Re: [boost] [random] fix assert in uniform_real.hpp

From: Neal Becker (ndbecker2_at_[hidden])
Date: 2007-05-14 07:47:30

Matthias Troyer wrote:

> 
> On 13 May 2007, at 10:45, Neal Becker wrote:
> 
>> Matthias Troyer wrote:
>>
>>>
>>> On 11 May 2007, at 23:13, Neal Becker wrote:
>>>
>>>> diff -r 5fcada374f41 boost/random/uniform_real.hpp
>>>> --- a/boost/random/uniform_real.hpp     Thu May 10 08:59:36 2007
>>>> -0400
>>>> +++ b/boost/random/uniform_real.hpp     Fri May 11 11:12:44 2007
>>>> -0400
>>>> @@ -40,7 +40,7 @@ public:
>>>>  #ifndef BOOST_NO_LIMITS_COMPILE_TIME_CONSTANTS
>>>>      BOOST_STATIC_ASSERT(!
>>>> std::numeric_limits<RealType>::is_integer);
>>>>  #endif
>>>> -    assert(min_arg < max_arg);
>>>> +    assert(min_arg <= max_arg);
>>>>    }
>>>>
>>>>    // compiler-generated copy ctor and assignment operator are fine
>>>
>>> This does not make sense to me. uniform_real should create numbers u
>>> that are
>>>
>>> min_arg <= u < maxarg
>>>
>>> Thus choosing minarg == maxarg does not make sense since the set of
>>> possible values would be empty.
>>>
>> It's a valid limiting case.
>>
>> Here's the use case:
>>
>> Your program has a random number generator.  The range is set by
>> command
>> line argument.  For test purposes, you set the range to be
>> infinitesimal.
>> As it is now, you get the assertion failure.  There is no good
>> reason for
>> this to fail, AFAICT.  Allowing it is useful and harmless.
> 
> Sorry, the range is not infinitesimal but is empty. What value should
> the generator return when minarg == maxarg? In that case there is no
> legal value. Should the generator check and throw when invoked?
> 

Agreed, I forgot that range was [min..max)