sqrt interpreter Q

Thu Feb 7 17:34:00 GMT 2002

Tom Tromey wrote:
>>>>>>"Andreas" == Andreas Tobler <toa@pop.agri.ch> writes:
>>>>>>>>Andreas> Doing it with gij I get the following:
>Andreas> [titanium:~] andreast% gij my_sqrt
>Andreas> sqrt_int: -2147483648
>Andreas> sqrt_double: NaN
>Andreas> sqrt_long: -9223372034707292160
>>Andreas> Is this obviously a problem in the libffi ? Or somewhere else?
>>Without debugging it is hard to say. But I would say that it is a
>good guess.
>>Try doing the same thing via reflection. If that also gets the
>incorrect answer then libffi is more strongly implicated.
>
For some reason my answer to this didn't make it to the list. I'm pretty 
sure this is not a libffi bug but rather an endian bug in either either 
"gcj -C" or libjava. The same problem occurs on Linux PPC. Reflection 
does have similar problems with sub-int sized types, but that may be a 
different issue.
regards
Bryce.