Boost Property Tree: Remove attribute from a node

Question

I have the following XML file:

<?xml version="1.0" encoding="utf-8"?>
<gexf>
  <graph>
    <nodes>
      <node id="0" label="0" start="0" end="25"/>
      <node id="1" label="1" start="1"/>
      <node id="2" label="2" start="2"/>
      ...
    </nodes>
    <edges>
      <edge id="0" source="0" target="1" start="7" end="19"/>
      <edge id="1" source="0" target="2" start="8" end="20"/>
      ...
    </edges>
  </graph>
</gexf>

I want to remove the start and end attributes from the edge with source="0" and target="1".

The way I've tried to do this is in the following code. Assuming the XML file is named ptree_test.gexf I read it in, find the correct edge in the tree, and then attempt to use erase to get rid of the attributes.

#include <boost/property_tree/ptree.hpp>
#include <boost/property_tree/xml_parser.hpp>
#include <iostream>

using boost::property_tree::ptree;

int main(int argc, char *argv[]) {

  ptree pt;

  read_xml("ptree_test.gexf", pt);

  // Now find edge (0, 1) and delete the start and end attributes
  ptree edge;
  int id1, id2;
  id1 = 0;
  id2 = 1;

  for(auto &e : pt.get_child("gexf.graph.edges")) {
    int s, t;
    s = e.second.get<int>("<xmlattr>.source");
    t = e.second.get<int>("<xmlattr>.target");

    // Check if this is the correct edge
    // Two checks because it can be reversed
    if((id1 == s && id2 == t) || (id1 == t && id2 == s)) {
      edge = e.second;
      break;
    }
  }

  for(auto & attr : edge.get_child("<xmlattr>")) {
    if(attr.first == "end" || attr.first == "start") {
      edge.erase(attr.first);
    }
  }

  write_xml(std::cout, pt);
  return 0;
}

This does not work. It doesn't remove the attribute. In fact, if I put in a debug statement that prints the return of edge.erase(attr.first) it shows 0.

Boost.PropertyTree is ***not a quick-and-dirty XML parser***. Stop treating it like it is. The XML serializer is for storing and unstoring data. It's not meant for loading or modifying XML in an arbitrary way. In short, if you need to manipulate a GEFX XML file, [use a *real* XML parser](http://stackoverflow.com/questions/9387610/what-xml-parser-should-i-use-in-c). — Nicol Bolas, Feb 08 '13 at 20:29
@NicolBolas I'm actually not ever reading the file. I just wrote the example like that because it made it shorter and easier to understand. In my library I'm composing the XML file in memory and then writing it out when done. During that process there may be cases when I need to change it in the way specified. I'm using the boost property tree because boost is already a dependency for my library. Rather than introduce another dependency I figured I'd just stick with it for the XML stuff. If this is not possible I'll use something else. Are you saying this is impossible? — Sean Lynch, Feb 08 '13 at 21:41

score 0 · Answer 1 · edited May 23 '17 at 11:59

0

Before answering, I would like to again discourage you from using Boost.PropertyTree as a quick-and-dirty XML processing system. Please use a real XML parser; there are many to choose from, and some are quite effective and require little dependency maintenance.

Anyway, your problem comes from your use of erase. You are trying to erase an element from a list that you're iterating over. That's not going to work. Not without special coding for your loop.

So you can't use a range-based for-loop here. You have to use a real for-loop over the iterators.

auto &children = edge.get_child();
for(auto &attrIt = children.begin(); attrIt != children.end();)
{
  auto &attr = *attrIt;
  //Do stuff here.


  if(attr.first == "end" || attr.first == "start")
    attrIt = children.erase(attrIt);
  else
    ++attrIt;
}

edited May 23 '17 at 11:59

Community

1
1

answered Feb 08 '13 at 22:09

Nicol Bolas

449,505
63
781
982

I think after thinking about this more and looking at your answer, I'm going to [use a real XML parser](http://stackoverflow.com/questions/9387610/what-xml-parser-should-i-use-in-c). But looking at the documentation for boost property tree, two of the signatures take iterators, but one takes a key: `size_type erase(const key_type &)` so I'm not sure why this doesn't work - not to mention that my example wouldn't compile if it expected an iterator. – Sean Lynch Feb 08 '13 at 23:13
Right, some containers (typically associative ones) do support direct erasure of keys. – StilesCrisis Feb 08 '13 at 23:35
@SeanLynch: You're right; your problem is deleting from a list you're iterating over. – Nicol Bolas Feb 08 '13 at 23:58
I am (honestly) curious why you think the use of `Boost.PropertyTree` as XML reader should be discouraged. Is it a matter of efficiency or incompletness of `BPT` with respect to a general XML file? – alfC Jan 02 '16 at 08:19
@alfC: I could just as easily turn the question around on you. Boost.PropertyTree is a data type that's very clearly *not* intended for XML data. So why should using it as an XML reader *not* be discouraged? Because it's quick and easy? So is downloading and installing Rapid/PugiXML. – Nicol Bolas Jan 02 '16 at 13:27
I don't know all the details of XML and all the details of BPT. But if BPT can be mapped to and XML and vice versa, do you see any other problem? Isn't an XML a property tree (in the BPT definition)? Also, BPT uses RapidXML internally so from a technical point of view BPT can in principle read (write?) the same XMLs with the advantages of C++/STL container syntax. http://www.boost.org/doc/libs/1_60_0/doc/html/property_tree/parsers.html#property_tree.parsers.xml_parser – alfC Jan 02 '16 at 17:52
Or your concern of BPT xml comes from the fact that schemas cannot be specified? – alfC Jan 02 '16 at 18:02
1

@alfC: You linked to a page that explains exactly why you should not use BPT to read arbitrary XML. It says very explicitly, "The XML storage encoding does not round-trip perfectly." BPT's XML parsing is intended for reading XML *written by BPT*. As for "the advantages of C++/STL container syntax", PugiXML provides iterator access to elements and attributes. – Nicol Bolas Jan 02 '16 at 18:14
Ok, it makes sense now. – alfC Jan 02 '16 at 18:18
BTW, I put another answer based on yours. Also, maybe it is the "encoding" issue that you mentioned, I had to add load and save options just to make the XML file legible. – alfC Jan 02 '16 at 18:59

alfC · Answer 2 · 2016-01-02T19:11:23.720

The main problem is that you are making a copy of the subtree in this line:

  edge = e.second;

and then modifying that copy instead of the original. Later, as @NicolBolas said you need an interator to erase. The full code looks like this:

int main(){
        boost::property_tree::ptree pt;
        read_xml("ptree_test.gexf", pt, boost::property_tree::xml_parser::trim_whitespace);
        int id1, id2;
        id1 = 0;
        id2 = 1;
        for(auto &e : pt.get_child("gexf.graph.edges")) {
            int s, t;
            s = e.second.get<int>("<xmlattr>.source");
            t = e.second.get<int>("<xmlattr>.target");
            // Check if this is the correct edge
            // Two checks because it can be reversed
            if((id1 == s && id2 == t) || (id1 == t && id2 == s)){
                auto &children = e.second.get_child("<xmlattr>");
                for(auto attrIt = children.begin(); attrIt != children.end(); ++attrIt){
                  if(attrIt->first == "end" || attrIt->first == "start")
                    attrIt = children.erase(attrIt);
                }
                break; // maybe shouldn't be here to keep looking?
            }
        }
        write_xml("ptree_test_copy.gexf", pt, std::locale(), bpt::xml_writer_settings<std::string>{'\t', 1});
}

Boost Property Tree: Remove attribute from a node

2 Answers2

Linked