One benefit of Float64 vs Float is that it helps make clear what a floating 
point number actually is: 64 bits. It can't represent all possible real 
numbers, and that's why we have things like 0.1 != 1//10.

I don't think typing the extra "64" is an undue burden: for most code using 
Number, Real, FloatingPoint types are typically sufficient for dispatch, 
and if you are dispatching on different floating point types, the 
bit-length would probably be useful for clarity. A grep shows that 
"Float64" occurs only 750 times in all of the Base library, and that 
includes its definition and lots of external library calls (like libmath 
and LAPACK calls) which most people won't encounter.

The main reasons I typically use Float64 in non-library code are for 
creating blank arrays using Array(Float64,m,n), and explicit conversions on 
comprehensions Float64[f(x) for x in X]. There might be arguments for 
making such calls simpler (such as making Array Float64 by default, as it 
is for zeros and ones), but I don't think the Float alias would help with 
this.

-Simon

Reply via email to