Rice distribution

[Paolo Carlini]

Hi,

On 09/13/2012 03:05 PM, Ed Smith-Rowland wrote:
> On 09/05/2012 10:29 PM, Ulrich Drepper wrote:
>> On Wed, Sep 5, 2012 at 6:34 PM, Ed Smith-Rowland <3dw4rd@verizon.net> 
>> wrote:
>>> I wrote some distributions a couple months back.  I wrote them as
>>> (name-uglified) extensions to libstdc++.
>>>
>>> Here is the Rice distribution.
>> I never used this distribution myself but that doesn't mean anything.
>> A few nits:
>>
>> - you should add __generate functions (see the current svn code)
> Forgot about that.  Fixed.
>>
>> - in the operator function, did you consider using hypot?  I'm not
>> sure it's necessary, the x and y numbers might not require it. hypot
>> is usually more accurate.  The multiplications are throwing away a lot
>> of bits which the sqrt call might need.
> Hypot is a good idea.  I'm not worried about overflow heresuch as 
> student-t.
>>
>> - the operator== has the bug which we fixed in the library.  You
>> cannot define a template friend function in the template class
>> definition.  Look at the current libstdc++ code.  Likewise for the
>> other friend functions.
> i think I got this right.  Other distros have things as I have them here.
>>
>> - how about moving the operator() code and perhaps even the operator==
>> code into random.tcc?
> They are small.  i left them where they wre but I could still move them.
>>
>> - move the operator!= outside the class definition, just like for the
>> other classes
> Done.
>>
>> - with the definition being in ext/random, the test cases shouldn't
>> reference the non-existing rice_distribution.h header
> I sent an old patch I guess.  Fixed.

I think the updated patch is Ok (as a general rule, tough, remember to 
always submit a ChangeLog entry together with the patch proper), you 
tested it, right? (better multilib, to catch issues having to do with 
32-bit longs, etc) In that case it's Ok to commit.

Thanks,
Paolo.