Richard Fateman wrote:
> Raymond Toy wrote:
>
>> ....
>>
>
>
>> But when invertbigfloat calls fpquotient, there is a
>> problem. fpquotient is called:
>>
>> (FPQUOTIENT (36028797018963968 1) (535045584704600947821 2))
>>
>>
>> The first argument is 1b0, using fpprec=56 bits. The second arg is
>> gamma(1/4), using 69 bits. But I'm pretty sure these fp functions
>> assume all args are supposed to have the same precision.
>>
>>
>
> yes, this would be a real problem if the two numbers are different
> precisions and the headers are missing.
> If you put the headers back, and did the division, e.g.
> ( ...divide.. ((bigfloat simp 56) 360..... 1) (bigfloat simp 69)
> 53..... 2))
>
>
In this particular case, the bogus call comes from invertbigfloat. My
proposed solution is to call bigfloatp on the argument to convert it to
the correct precision. (I would never have suspected bigfloatp would
actually convert the bigfloat!)
This fixes the immediate problem. There may also be some issues in
exptbigfloat, but I haven't looked at that yet.
Ray