lua-users home
lua-l archive

[Date Prev][Date Next][Thread Prev][Thread Next] [Date Index] [Thread Index]


Hello,

Sorry, but I need to correct one point here...

>  Wed, 22 Jan 2003 15:16:27 -0200
>  Luiz Henrique de Figueiredo <lhf@tecgraf.puc-rio.br> wrote:
>
[...]
>  
>  On the other hand, I do have a draft IEEE math library that I'll probably
>  distribute when Lua 5.0 final is released. It's pretty simple anyway.
>  
>  Finally, you can write those in Lua:
>  
>   function isnan(x) return x~=x end
>   function isinfinity(x) return x+1==x end

which mathematically  would be right, but numerically  would be wrong.
By  this, you  would get  something like  the inverse  of  the machine
epsilon which is defined as the smallest number for which

1+eps > 1 

in  the  floating  point  arithmetic.   In ANSI  C,  this  is  exactly
DBL_EPSILON  resp.   FLT_EPSILON,  which  for  IEEE  have  the  values
2.2204460492503131e-16 resp. 1.19209290e-07.


If the  IEEE rules (which I don't  know well) 1.79769313486231570e+308
would state that

infinity>x 

for all floating point numbers, we could use something like

function isinfinity(x) return x>DBL_MAX end

where  DBL_MAX takes  the value  1.79769313486231570e+308  in float.h.
But please  don't enter  this value  from within Lua,  as there  is at
least  some danger  that sscanf  will mess  up the  value.  Rather, it
should be entered  using lua_pushvalue from within C  to get the exact
valu.


Perhaps it  was a wise  decision by the  ANSI commitee not  to include
this stuff into ANSI C because on many systems, this functionality can
be    toggled     via    assembler    statements     or    environment
variables.  E.g.  many  numerical  anlaysts rather  like  to  generate
exceptions on 1/0 than checking  for infinity or nan, and switch these
feature off.  On VAXen, you have floating numbers not covered by IEEE.

Juergen Fuhrmann