loss)

# 2. 优化器配置
fluid.optimizer.SGD(learning_rate=0.01).minimize(avg_loss)

# 3. 执行环境准备
place = fluid.CPUPlace()
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())

# 4. 执行网络
x = numpy.random.random(size=(8, 12)).astype('float32')
y = numpy.random.random(size=(8, 1)).astype('float32')
loss_data, = exe.run(fluid.default_main_program(), feed={'X': x, 'Y': y}, fetch_list=[avg_loss.name])

代码执行后的报错信息如下：

Traceback (most recent call last):
  File "paddle_error_case1.py", line 24, in <module>
    loss_data, = exe.run(fluid.default_main_program(), feed={'X': x, 'Y': y}, fetch_list=[avg_loss.name])
  File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/executor.py", line 1079, in run
    six.reraise(*sys.exc_info())
  File "/usr/local/lib/python3.5/dist-packages/six.py", line 696, in reraise
    raise value
  File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/executor.py", line 1074, in run
    return_merged=return_merged)
  File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/executor.py", line 1162, in _run_impl
    use_program_cache=use_program_cache)
  File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/executor.py", line 1237, in _run_program
    fetch_var_name)
paddle.fluid.core_avx.EnforceNotMet: 

--------------------------------------------
C++ Call Stacks (More useful to developers):
--------------------------------------------
0   std::string paddle::platform::GetTraceBackString<std::string const&>(std::string const&, char const*, int)
1   paddle::platform::EnforceNotMet::EnforceNotMet(std::string const&, char const*, int)
2   paddle::operators::MulOp::InferShape(paddle::framework::InferShapeContext*) const
3   paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, paddle::platform::Place const&, paddle::framework::RuntimeContext*) const
4   paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, paddle::platform::Place const&) const
5   paddle::framework::OperatorBase::Run(paddle::framework::Scope const&, paddle::platform::Place const&)
6   paddle::framework::Executor::RunPartialPreparedContext(paddle::framework::ExecutorPrepareContext*, paddle::framework::Scope*, long, long, bool, bool, bool)
7   paddle::framework::Executor::RunPreparedContext(paddle::framework::ExecutorPrepareContext*, paddle::framework::Scope*, bool, bool, bool)
8   paddle::framework::Executor::Run(paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool, std::vector<std::string, std::allocator<std::string > > const&, bool, bool)

------------------------------------------
Python Call Stacks (More useful to users):
------------------------------------------
  File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/framework.py", line 2799, in append_op
    attrs=kwargs.get("attrs", None))
  File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/layer_helper.py", line 43, in append_op
    return self.main_program.current_block().append_op(*args, **kwargs)
  File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/layers/nn.py", line 349, in fc
    "y_num_col_dims": 1})
  File "paddle_error_case1.py", line 9, in <module>
    predict = fluid.layers.fc(input=x, size=1, act=None)

----------------------
Error Message Summary:
----------------------
InvalidArgumentError: After flatten the input tensor X and Y to 2-D dimensions matrix X1 and Y1, the matrix X1's width must be equal with matrix Y1's height. But received X's shape = [8, 12], X1's shape = [8, 12], X1's width = 12; Y's shape = [13, 1], Y1's shape = [13, 1], Y1's height = 13.
  [Hint: Expected x_mat_dims[1] == y_mat_dims[0], but received x_mat_dims[1]:12 != y_mat_dims[0]:13.] at (/work/paddle/paddle/fluid/operators/mul_op.cc:83)
  [operator < mul > error]

参考飞桨报错信息分析流程对这个错误示例进行剖析。

1. 首先分析代码核心错误概要。依据统一的报错结构，开发者可以快速的找到报错原因。

从示例中可获得如下信息：

这是一个参数错误；出错的Op是mul；mul Op输入的Tensor X矩阵的宽度，即第2维的大小需要和输入Tensor Y矩阵的高度，即第一维的大小相等，才可以进行正常的矩阵乘法；给出了具体的输入X与Y的维度信息即出错维度的值，有一处的维度写错了，可能是13误写成了12。

目前飞桨有12种错误类型，更多介绍请查看《报错信息文案书写规范》，链接如下：https://github.com/[PaddlePaddle](https://www.oschina.net/action/visit/ad?id=1185)/[Paddle](https://www.oschina.net/action/visit/ad?id=1185)/wiki/[Paddle](https://www.oschina.net/action/visit/ad?id=1185)-Error-Message-Writing-Specification

2. 其次分析Paddle 编译时Python错误信息栈，发现出错的代码位置如下：

Paddle插入的Python错误信息栈为了和C++栈的调用顺序保持一致，最下面的信息是用户代码的位置，这和原生python错误信息栈的顺序有所区别。这里我们可以得知，是调用fc的时候出错的，fc中包含一个乘法运算和一个加法运算，根据前面的信息可以得知是此处的乘法运算的输入数据存在问题。至此，通过检查代码，可以找到错误位置：

将代码中的12改为13，即可解决该问题。

（可选）通常出错场景较为简单时，C++错误信息栈可以不关心。但如果用户在解决时遇到困难，需要飞桨开发人员协助解决时，需要反馈此信息，帮助开发人员快速得知底层的出错执行逻辑。例如在这个例子中，我们能够得知程序的执行路径为Run -> RunPreParedContext -> Run -> RunImpl -> MulOp::InferShape，InferShape是检查算子输入输出及参数维度的方法，由此可以推断出，本错误是由于Mul算子的输入参数维度出错导致。

飞桨命令式编程模式

（动态图）报错解读

动态图不区分网络模型的编译期和执行期，报错信息中不需要再插入编译时的python信息栈。执行如下动态图示例代码：

import numpy
import paddle.fluid as fluid

place = fluid.CPUPlace()
with fluid.dygraph.guard(place):
    x = numpy.random.random(size=(10, 2)).astype('float32')
    linear = fluid.dygraph.Linear(1, 10)
    data = fluid.dygraph.to_variable(x)
    res = linear(data)

代码执行后的报错信息如下：

/work/scripts {master} python paddle_error_case2.py 
Traceback (most recent call last):
  File "paddle_error_case2.py", line 9, in <module>
    res = linear(data)
  File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/dygraph/layers.py", line 600, in __call__
    outputs = self.forward(*inputs, **kwargs)
  File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/dygraph/nn.py", line 965, in forward
    'transpose_Y', False, "alpha", 1)
paddle.fluid.core_avx.EnforceNotMet: 

--------------------------------------------
C++ Call Stacks (More useful to developers):
--------------------------------------------
0   std::string paddle::platform::GetTraceBackString<std::string const&>(std::string const&, char const*, int)
1   paddle::platform::EnforceNotMet::EnforceNotMet(std::string const&, char const*, int)
2   paddle::operators::MatMulOp::InferShape(paddle::framework::InferShapeContext*) const
3   paddle::imperative::PreparedOp::Run(paddle::imperative::NameVarBaseMap const&, paddle::imperative::NameVarBaseMap const&, paddle::framework::AttributeMap const&)
4   paddle::imperative::Tracer::TraceOp(std::string const&, paddle::imperative::NameVarBaseMap const&, paddle::imperative::NameVarBaseMap const&, paddle::framework::AttributeMap, paddle::platform::Place const&, bool)
5   paddle::imperative::Tracer::TraceOp(std::string const&, paddle::imperative::NameVarBaseMap const&, paddle::imperative::NameVarBaseMap const&, paddle::framework::AttributeMap)

----------------------
Error Message Summary:
----------------------
InvalidArgumentError: Input X's width should be equal to the Y's height, but received X's shape: [10, 2],Y's shape: [1, 10].
  [Hint: Expected mat_dim_x.width_ == mat_dim_y.height_, but received mat_dim_x.width_:2 != mat_dim_y.height_:1.] at (/work/paddle/paddle/fluid/operators/matmul_op.cc:411)
  [operator < matmul > error]

同样，我们可以依据前面讲述的步骤对报错进行分析。