如何使用带有 eigen 数学库的 gdb 进行调试

Posted

技术标签:

【中文标题】如何使用带有 eigen 数学库的 gdb 进行调试【英文标题】:How to debug with gdb with eigen math library 【发布时间】:2014-11-28 01:00:44 【问题描述】:

我将 code::blocks (CB) 或 Visual Studio (VS) 用于带有 eigen 库的 C++ 程序。但是,在调试时,我看不到数组、矩阵等的内容。我查看了以下帖子:

Using GDB with Eigen C++ library

我不是 C++ 专家,但我知道我需要一种叫做打印机的东西。

https://android.googlesource.com/platform/external/eigen/+/b015e75e8c7ba1ab4ddb91e9372a57e76f3fd159/debug/gdb/printers.py 有源代码。但是我不知道如何使用这个源代码在 CB 或 VS 中使用带有 eigen 库的 gdb 进行调试。任何想法如何做到这一点?

更新: vsoftco 提到了一个网页 https://android.googlesource.com/platform/external/eigen/+/b015e75e8c7ba1ab4ddb91e9372a57e76f3fd159/debug,它有用于 CB 和 VS 的 gdb 的 python 打印机。如果有人知道如何使用它们来查看特征库数组的内容,请发表评论。

【问题讨论】:

您是否已验证您正在使用调试信息构建 Eigen?对于 g++,这意味着添加标志 -g -ggdb。 VS 见***.com/a/4662345/2197564 【参考方案1】:

Eigen::Matrix 类不是聚合类,因此您不能仅使用调试器查看其内容。但是,您应该能够使用调试器介入,并且可以使用cout 或其他方法来显示内容。

你提到的链接是gdb的python插件,允许gdb打印Eigen类型的内容。但是当你使用 VS(它有它的内部调试器并且不使用gdb)时,它没有理由在你的情况下工作。

您可以尝试切换到 MinGW 和 g++/gdb,或者可以查看此链接 How can I use GDB from inside Visual Studio C++ (Express) to debug my GCC Makefile projects? 以获取有关在 VS 下安装 gdb 的一些建议。

PS:看来VS的解决方案也有,

https://android.googlesource.com/platform/external/eigen/+/b015e75e8c7ba1ab4ddb91e9372a57e76f3fd159/debug

【讨论】:

如果我将 GNU gccCode::Blockseigen 一起使用,如何调试以查看数组的内容? @c202933 您还需要使用gdb 作为调试器,并将该python 补丁应用到gdb。如何做后者我真的不知道,但应该在你链接的那个页面上解释。但是checkout同一个页面,好像还有VS的补丁,android.googlesource.com/platform/external/eigen/+/…【参考方案2】:

gdb 的 python 打印机对我有用。请注意,printers.py 脚本是用 Python 2.7 编写的,而您的 gdb 可能正在运行 python 3.5 或更高版本...使用 2to3 转换器或简单地将其复制到名为 printers3.py 的新文件中:

# -*- coding: utf-8 -*-
# This file is part of Eigen, a lightweight C++ template library
# for linear algebra.
#
# Copyright (C) 2009 Benjamin Schindler <bschindler@inf.ethz.ch>
#
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at http://mozilla.org/MPL/2.0/.
# Pretty printers for Eigen::Matrix
# This is still pretty basic as the python extension to gdb is still pretty basic. 
# It cannot handle complex eigen types and it doesn't support any of the other eigen types
# Such as quaternion or some other type. 
# This code supports fixed size as well as dynamic size matrices
# To use it:
#
# * Create a directory and put the file as well as an empty __init__.py in 
#   that directory.
# * Create a ~/.gdbinit file, that contains the following:
#      python
#      import sys
#      sys.path.insert(0, '/path/to/eigen/printer/directory')
#      from printers3 import register_eigen_printers
#      register_eigen_printers (None)
#      end
import gdb
import re
import itertools

class EigenMatrixPrinter:
    "Print Eigen Matrix or Array of some kind"
    def __init__(self, variety, val):
        "Extract all the necessary information"

        # Save the variety (presumably "Matrix" or "Array") for later usage
        self.variety = variety

        # The gdb extension does not support value template arguments - need to extract them by hand
        type = val.type
        if type.code == gdb.TYPE_CODE_REF:
            type = type.target()
        self.type = type.unqualified().strip_typedefs()
        tag = self.type.tag
        regex = re.compile('\<.*\>')
        m = regex.findall(tag)[0][1:-1]
        template_params = m.split(',')
        template_params = [x.replace(" ", "") for x in template_params]

        if template_params[1] == '-0x00000000000000001' or template_params[1] == '-0x000000001' or template_params[1] == '-1':
            self.rows = val['m_storage']['m_rows']
        else:
            self.rows = int(template_params[1])

        if template_params[2] == '-0x00000000000000001' or template_params[2] == '-0x000000001' or template_params[2] == '-1':
            self.cols = val['m_storage']['m_cols']
        else:
            self.cols = int(template_params[2])

        self.options = 0 # default value
        if len(template_params) > 3:
            self.options = template_params[3];

        self.rowMajor = (int(self.options) & 0x1)

        self.innerType = self.type.template_argument(0)

        self.val = val

        # Fixed size matrices have a struct as their storage, so we need to walk through this
        self.data = self.val['m_storage']['m_data']
        if self.data.type.code == gdb.TYPE_CODE_STRUCT:
            self.data = self.data['array']
            self.data = self.data.cast(self.innerType.pointer())

    class _iterator:
        def __init__ (self, rows, cols, dataPtr, rowMajor):
            self.rows = rows
            self.cols = cols
            self.dataPtr = dataPtr
            self.currentRow = 0
            self.currentCol = 0
            self.rowMajor = rowMajor

        def __iter__ (self):
            return self

        def __next__(self):

            row = self.currentRow
            col = self.currentCol
            if self.rowMajor == 0:
                if self.currentCol >= self.cols:
                    raise StopIteration

                self.currentRow = self.currentRow + 1
                if self.currentRow >= self.rows:
                    self.currentRow = 0
                    self.currentCol = self.currentCol + 1
            else:
                if self.currentRow >= self.rows:
                    raise StopIteration

                self.currentCol = self.currentCol + 1
                if self.currentCol >= self.cols:
                    self.currentCol = 0
                    self.currentRow = self.currentRow + 1


            item = self.dataPtr.dereference()
            self.dataPtr = self.dataPtr + 1
            if (self.cols == 1): #if it's a column vector
                return ('[%d]' % (row,), item)
            elif (self.rows == 1): #if it's a row vector
                return ('[%d]' % (col,), item)
            return ('[%d,%d]' % (row, col), item)

    def children(self):

        return self._iterator(self.rows, self.cols, self.data, self.rowMajor)

    def to_string(self):
        return "Eigen::%s<%s,%d,%d,%s> (data ptr: %s)" % (self.variety, self.innerType, self.rows, self.cols, "RowMajor" if self.rowMajor else  "ColMajor", self.data)
class EigenQuaternionPrinter:
    "Print an Eigen Quaternion"

    def __init__(self, val):
        "Extract all the necessary information"
        # The gdb extension does not support value template arguments - need to extract them by hand
        type = val.type
        if type.code == gdb.TYPE_CODE_REF:
            type = type.target()
        self.type = type.unqualified().strip_typedefs()
        self.innerType = self.type.template_argument(0)
        self.val = val

        # Quaternions have a struct as their storage, so we need to walk through this
        self.data = self.val['m_coeffs']['m_storage']['m_data']['array']
        self.data = self.data.cast(self.innerType.pointer())

    class _iterator:
        def __init__ (self, dataPtr):
            self.dataPtr = dataPtr
            self.currentElement = 0
            self.elementNames = ['x', 'y', 'z', 'w']

        def __iter__ (self):
            return self

        def __next__(self):
            element = self.currentElement

            if self.currentElement >= 4: #there are 4 elements in a quanternion
                raise StopIteration

            self.currentElement = self.currentElement + 1

            item = self.dataPtr.dereference()
            self.dataPtr = self.dataPtr + 1
            return ('[%s]' % (self.elementNames[element],), item)

    def children(self):

        return self._iterator(self.data)

    def to_string(self):
        return "Eigen::Quaternion<%s> (data ptr: %s)" % (self.innerType, self.data)
def build_eigen_dictionary ():
    pretty_printers_dict[re.compile('^Eigen::Quaternion<.*>$')] = lambda val: EigenQuaternionPrinter(val)
    pretty_printers_dict[re.compile('^Eigen::Matrix<.*>$')] = lambda val: EigenMatrixPrinter("Matrix", val)
    pretty_printers_dict[re.compile('^Eigen::Array<.*>$')]  = lambda val: EigenMatrixPrinter("Array",  val)
def register_eigen_printers(obj):
    "Register eigen pretty-printers with objfile Obj"
    if obj == None:
        obj = gdb
    obj.pretty_printers.append(lookup_function)
def lookup_function(val):
    "Look-up and return a pretty-printer that can print va."

    type = val.type

    if type.code == gdb.TYPE_CODE_REF:
        type = type.target()

    type = type.unqualified().strip_typedefs()

    typename = type.tag
    if typename == None:
        return None

    for function in pretty_printers_dict:
        if function.search(typename):
            return pretty_printers_dict[function](val)

    return None
pretty_printers_dict = 
build_eigen_dictionary ()

【讨论】:

以上是关于如何使用带有 eigen 数学库的 gdb 进行调试的主要内容,如果未能解决你的问题,请参考以下文章

为啥我的Eigen的速度特别慢

Eigen库的使用笔记

slam学习之Eigen库的简单总结

线性代数c++模板库 eigen如何使用

使用 gdb 检查调用者帧

带有 Eigen 和 IPOPT 的矢量化标志