''Ogle this student code, to convert a number to base 36:''

 #include <iostream>
 #include <string>

 using std::abs;
 using std::cout;
 using std::endl;
 using std::reverse;
 using std::string;

 char charTable[] = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";

 void printBase(int i, int base)
 {
  if ((abs(base) <= 1) || (abs(base) > 36) || (i == 0))
  {
    cout << "0" << endl;
    return;
  }

  string str = "";
  if ((base > 0) && (i < 0))
  {
    str += "-";
    i = -i;
  }

  do {
    if (i % base >= 0)
    {
      str += charTable[i % base];
      i = i / base;
    } else {
      str += charTable[i % base - base];
      i = i /base + 1;
    }
  } while(i);
  reverse(str.begin(), str.end());
  cout << str << endl;
 }

 int main()
 {
  for(int i = -55; i <= 55; i++)
  {
    cout << i << " = ";
    printBase(i,-10);
  }
 }

--manoj

''Suppose TestDrivenDevelopment generated it. Would TDD be sufficient to cover its branches?''

''answers reposted from news:comp.lang.c++ without permission''

''Would BranchTesting be inadequate for converting an integer to a string???''

Certainly inadequate as a testing discipline in general, and almost 
certainly inadequate for testing a function that converts an integer to 
a string. Whether it gives you adequate coverage in any particular case 
depends on the case. Applied to the original code it is inadequate.

For example, branch analysis could be satisfied with test cases that 
only exercise the do loop exactly twice: once to branch back to the head 
of the loop and once to branch out of it. That doesn't test what happens 
when you need exactly one digit or three or more digits to represent the 
value. Of course, doing that one thoroughly requires path testing, which 
is known to be prohibitively expensive <g>, although perhaps workable 
for this example.

For another example, if the charTable had some incorrect values, a test 
suite based only on branch testing wouldn't necessarily find them.

--PeteBecker

At least insufficient. For example, branch testing would not detect
the error which is in the original code when passing
''std::numeric_limits<int>::min()'': the value range for integers is
[normally] asymmetric and "i = -i" is the identity operation for
the two values ''0'' and ''std::numeric_limits<int>::min()'' (well, for
the latter value the behavior may actually be undefined or
implementation defined). That is, in addition to branch testing you
need to perform at least some testing for CornerCase''''''s.

--DietmarKuehl