3

I must be going crazy here... Here is my XML file, called Original.xml:

<root>
  <metadata>Trying to change this</metadata>
  <body>
    <salad>Greek Caesar</salad>
  </body>
</root>

I am trying to modify the contents within the metadata tag.

Here is my entire piece of code that I have that WORKS:

#include <iostream>

#include <rapidxml/rapidxml_print.hpp>
#include <rapidxml/rapidxml_utils.hpp>

int main()
{
  // Open 'Original.xml' to read from
  rapidxml::file<> xmlFile("Original.xml");
  rapidxml::xml_document<> doc;
  doc.parse<0>(xmlFile.data());

  // Get to <metadata> tag
  //                                       <root>        <metadata>    ???
  rapidxml::xml_node<>* metadataNode = doc.first_node()->first_node()->first_node();
  
  // Always correctly prints: 'Trying to change this'
  std::cout << "Before: " << metadataNode->value() << std::endl;

  // Modify the contents within <metadata>
  const std::string newMetadataValue = "Did the changing";
  metadataNode->value(newMetadataValue.c_str());

  // Always correctly prints: 'Did the changing'
  std::cout << "After: " << metadataNode->value() << std::endl;

  // Save output to 'New.xml'
  std::ofstream newXmlFile("New.xml");
  newXmlFile << doc;
  newXmlFile.close();
  doc.clear();

  return 0;
}

New.xml will now look like this:

<root>
    <metadata>Did the changing</metadata>
    <body>
        <salad>Greek Caesar</salad>
    </body>
</root>

That's the desired behavior I want.

What I don't understand is why I need a third first_node() call to SAVE the information inside metadata.

If I remove the third first_node() call, which is marked by the ??? above, New.xml will keep the old <metadata> string: "Trying to change this".

Yet, in this scenario, both std::cout calls on metadataNode->value() will still correctly print the intended strings; meaning, the first one will print "Trying to change this" and the second will correctly print "Did the changing".

Why in the world do I need to use n+1 calls to first_node() to SAVE the new value at the desired node where n is the number of nodes traversed (from the root) to get to the desired node? Why is that if I have n first_node() calls, I can successfully modify the value at the desired node in RAM only?

Possible bug? On whose end?

FriskySaga
  • 409
  • 1
  • 8
  • 19

1 Answers1

3

In the XML tree model, text elements are nodes as well. This makes sense when you have mixed content elements: <a>some<b/>text<c/>nodes</a>.

Basically:

#include <rapidxml/rapidxml_print.hpp>
#include <rapidxml/rapidxml_utils.hpp>

int main() {
    rapidxml::file<> xmlFile("Original.xml");
    rapidxml::xml_document<> doc;
    doc.parse<0>(xmlFile.data());

    auto root      = doc.first_node();
    auto metadata  = root->first_node();
    auto text_node = metadata->first_node();

    text_node->value("Did the changing");

    std::ofstream newXmlFile("New.xml");
    newXmlFile << doc;
}

But wait, there's more

Sadly, this is a problem unless your input has exactly the expected properties.

Assuming this ok sample:

char sample1[] = R"(<root><metadata>Trying to change this</metadata></root>)";

If metadata element were empty, you'd crash:

char sample2[] = R"(<root><metadata></metadata></root>)";
char sample3[] = R"(<root><metadata/></root>)";

Indeed this triggers ASAN failures:

/home/sehe/Projects/stackoverflow/test.cpp:17:25: runtime error: member access within null pointer of type 'struct xml_node'
/home/sehe/Projects/stackoverflow/test.cpp:17:25: runtime error: member call on null pointer of type 'struct xml_base'
/usr/include/rapidxml/rapidxml.hpp:762:24: runtime error: member call on null pointer of type 'struct xml_base'
/usr/include/rapidxml/rapidxml.hpp:753:21: runtime error: member access within null pointer of type 'struct xml_base'
AddressSanitizer:DEADLYSIGNAL

If there's a surprise, it will .... do surprising things!

char sample4[] = R"(<root><metadata><surprise/></metadata></root>)";

Ends up erroneously generating:

<root>
        <metadata>
                <surprise>changed</surprise>
        </metadata>
</root>

And that's not the end of it:

#include <rapidxml/rapidxml_print.hpp>
#include <rapidxml/rapidxml_utils.hpp>
#include <iostream>

namespace {
    char sample1[] = R"(<root><metadata>Trying to change this</metadata></root>)";
    char sample2[] = R"(<root><metadata><surprise/></metadata></root>)";
    char sample3[] = R"(<root><metadata>mixed<surprise/>bag</metadata></root>)";
    char sample4[] = R"(<root><metadata><![CDATA[mixed<surprise/>bag]]></metadata></root>)";
    char sample5[] = R"(<root><metadata><!-- comment please -->outloud<!-- hidden --></metadata></root>)";
    //These crash:
    //char sampleX[] = R"(<root><metadata></metadata></root>)";
    //char sampleY[] = R"(<root><metadata/></root>)";
}

int main() {
    for (char* xml : {sample1, sample2, sample3, sample4, sample5}) {
        std::cout << "\n=== " << xml << " ===\n";
        rapidxml::xml_document<> doc;
        doc.parse<0>(xml);

        auto root      = doc.first_node();
        auto metadata  = root->first_node();
        auto text_node = metadata->first_node();

        text_node->value("changed");
        print(std::cout << " --> ", doc, rapidxml::print_no_indenting);
        std::cout << "\n";
    }
}

Prints

=== <root><metadata>Trying to change this</metadata></root> ===
 --> <root><metadata>changed</metadata></root>

=== <root><metadata><surprise/></metadata></root> ===
 --> <root><metadata><surprise>changed</surprise></metadata></root>

=== <root><metadata>mixed<surprise/>bag</metadata></root> ===
 --> <root><metadata>changed<surprise/>bag</metadata></root>

=== <root><metadata><![CDATA[mixed<surprise/>bag]]></metadata></root> ===
 --> <root><metadata><![CDATA[changed]]></metadata></root>

=== <root><metadata><!-- comment please -->outloud<!-- hidden --></metadata></root> ===
 --> <root><metadata>changed</metadata></root>

HOW TO GET IT ROBUST?

  • Firstly, use queries to find your target. Sadly rapidxml doesn't support this; See What XML parser should I use in C++?

  • Secondly, check the node type before editing

  • Thirdly, replace the entire node if you can, that makes you independent of what was previously there

  • Lastly, be sure to actually allocate your new node from the document so you don't get lifetime issues.

     auto root = doc.first_node();
    
     if (auto* old_meta = root->first_node()) {
         assert(old_meta->name() == std::string("metadata"));
         print(std::cout << "Removing metadata node: ", *old_meta, fmt);
         std::cout << "\n";
         root->remove_first_node();
     }
    
     auto newmeta = doc.allocate_node(rapidxml::node_element, "metadata", "changed");
     root->prepend_node(newmeta);
    

PUTTING IT ALL TOGETHER:

#include <rapidxml/rapidxml.hpp>
#include <rapidxml/rapidxml_print.hpp>
#include <rapidxml/rapidxml_utils.hpp>
#include <iostream>

namespace {
    std::string cases[] = {
     R"(<root><metadata>Trying to change this</metadata></root>)",
     R"(<root><metadata><surprise/></metadata></root>)",
     R"(<root><metadata>mixed<surprise/>bag</metadata></root>)",
     R"(<root><metadata><![CDATA[mixed<surprise/>bag]]></metadata></root>)",
     R"(<root><metadata><!-- comment please -->outloud<!-- hidden --></metadata></root>)",
     R"(<root>
  <metadata>Trying to change this</metadata>
  <body>
    <salad>Greek Caesar</salad>
   </body>
</root>)",
     //These no longer crash:
     R"(<root><metadata></metadata></root>)",
     R"(<root><metadata/></root>)",
     // more edge-cases in the predecessor chain
     R"(<root></root>)",
     R"(<root><no-metadata/></root>)",
     R"(<bogus/>)",
    };
}

int main() {
    auto const fmt = rapidxml::print_no_indenting;
    for (auto& xml : cases) {
        std::cout << "Input: " << xml << "\n";

        rapidxml::xml_document<> doc;
        doc.parse<0>(xml.data());

        if (auto root = doc.first_node()) {
            if (root->name() == std::string("root")) {
                if (auto* old_meta = root->first_node()) {
                    if (old_meta->name() == std::string("metadata")) {
                        root->remove_first_node();
                    } else {
                        std::cout << "WARNING: Not removing '" << old_meta->name() << "' element where 'metadata' expected\n";
                    }
                }

                auto newmeta = doc.allocate_node(rapidxml::node_element, "metadata", "changed");
                root->prepend_node(newmeta);
            } else {
                std::cout << "WARNING: '" << root->name() << "' found where 'root' expected\n";
            }
        }
        
        print(std::cout << "Output: ", doc, fmt);
        std::cout << "\n--\n";
    }
}

Prints

Input: <root><metadata>Trying to change this</metadata></root> ===
Output: <root><metadata>changed</metadata></root>
--
Input: <root><metadata><surprise/></metadata></root> ===
Output: <root><metadata>changed</metadata></root>
--
Input: <root><metadata>mixed<surprise/>bag</metadata></root> ===
Output: <root><metadata>changed</metadata></root>
--
Input: <root><metadata><![CDATA[mixed<surprise/>bag]]></metadata></root> ===
Output: <root><metadata>changed</metadata></root>
--
Input: <root><metadata><!-- comment please -->outloud<!-- hidden --></metadata></root> ===
Output: <root><metadata>changed</metadata></root>
--
Input: <root>
  <metadata>Trying to change this</metadata>
  <body>
    <salad>Greek Caesar</salad>
   </body>
</root> ===
Output: <root><metadata>changed</metadata><body><salad>Greek Caesar</salad></body></root>
--
Input: <root><metadata></metadata></root> ===
Output: <root><metadata>changed</metadata></root>
--
Input: <root><metadata/></root> ===
Output: <root><metadata>changed</metadata></root>
--
Input: <root></root> ===
Output: <root><metadata>changed</metadata></root>
--
Input: <root><no-metadata/></root> ===
WARNING: Not removing 'no-metadata' element where 'metadata' expected
Output: <root><metadata>changed</metadata><no-metadata/></root>
--
Input: <bogus/> ===
WARNING: 'bogus' found where 'root' expected
Output: <bogus/>
--

SUMMARY

XML is extensible. It's Markup. It's Language. It's not simple :)

sehe
  • 374,641
  • 47
  • 450
  • 633
  • 1
    For fun: [here's how PugiXML would do](https://paste.ubuntu.com/p/dQFkQ7jtZD/) the matching with a direct `doc.select_single_node("//root/metadata")` - or even `doc.select_single_node("//root/metadata/test()")`if you wanted to get directly at a required text node. It has the same effects, safely. Note also the lack of allocation acrobatics. Output: https://paste.ubuntu.com/p/jwtxXzppQx/ – sehe Jul 08 '20 at 01:34
  • 1
    An absolutely thorough answer. Still working to digest this information but I have a newfound appreciation for the beautiful ways you've shown me to approach testing. – FriskySaga Jul 09 '20 at 00:09
  • "use queries to find your target." While RapidXML doesn't support full queries, you absolutely should use the name parameter to `first_node` wherever possible. `auto metadata = root->first_node("metadata");` – Roddy Jul 10 '20 at 08:15
  • @Roddy note that it does change the meaning of the code though. Queries doesn't mind if other nodes (with unexpected names) are there. In general I think that's how XML was intended to be used (the X is for extendible). – sehe Jul 10 '20 at 11:03
  • @sehe, Yeah. Don't get me started on how XML was intended to be used, though :) Everywhere i see it, it's banging an JSON-shaped peg into an XML hole. I've never, EVER seen it used as genuine markup. – Roddy Jul 10 '20 at 15:47
  • @Roddy +100 [It's still markup if it describes data]. And yeah I _have_ seen it as intended to markup semantic data. And it was horrible. Ever heard of XBRL? Combined with XSLT-FO it is the definition of hell... interesting. – sehe Jul 10 '20 at 15:50