RFC: Building temporary char** property arrays

From: Ben Martin <monkeyiq_at_users.sourceforge.net>
Date: Sat Dec 11 2010 - 00:51:23 CET

  I notice a fair bit in some parts of Abiword that I'm hacking on code
like the following:
const gchar* ppAtts[10];
int i=0;
ppAtts[i++] = "foo";
ppAtts[i++] = 0;

Where ppAtts is a NULL terminated list of pointers to char* which are
not duplicated/freed. ie, const strings or std::string::c_str() where
the string object lasts beyond the scope that ppAtts is required for.

I was thinking of replacing some / phasing in the use of boost::array
for these. The gains being that the allocation of the array on the stack
is retained, explicit null termination becomes optional, and the "top"
array index can be maintained by the class for you (avoiding i++ in the
above). So the code becomes something like this:

propertyArray<> ppAtts;
ppAtts.push_back( "foo" );
// optional
ppAtts.push_back( 0 );


As boost::array is a template I have left my prototype as one too, with
a default size of 32 which should cover the bulk of these temporary
arrays. The main motivation was to avoid tracking multiple array/index
pairs as autovars and also in cases where there is one or more if()
blocks that complicate the flow while building the array. In these cases
knowing that explicit null termination is not required might avoid
subtle bugs depending on the code paths executed.

At the moment I zero the array at construction time, but the null
termination could be added in the data() method to avoid initialization
overhead. I used delegation to boost::array rather than inheritance to
allow better control over m_highestUsedIndex and possible lazy null

Thoughts welcome... class below...

#include <boost/array.hpp>

 * As at 2010 there are many pieces of code that declare an array of
 * char* on the stack to set 2-10 parameters and pass this char** to a
 * function to use. It is ugly and somewhat error prone to maintain
 * the index into this array manually and also if code happens to
 * overrun the array there is no error detection for that. By using
 * boost::array at least these out of bounds cases can be cought and
 * code does not have to worry about NULL terminating the array
 * itself. The array is allocated as part of the object on the stack
 * though so we avoid memory allocation and zeroing issues along with
 * possible fragmentation over time. Note that the [i++] style should
 * also work with propertyArray, and count() gives you the index of
 * the first NULL value in the array.
 * Example usage:
 * propertyArray<> ppAtts;
 * ppAtts.push_back( PT_REVISION_ATTRIBUTE_NAME );
 * ppAtts.push_back( "foo" );
 * ppAtts.push_back( 0 ); // optional
 * somethingThatWantsArrayOfCharPointers(ppAtts.data());
 * and the array goes away here.
 * This replaces the style below, note that the code is similar,
 * though the explicit index "i" is not needed above.
 * const gchar* ppAtts[10];
 * bzero(ppAtts, 10 * sizeof(gchar*));
 * int i=0;
 * ppAtts[i++] = "foo";
 * ppAtts[i++] = 0;
 * somethingThatWantsArrayOfCharPointers(ppAtts);
template< std::size_t N = 32 >
class propertyArray
    typedef const gchar* T;
    typedef boost::array< const gchar*, N > m_boostArray_t;
    typedef typename m_boostArray_t::value_type value_type;
    typedef typename m_boostArray_t::iterator iterator;
    typedef typename m_boostArray_t::const_iterator const_iterator;
    typedef typename m_boostArray_t::reference reference;
    typedef typename m_boostArray_t::const_reference const_reference;
    typedef std::size_t size_type;

        m_highestUsedIndex( 0 )
        m_array.assign( 0 );
    void push_back( const gchar* v )
        std::size_t sz = count();
        m_array.at(sz) = v;

    std::size_t count() const
        return m_highestUsedIndex;

    const_reference operator[](size_type i) const
        return m_array.at(i);
    static size_type size() { return N; }
    T* data() { return m_array.data(); }

    std::size_t m_highestUsedIndex;
    m_boostArray_t m_array;

Received on Sat Dec 11 00:51:42 2010

This archive was generated by hypermail 2.1.8 : Sat Dec 11 2010 - 00:51:42 CET