Georges Khaznadar wrote: > Hello Thiemo, > > Thank you for this patch. It is correct for the amd64 architecture too, > as far as the test suite can assert it. > > I have included your patch in the revision 2.4 which is now available in > source form (and as binaries for i386) at ofset's repository, see > http://debian.ofset.org/ . > > I suppose that the program as you modified it should work even with 32 > bits architectures,
It is not a 32-bit vs. 64-bit issue in the usual sense. Most architectures supported by Debian provide only 64-bit wide FP values, and make "long double" an alias for "double". This is a hardware limitation. The exceptions, as far as I know, are: i386, amd64: 80-bit wide "long doubles" sparc: 128-bit wide "long doubles" > as the relevant variables are still declared as long > double, which is mandatory The relevant standards only require a "long double" to be equal or larger than "double", there is no guarantee about the actual size. (Checking the actual values in float.h is a way to know for sure.) > (as enthalpies of reactions can be computed > by combining two redox reaction halves, their values can be out of the > range allowed for the type double, but will remain still in the range > allowed for long double). How bad is this effect? Will it make the package useless on most architectures, or will it hit only in some cornercases? Thiemo -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
