Implementing IXmlWriter Part 13: Putting IXmlWriter Behind A Pimpl Firewall

This is part 13 of my Implementing IXmlWriter post series.

As the private members of IXmlWriter are getting too numerous and too likely to change by my judgment, today I will put last time’s IXmlWriter behind a compilation firewall (pimpl).

The idea behind the pimpl idiom is to hide as much of the class definition as possible in order to avoid requiring users of the class to recompile if the class’s private members are changed. It is accomplished by moving all private members (functions, data, etc.) into a separate class (called the implementation or pimpl class) hidden from the class definition, and replacing these members with an opaque pointer to a forward declaration of this class. It works because a C++ compiler does not need to have the full definition of a class visible in order to allocate space for a pointer to the class; every pointer is a constant, fixed size (often 4 bytes).

Guru of the Week (GotW) #24: Compilation Firewalls contains some discussion about what should belong within the pimpl class, which led me to choose to move all private type definitions, functions, and data into the pimpl class. GotW #24 also mentions some negative performance impact with respect to instantiation and the extra indirections required for accessing private members. The performance impact associated with instantiation can be mitigated by following the lessons of GotW #28: The Fast Pimpl Idiom, but I believe that instances of IXmlWriter will be created relatively infrequently and thus the extra effort required is not worth it. The impact based on the extra indirections we will just have to live with; I highly doubt a profile would identify the extra indirections as a performance bottleneck for IXmlWriter.

The steps required to modify IXmlWriter to use the pimpl idiom are:

  1. In the implementation file, create an internal class (a struct StringXmlWriter::StringXmlWriterImpl, in my case) and move all private members, including the implementations of the private functions, into this class. If applicable, write constructors for the pimpl class which perform the same initializations of private members that the constructors on your original class did.
  2. Within the class definition in the header file, add a forward declaration to this implementation class (e.g. struct StringXmlWriterImpl;) and a member which contains a pointer to this class (e.g. StringXmlWriterImpl* m_pimpl;). This should be the only private member left in the class. (Why not use std::auto_ptr<StringXmlWriterImpl>? See this code comments forum thread.)
  3. Now that you have a pointer variable as a member of the class, the default copy constructor and assignment operators are wrong. Either disable them (as I have done) or write correct implementations.
  4. Modify every constructor of your original class to create an instance of the pimpl class using new and the appropriate pimpl constructor.
  5. Add a destructor to your original class which deletes the instance of the pimpl class.
  6. In the implementation file, modify all references to private types and data to indirect through the pimpl instance. For example, m_xmlStr = ...; becomes m_pimpl->m_xmlStr = ...;.

Here’s the resulting header file:

// StringXmlWriter.h

class StringXmlWriter
{
public:
    enum Formatting
    {
        Formatting_Indented,
        Formatting_None
    };

    StringXmlWriter(Formatting formatting = Formatting_None,
                    int indentation = 2,
                    char indentChar = ' ');
    ~StringXmlWriter();

    std::string GetXmlString() const;
    void WriteAttributeString(const std::string& localName,
                              const std::string& text);
    void WriteAttributeString(const std::string& localName,
                              const std::string& ns,
                              const std::string& text);
    void WriteComment(const std::string& text);
    void WriteElementString(const std::string& localName,
                            const std::string& text);
    void WriteElementString(const std::string& localName,
                            const std::string& ns,
                            const std::string& text);
    void WriteEndAttribute();
    void WriteEndDocument();
    void WriteEndElement();
    void WriteStartAttribute(const std::string& localName);
    void WriteStartAttribute(const std::string& localName,
                             const std::string& ns);
    void WriteStartDocument();
    void WriteStartElement(const std::string& localName);
    void WriteStartElement(const std::string& localName,
                           const std::string& ns);
    void WriteString(const std::string& text);

private:
    struct StringXmlWriterImpl;
    StringXmlWriterImpl* m_pimpl;

private:
    // Disable copy construction and assignment
    StringXmlWriter(const StringXmlWriter&);
    StringXmlWriter& operator=(const StringXmlWriter&);
};

Here’s the resulting implementation file:

// StringXmlWriter.cpp

#include "StringXmlWriter.h"

#define ARRAYSIZE(x) ( sizeof(x) / sizeof(x[0]) )

struct CharTranslation
{
    char OriginalChar;
    const char* ReplacementString;
};

static const CharTranslation AttributeValueTranslations[] =
{
    { '"', "&quot;" },
    { '&', "&amp;" },
};

static const CharTranslation CharDataTranslations[] =
{
    { '&', "&amp;" },
    { '<', "&lt;" },
    { '>', "&gt;" },
};

struct OriginalCharEquals :
    public std::binary_function<CharTranslation, char, bool>
{
    bool operator() (const CharTranslation& translation, char ch) const
    {
        return (translation.OriginalChar == ch);
    }
};

static std::string TranslateString(const std::string& originalStr,
                                   const CharTranslation* translations,
                                   int numTranslations)
{
    // Actually one past end, needed for proper std::find_if semantics
    const CharTranslation* endTranslations = translations + numTranslations;

    std::string translatedStr;

    for (std::string::const_iterator stringIter = originalStr.begin();
         stringIter != originalStr.end();
         ++stringIter)
    {
        char ch = *stringIter;

        const CharTranslation* translation = std::find_if
            (
            translations,
            endTranslations,
            std::bind2nd(OriginalCharEquals(), ch)
            );
        if (translation != endTranslations)
        {
            translatedStr += translation->ReplacementString;
        }
        else
        {
            translatedStr += ch;
        }
    }

    return translatedStr;
}

struct StringXmlWriter::StringXmlWriterImpl
{
    StringXmlWriterImpl(Formatting formatting,
                        int indentation,
                        char indentChar) :
        m_writeState(WriteState_Start),
        m_formatting(formatting),
        m_contentWritten(false)
    {
        for (int i = 0; i < indentation; ++i)
        {
            m_indentStr += indentChar;
        }
    }

    // PRIVATE TYPES
    // =============
    enum WriteState
    {
        WriteState_Attribute, // An attribute value is being written
        WriteState_Content, // Element content is being written
        WriteState_Element, // An element start tag has been written (and is unclosed)
        WriteState_Prolog, // The prolog is being written
        WriteState_Start, // No Write() methods have been called
    };

    struct OpenElement
    {
        explicit OpenElement(const std::string& localName) :
            QName(localName)
        {
        }

        explicit OpenElement(const std::string& localName,
                             const std::string& prefix) :
            QName(prefix.empty() ? localName : prefix + “:” + localName)
        {
        }

        // The qualified name (namespace prefix-included) of the
        // opened element
        std::string QName;
        // All namespaces declared in this element (maps namespace
        // to namespace prefix)
        typedef std::map<std::string, std::string> Namespaces_t;
        Namespaces_t Namespaces;
    };

    // PRIVATE MEMBERS
    // ===============
    WriteState m_writeState;

    // Need to use a vector instead of a stack because we must be able
    // to iterate over each opened element in the stack to see if a
    // namespace has already been declared.
    typedef std::vector<OpenElement> OpenedElements_t;
    OpenedElements_t m_openedElements;

    // Needed to track whether content was written inside the XML element
    // so we know how to handle indentation.
    bool m_contentWritten;

    // The style of formatting we are using.
    StringXmlWriter::Formatting m_formatting;

    // The string to use for a single level of indentation.
    std::string m_indentStr;

    // The XML fragment that this class has generated so far.  There’s no
    // guarantee it will be valid unless WriteEndDocument() is called.
    std::string m_xmlStr;	

    // PRIVATE FUNCTIONS
    // =================
    std::string GetExistingNamespacePrefix(const std::string& ns)
    {
        for (OpenedElements_t::const_iterator openElemIter = m_openedElements.begin();
            openElemIter != m_openedElements.end();
            ++openElemIter)
        {
            OpenElement::Namespaces_t::const_iterator nsIter =
                openElemIter->Namespaces.find(ns);
            if (nsIter != openElemIter->Namespaces.end())
            {
                return nsIter->second;
            }
        }

        return “”;
    }

    std::string GetNextNamespacePrefix(const std::string& ns)
    {
        std::string nsPrefix;

        for (int i = 1; ; ++i)
        {
            std::stringstream ss;
            ss << “ns” << i;
            std::string nsPrefix = ss.str();
            if (!NamespacePrefixExists(nsPrefix))
                return nsPrefix;
        }
    }

    bool NamespacePrefixExists(const std::string& nsPrefix)
    {
        for (OpenedElements_t::const_iterator iter = m_openedElements.begin();
            iter != m_openedElements.end();
            ++iter)
        {
            for (OpenElement::Namespaces_t::const_iterator nsIter = iter->Namespaces.begin();
                nsIter != iter->Namespaces.end();
                ++nsIter)
            {
                if (nsIter->second == nsPrefix)
                    return true;
            }
        }

        return false;
    }

    void CloseOpenElement()
    {
        m_xmlStr += ‘>’;
        m_writeState = WriteState_Content;
    }

    void NewlineAndIndent()
    {
        assert(m_formatting == Formatting_Indented);

        if (!m_xmlStr.empty())
            m_xmlStr += ‘\n’;

        for (int i = 0; i != m_openedElements.size(); ++i)
        {
            m_xmlStr += m_indentStr;
        }
    }
};

StringXmlWriter::StringXmlWriter(Formatting formatting,
                                 int indentation,
                                 char indentChar) :
    m_pimpl(new StringXmlWriterImpl(formatting, indentation, indentChar))
{
}

StringXmlWriter::~StringXmlWriter()
{
    delete m_pimpl;
}

std::string StringXmlWriter::GetXmlString() const
{
    return m_pimpl->m_xmlStr;
}

void StringXmlWriter::WriteAttributeString(const std::string& localName,
                                           const std::string& text)
{
    WriteStartAttribute(localName);
    WriteString(text);
    WriteEndAttribute();
}

void StringXmlWriter::WriteAttributeString(const std::string& localName,
                                           const std::string& ns,
                                           const std::string& text)
{
    WriteStartAttribute(localName, ns);
    WriteString(text);
    WriteEndAttribute();
}

void StringXmlWriter::WriteComment(const std::string& text)
{
    switch (m_pimpl->m_writeState)
    {
    case StringXmlWriterImpl::WriteState_Element:
        // An element is currently open.  Close the element so we can open
        // a new one.
        m_pimpl->CloseOpenElement();
        // FALL THROUGH
    case StringXmlWriterImpl::WriteState_Content:
    case StringXmlWriterImpl::WriteState_Prolog:
    case StringXmlWriterImpl::WriteState_Start:
        if (m_pimpl->m_formatting == Formatting_Indented)
        {
            m_pimpl->NewlineAndIndent();
        }

        m_pimpl->m_xmlStr += “<!–”;
        m_pimpl->m_xmlStr += text;
        m_pimpl->m_xmlStr += “–>”;
        break;
    default:
        // It doesn’t make sense to allow writing comments when writing an
        // attribute value.
        // TODO: Generate error
        break;
    }
}

void StringXmlWriter::WriteElementString(const std::string& localName,
                                         const std::string& text)
{
    WriteStartElement(localName);
    WriteString(text);
    WriteEndElement();
}

void StringXmlWriter::WriteElementString(const std::string& localName,
                                         const std::string& ns,
                                         const std::string& text)
{
    WriteStartElement(localName, ns);
    WriteString(text);
    WriteEndElement();
}

void StringXmlWriter::WriteEndAttribute()
{
    switch (m_pimpl->m_writeState)
    {
    case StringXmlWriterImpl::WriteState_Attribute:
        m_pimpl->m_xmlStr += ‘”‘;
        m_pimpl->m_writeState = StringXmlWriterImpl::WriteState_Element;
        break;
    default:
        // TODO: Generate error
        break;
    }
}

void StringXmlWriter::WriteEndDocument()
{
    switch (m_pimpl->m_writeState)
    {
    case StringXmlWriterImpl::WriteState_Attribute:
        WriteEndAttribute();
        // FALL THROUGH
    case StringXmlWriterImpl::WriteState_Content:
    case StringXmlWriterImpl::WriteState_Element:
        while (!m_pimpl->m_openedElements.empty())
        {
            WriteEndElement();
        }
        break;
    case StringXmlWriterImpl::WriteState_Start:
    case StringXmlWriterImpl::WriteState_Prolog:
        // DO NOTHING
        break;
    default:
        // TODO: Generate error
        break;
    }

    m_pimpl->m_writeState = StringXmlWriterImpl::WriteState_Start;
}

void StringXmlWriter::WriteEndElement()
{
    switch (m_pimpl->m_writeState)
    {
    case StringXmlWriterImpl::WriteState_Content:
        {
            std::string qname = m_pimpl->m_openedElements.back().QName;
            m_pimpl->m_openedElements.pop_back();

            if (!m_pimpl->m_contentWritten &&
                m_pimpl->m_formatting == Formatting_Indented)
            {
                m_pimpl->NewlineAndIndent();
            }

            m_pimpl->m_xmlStr += “</”;
            m_pimpl->m_xmlStr += qname;
            m_pimpl->m_xmlStr += ‘>’;
            m_pimpl->m_writeState = StringXmlWriterImpl::WriteState_Content;
            break;
        }
    case StringXmlWriterImpl::WriteState_Element:
        {
            m_pimpl->m_xmlStr += “/>”;
            m_pimpl->m_openedElements.pop_back();
            m_pimpl->m_writeState = StringXmlWriterImpl::WriteState_Content;
            break;
        }
    default:
        // TODO: Generate error
        break;
    }
}

void StringXmlWriter::WriteStartAttribute(const std::string& localName)
{
    WriteStartAttribute(localName, “”);
}

void StringXmlWriter::WriteStartAttribute(const std::string& localName,
                                          const std::string& ns)
{
    switch (m_pimpl->m_writeState)
    {
    case StringXmlWriterImpl::WriteState_Element:
        {
        std::string nsPrefix;
        bool mustDeclareNamespace = false;

        if (!ns.empty()) {
            nsPrefix = m_pimpl->GetExistingNamespacePrefix(ns);
            if (nsPrefix.empty()) {
                nsPrefix = m_pimpl->GetNextNamespacePrefix(ns);
                m_pimpl->m_openedElements.back().Namespaces[ns] = nsPrefix;
                mustDeclareNamespace = true;
            }
        }

        if (mustDeclareNamespace) {
            m_pimpl->m_xmlStr += ” xmlns:”;
            m_pimpl->m_xmlStr += nsPrefix;
            m_pimpl->m_xmlStr += “=”";
            m_pimpl->m_xmlStr += ns;
            m_pimpl->m_xmlStr += ‘”‘;
        }

        m_pimpl->m_xmlStr += ‘ ‘;
        if (!nsPrefix.empty()) {
            m_pimpl->m_xmlStr += nsPrefix;
            m_pimpl->m_xmlStr += ‘:’;
        }
        m_pimpl->m_xmlStr += localName;
        m_pimpl->m_xmlStr += “=”";
        m_pimpl->m_writeState = StringXmlWriterImpl::WriteState_Attribute;
        break;
        }
    default:
        // TODO: Generate error
        break;
    }
}

void StringXmlWriter::WriteStartDocument()
{
    switch (m_pimpl->m_writeState)
    {
    case StringXmlWriterImpl::WriteState_Start:
        m_pimpl->m_xmlStr += “<?xml version=”1.0”?>”;
        m_pimpl->m_writeState = StringXmlWriterImpl::WriteState_Prolog;
        break;
    default:
        // TODO: Generate error
        break;
    }
}

void StringXmlWriter::WriteStartElement(const std::string& localName)
{
    WriteStartElement(localName, “”);
}

void StringXmlWriter::WriteStartElement(const std::string& localName,
                                        const std::string& ns)
{
    switch (m_pimpl->m_writeState)
    {
    case StringXmlWriterImpl::WriteState_Element:
        // An element is currently open.  Close the element so we can open
        // a new one.
        m_pimpl->CloseOpenElement();
        // FALL THROUGH
    case StringXmlWriterImpl::WriteState_Content:
    case StringXmlWriterImpl::WriteState_Prolog:
    case StringXmlWriterImpl::WriteState_Start:
        {
        if (m_pimpl->m_formatting == Formatting_Indented)
        {
            m_pimpl->NewlineAndIndent();
        }

        std::string nsPrefix;
        bool mustDeclareNamespace = false;

        if (!ns.empty()) {
            nsPrefix = m_pimpl->GetExistingNamespacePrefix(ns);
            if (nsPrefix.empty()) {
                nsPrefix = m_pimpl->GetNextNamespacePrefix(ns);
                mustDeclareNamespace = true;
            }
        }

        StringXmlWriterImpl::OpenElement openElement(localName, nsPrefix);
        if (mustDeclareNamespace) {
            openElement.Namespaces[ns] = nsPrefix;
        }

        m_pimpl->m_openedElements.push_back(openElement);

        m_pimpl->m_xmlStr += ‘<’;
        if (!nsPrefix.empty()) {
            m_pimpl->m_xmlStr += nsPrefix;
            m_pimpl->m_xmlStr += ‘:’;
        }
        m_pimpl->m_xmlStr += localName;
        if (mustDeclareNamespace) {
            m_pimpl->m_xmlStr += ” xmlns:”;
            m_pimpl->m_xmlStr += nsPrefix;
            m_pimpl->m_xmlStr += “=”";
            m_pimpl->m_xmlStr += ns;
            m_pimpl->m_xmlStr += ‘”‘;
        }

        m_pimpl->m_writeState = StringXmlWriterImpl::WriteState_Element;
        m_pimpl->m_contentWritten = false;
        break;
        }
    default:
        // TODO: Generate error
        break;
    }
}

void StringXmlWriter::WriteString(const std::string& text)
{
    switch (m_pimpl->m_writeState)
    {
    case StringXmlWriterImpl::WriteState_Attribute:
        m_pimpl->m_xmlStr += TranslateString
            (
            text,
            AttributeValueTranslations,
            ARRAYSIZE(AttributeValueTranslations)
            );
        break;
    case StringXmlWriterImpl::WriteState_Element:
        // An element is currently open.  Close the element so we can start
        // writing the element content.
        m_pimpl->CloseOpenElement();
        // FALL THROUGH
    case StringXmlWriterImpl::WriteState_Content:
        m_pimpl->m_xmlStr += TranslateString
            (
            text,
            CharDataTranslations,
            ARRAYSIZE(CharDataTranslations)
            );
        m_pimpl->m_contentWritten = true;
        break;
    default:
        // TODO: Generate error
        break;
    }
}

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s