Re: How to create nan/inf

iirc, FFFFFFFFFFFFFFFF is NaN on Windows boxes (at least with Intel Visual Fortran and C), but I can't swear that's guaranteed to be true on all x86 machines, let alone all machines using IEEE arithmetic. I know that some platforms, e.g., IBM mainframes, don't have floating point models capable of representing NaN or Inf, so a completely portable solution is impossible.

Information about American English usage here and here. Floating point issues? Please read this before posting. — emc

Comment on Re: How to create nan/inf

Replies are listed 'Best First'.
Re^2: How to create nan/inf by syphilis (Archbishop) on Aug 09, 2008 at 06:22 UTC
so a completely portable solution is impossible That's not really an issue for me. I was just after some way of getting test 6 of PowInt.t to pass on Windows, leaving all other systems unaffected. I think I've now achieved that. One of the things that was throwing me was that, on Linux, the strings 'nan' and 'inf' are converted to a nan/inf when used in numeric context. That is, on Linux: `perl -MDevel::Peek -e 'Dump "inf" + 1';` [download] outputs: `SV = NV(0x...) at 0x.... REFCNT = 1 FLAGS = (NOK,READONLY,pNOK) NV = inf` [download] But there's no comparable behaviour (that I can find) on Windows. The strings '1.#INF' and '1.#QNAN' are not interpreted as inf/nan when used in numeric context. Instead they're just numified according to the usual rules that apply to the numification of strings - ie they take on a numeric value of 1: `C:\>perl -MDevel::Peek -e "Dump '1.#INF' + 1" SV = IV(0x218fe28) at 0x218fe2c REFCNT = 1 FLAGS = (PADTMP,IOK,READONLY,pIOK) IV = 2` [download] I wasn't expecting such different behaviour - and that added significantly to my confusion. Cheers, Rob	[reply] [d/l] [select]
Re^2: How to create nan/inf by ikegami (Patriarch) on Aug 09, 2008 at 06:45 UTC
Double `FFFFFFFFFFFFFFFF` is a IEEE 754 NaN. Anything of the form `[7F]FFFxxxxxxxxxxxx` where "`x`" is nonzero is a NaN. There are at least two kinds of NaN: signaled and quiet. IEEE 754 doesn't define the how to encode them, but recommends using the most significant bit of "`x`" (0 for signaled).	[reply] [d/l] [select]