Overload stream operator for nested template C++ STL - c++

Question

I've got a data file that is a single line consisting of a nested series of doubles eg.
[[0.127279,0.763675,0.636396],[0.254558,0.890955,0.636396],
[0.127279,0.636396,0.763675],[0.254558,0.763675,0.763675],
[0.381838,0.890955,0.763675],[0.127279,0.509117,0.890955],
[0.254558,0.636396,0.890955],[0.509117,0.890955,0.890955]]
I'd like to be able to read this into a STL vector<vector<double> > using the stream operator which is templated across the inner type of A:
vector<vector<double> > A;
FIN >> A;
I've figured out a way to do this when the vector is not nested, ie. a simple vector<T> as so:
template <class T>
istream& operator>>(istream& s, vector<T> &A){
T x;
string token; char blank;
s >> blank; // Gobble the first '['
while( getline(s, token, ',') ) {
istringstream input(token);
input >> x;
A.push_back(x);
}
s >> blank; // Gobble the last ']'
return s;
}
But I'm having problem with the istream& operator>>(istream& s, vector<vector<T> >&A) part as I can't seem to catch the inner ]'s properly. I'm sure that Boost has a way of doing this, but I'd like to see a solution with the STL for pedagogical purposes.
Note: I'm aware that overloading the stream operator for vector<T> can have far-reaching undesirable consequences and that the implementation should be wrapped up in its own class - I'm using this example above as it stands to clarify the question.
EDIT:
I'd like the method to be robust enough to handle a input array whose size (and inner array) size is not known in advance, but inferred from reading the stream.

Answer 1

Actually, the problem with your code that you want to use the same function for both, when T is:
vector<double>
double
But the logic which needs to read the data into vector and double is slightly different. So you cannot do that, at least not with that simple logic:
I would prefer to write two functions, to handle both cases separately. After all, even in your case, the compiler will generate two different functions for each value of T.
template <class T>
istream& operator>>(istream& s, vector<T> &A)
{
T x;
string token; char blank;
s >> blank; // Gobble the first '['
while( getline(s, token, ',') )
{
istringstream input(token);
input >> x;
A.push_back(x);
}
// s >> blank; // Gobble the last ']'
return s;
}
template <class T>
istream& operator>>(istream& s, vector<vector<T>> &A)
{
vector<T> x;
string token;
char blank;
s >> blank; // Gobble the first '['
while( getline(s, token, ']') )
{
istringstream input(token);
input >> x;
s >> blank; //read , after [...]
A.push_back(x);
x.clear();
}
s >> blank; // Gobble the last ']'
return s;
}
Test code:
int main() {
vector<vector<double>> A;
cin >> A;
for(size_t i = 0 ;i < A.size(); ++i)
{
for(size_t j = 0 ; j < A[i].size(); ++j)
cout << A[i][j] <<" ";
cout << endl;
}
return 0;
}
Input:
[[1,2,3],[4,5,6],
[7,8,9],[10,11,12],
[13,14,15],[16,17,18],
[19,20,21],[22,23,24]]
Output:
1 2 3
4 5 6
7 8 9
10 11 12
13 14 15
16 17 18
19 20 21
22 23 24
Online demo : http://ideone.com/iBbmw

Answer 2

In your particular example which is very simple.
Read the whole line into a string.
Replace all [ , ] and , with whitespace character.
Create a simple stringstream with whitespace replaced string.
Now you can have a a simple loop of
double x;
while( stringstreamp >> x )
{
}
And some special logic after reading three doubles to insert them them into a new array.

Answer 3

Some years later, and I was here struggling with the same problem.
Based on your contribution, I developed a modified version of the original template. This one is able to parse multidimensional arrays, even if they are spread across several lines.
template <class T>
istream& operator>>(istream& s, vector<T> &A){
while(true){
T x;
char c = s.peek();
if( c == '[' || c == ','){
s.ignore();
if(s >> x) A.push_back(x);
else throw invalid_argument("Bad, bad JS array!");
continue;
}
if( c == ']') {
s.ignore();
return s;
}
/* Ignore line-break */
s.ignore();
}
return s;
}
Hope this can be useful for someone.