Re: [boost] [constrained_value] need help with runtime error

Subject: Re: [boost] [constrained_value] need help with runtime error
From: Neal Becker (ndbecker2_at_[hidden])
Date: 2008-12-10 14:57:05

Steven Watanabe wrote:

> AMDG
> 
> Neal Becker wrote:
>> This code has an error:
>>
>> #include <boost/numeric/ublas/vector.hpp>
>> #include <boost/constrained_value.hpp>
>>
>> namespace cv = boost::constrained_value;
>> namespace ublas = boost::numeric::ublas;
>>
>> typedef cv::bounded<int>::type val_t;
>> typedef cv::bounded<int>::type::constraint_type constraint_type;
>>
>> int main() {
>>   int min = -200;
>>   int max = 200;
>>   val_t init (10, constraint_type (min, max));
>>   ublas::vector<val_t> v (10, init);
>>   ublas::vector<val_t> w = -v;
>> }
>>
>> This code will throw an exception.  The last line invokes:
>> ublas/functional.hpp:70 (scalar_negate), which for some reason calls
>>
>> constrained.hpp:224, which is:
>>     constrained(const value_type & v)
>>
>> That is, initialize only the value, not the constraint!  Any ideas?
>>   
> 
> It looks like the arithmetic operators may need to be overloaded
> rather than relying on implicit conversions?
> 
> The error should occur in a simpler case:
> 
> val_t x(10, constraint_type(min, max));
> val_t neg_x(-x);
> 

Yes, this gives the same error:

typedef cv::bounded<int>::type val_t;
typedef cv::bounded<int>::type::constraint_type constraint_type;

// template<typename V, typename C, typename E>
// cv::constrained<V,C,E> & operator- (cv::constrained<V,C,E> c) {
//   V tmp (c.value());
//   c = -tmp;
//   return c;
// }

int main() {
  int min = -200;
  int max = 200;
  val_t init (10, constraint_type (min, max));
  val_t neg_x(-init);

If the above operator overload is uncommented, then it does not throw on the last line above, but it STILL throws on the original vector operation:

 ublas::vector<val_t> v (10, init);
 ublas::vector<val_t> w = -v;

I can't figure out why the ublas scalar_negate doesn't use my operator-.