Er actually I mean NaN and +/-inf and negation .. as applied
go floats :)
I seek feedback, opinions, etc ...
/Users/johnskaller/felix/build/release/tut/tutorial/tut-02.02.05-0.flx: line 6,
cols 10 to 9
5: | NaN => { print "Not a Number"; }
6: | -inf .. -0.999 => { print "lt -1"; }
*
7: | -1.001 .. 1.001 => { print "Unit circle"; }
Q1: what to do with
NaN // not a legit value
+inf // legit xfloat value
-inf // legit xfloat value
-1.0 // legit float value
Denorm
Here: float is your ordinary floats (float, double, ldouble).
xfloat means floats together with + and - infinity.
This is a legitimate number system, but it has different
laws of arithmetic. For example
1.0 / 0.0 => +inf
in xfloat, but an error in float. Note that Complex usually doesn't
allow +/- inf, instead it has plain inf. This makes complex number
system complete.
NaN is not a floating value as such. It may be considered
part of float or xflloat to make 2 new number systems,
similar to
float_option = | Float of float | NaN
i.e. adding an error values .. BUT .. this is NOT how it
normally works. In particular NaN != NaN is actually
the defining property of NaN (more precisely,
NaN != x for any value x, **including** NaN).
As such, NaN is not a legitimate value when coupled
with the usual way of thinking about equality.
We may have
isNaN(x)
and similarly
match x with | NaN => ...
make sense. Of course NaN doesn't exist if you have
signalling NaNs and signalling is switched on.
Grrr .. then we have signalling and non-signalling NaNs ....
Denorm is a warning. Denorm values are legit values but
they have reduced precision due to the exponent maxing out.
Denorm is more of an issue if you switch strict IEEE floating behaviour
off. This is useful because it is much faster on some FPUs.
Denorm doesn't exist in xfloat because we just switch to +inf
or -inf on exponent overflow, and 0.0 on underflow.
Finally the easiest one: negative values. Yes, sure, Felix has
negative floats! The question is: does it have negative valued
literals and the answer at the moment is NO. Not for floats
or for integers.
Since matches ONLY allow literals, you cannot match on a negative
value directly. You have to do this instead:
match x with | ?a when a == -1.0 => ...
or equivalently
match x with | $(-1.0) => ..
Unfortunately you cannot do that with ranges:
match x with | $(-1.0) .. 1.0 => ...
doesn't work now. It used to but now, pattern matching is simplified
and generalised to work with ANY literals.
Q2: WTF do we do with all the slice notations!
In pattens:
1 .. 2 ==> range 1 to 2 inclusive
In for loops:
for i in 1 upto 2 ==> range 1 to 2 inclusive
In string substrings:
x.[1 to 2] ==> range 1 to 2
Strings also allow negatives. We also have a slice object in the library.
We also have an iterator that works on slices (but is implemented
with yield inside a for .. upto .. loop)
We seriously need to get rid of this mess!
--
john skaller
skal...@users.sourceforge.net
http://felix-lang.org
------------------------------------------------------------------------------
Better than sec? Nothing is better than sec when it comes to
monitoring Big Data applications. Try Boundary one-second
resolution app monitoring today. Free.
http://p.sf.net/sfu/Boundary-dev2dev
_______________________________________________
Felix-language mailing list
Felix-language@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/felix-language
