cppannotations/yo/classtemplates/concrete2template.yo

    An ordinary class may be used as the base class for deriving a template
        hi(class template: derived from ordinary class) class. The advantage
of such an inheritance tree is that the base class's members may all be
compiled beforehand. When objects of the class template are now instantiated
only the actually used members of the derived class template must be
instantiated.

    This approach may be used for all class templates having member functions
whose implementations do not depend on template parameters. These members may
be defined in a separate class which is then used as a base class of the
class template derived from it.

    As an illustration of this approach we'll develop such a class template
below. We'll develop a class tt(Table) derived from an ordinary class
tt(TableType). The class tt(Table) displays elements of some type in a
table having a configurable number of columns. The elements are either
displayed horizontally (the first em(k) elements occupy the first row) or
vertically (the first em(r) elements occupy a first column).

    When displaying the table's elements they are inserted into a stream.  The
table is handled by a separate class (tt(TableType)), implementing the table's
presentation. Since the table's elements are inserted into a stream, the
conversion to text (or tt(string)) is implemented in tt(Table), but the
handling of the strings themselves is left to tt(TableType). We'll cover some
characteristics of tt(TableType) shortly, concentrating on tt(Table)'s
interface first:
    itemization(
    it() The class tt(Table) is a class template, requiring only one template
type parameter: tt(Iterator) referring to an iterator to some data type:
        verbinsert(HEAD)(examples/table/table/table.h)
    it() tt(Table) doesn't need any data members. All data manipulations are
performed by tt(TableType).
    it() tt(Table) has two constructors. The constructor's first two
parameters are tt(Iterator)s used to iterate over the elements that must be
entered into the table. The constructors require us to specify
the number of columns we would like our table to have as well as a
em(FillDirection). tt(FillDirection) is an tt(enum),
defined by tt(TableType), having values tt(HORIZONTAL) and tt(VERTICAL). To
allow tt(Table)'s users to exercise control over headers, footers, captions,
horizontal and vertical separators, one constructor has tt(TableSupport)
reference parameter. The class tt(TableSupport) will be developed later as a
virtual class allowing clients to exercise this control. Here are the class's
constructors:
        verbinsert(CONS)(examples/table/table/table.h)
    it() The constructors are tt(Table)'s only two public members. Both
constructors use a base class initializer to initialize their tt(TableType)
base class and then call the class's private member tt(fill) to insert data
into the tt(TableType) base class object. Here are the constructor's
implementations:
        verbinsert(CONSIMP)(examples/table/table/table.h)
    it() The class's tt(fill) member iterates over the range of elements
rangett(begin, end), as defined by the constructor's first two parameters.
As we will see shortly, tt(TableType) defines a protected data member
tt(std::vector<std::string> d_string). One of the requirements of the data
type to which the iterators point is that this data type can be inserted into
streams. So, tt(fill) uses a tt(ostringstream) object to obtain the textual
representation of the data, which is then appended to tt(d_string):
        verbinsert(FILL)(examples/table/table/table.h)
    )

    This completes the implementation of the class tt(Table). Note that this
class template only has three members, two of them constructors. Therefore, in
most cases only two function templates will have to be instantiated: a
constructor and the class's tt(fill) member. For example, the following
defines a table having four columns, vertically filled by tt(string)s
extracted from the standard input stream:
        verb(
    Table<istream_iterator<string> >
        table(istream_iterator<string>(cin), istream_iterator<string>(),
              4, TableType::Vertical);
        )
    The fill-direction is specified as
tt(TableType::VERTICAL). It could also have been specified using tt(Table),
but since tt(Table) is a class template its specification would have been
slightly more complex: tt(Table<istream_iterator<string> >::VERTICAL).

    Now that the tt(Table) derived class has been designed, let's turn our
attention to the class tt(TableType). Here are its essential characteristics:
    itemization(
    it() It is a ordinary class, designed to operate as tt(Table)'s base
class.
    it() It uses various private data members, among which tt(d_colWidth), a
vector storing the width of the widest element per column and tt(d_indexFun),
pointing to the class's member function returning the element in
tt(table[row][column]), conditional to the table's fill
direction. tt(TableType) also uses a tt(TableSupport) pointer and a
reference. The constructor not requiring a tt(TableSupport) object uses the
tt(TableSupport *) to allocate a (default) tt(TableSupport) object and then
uses the tt(TableSupport &) as the object's alias. The other constructor
initializes the pointer to 0 and uses the reference data member to refer to
the tt(TableSupport) object provided by its parameter. Alternatively, a static
tt(TableSupport) object could have been used to initialize the reference data
member in the former constructor. The remaining private data members are
probably self-explanatory:
        verbinsert(DATA)(examples/table/tabletype/tabletype.h)
    it() The actual tt(string) objects populating the table are stored in a
protected data member:
        verbinsert(PROT)(examples/table/tabletype/tabletype.h)
    it() The (protected) constructors perform basic tasks: they initialize the
object's data members. Here is the constructor expecting a reference to a
tt(TableSupport) object:
        verbinclude(examples/table/tabletype/tabletype.cc)
    it() Once tt(d_string) has been filled, the table is initialized by
tt(Table::fill). The tt(init) protected member resizes tt(d_string) so
that its size is exactly tt(rows x columns), and it determines the maximum
width of the elements per column. Its implementation is straightforward:
        verbinclude(examples/table/tabletype/init.cc)
    it() The public member tt(insert) is used by the insertion operator
(oplshift()) to insert a tt(Table) into a stream. First it informs the
tt(TableSupport) object about the table's dimensions. Next it displays the
table, allowing the tt(TableSupport) object to write headers, footers and
separators:
        verbinclude(examples/table/tabletype/insert.cc)
    it() The tt(cplusplus.yo.zip) archive contains tt(TableSupport)'s full
implementation. This implementation is found in the directory
tt(yo/classtemplates/examples/table). Most of its remaining members are
private. Among those, these two members return table element
tt([row][column]) for, respectively, a horizontally filled table and a
vertically filled table:
      verbinsert(INDEX)(examples/table/tabletype/tabletype.h)
    )

    The support class tt(TableSupport) is used to display headers, footers,
captions and separators. It has four virtual members to perform those tasks
(and, of course, a virtual constructor):
    itemization(
    itt(hline(size_t rowIndex)): called just before the elements in row
tt(rowIndex) will be displayed.
    itt(hline()): called immediately after displaying the final row.
    itt(vline(size_t colIndex)): called just before the element in column
tt(colIndex) will be displayed.
    itt(vline()): called immediately after displaying all elements in a row.
    )
    The reader is referrred to the tt(cplusplus.yo.zip) archive for the full
implementation of the classes tt(Table), tt(TableType) and tt(TableSupport).
Here is a little program showing their use:
    verbinclude(examples/table/table.cc)