FloatQ do not read like they print (et vice et versa)
By Nicolas Cellier - Posted on June 8th, 2009
Tagged:
| Project: | GNU Smalltalk |
| Component: | Base classes |
| Category: | bug |
| Priority: | normal |
| Assigned: | Unassigned |
| Status: | fixed |
Description
The pi example gave me a clue.
st>3.1415926535897932384626433832795028841971693993751058209749445923078164062862q sin exponent
-61
st>-1.0q arcCos sin exponent
-64
What, I lost three bits?
Hmm fairly sure about my 256bits approximation of pi though...
st>-1.0q arcCos
3.1415926535897932385
st>3.1415926535897932385q
3.1415926535897932383
st> 3.1415926535897932385q sin exponent
-63
Yes, there might be a problem for reading these extended precision.
Updates
#1 submitted by Paolo Bonzini on Mon, 06/15/2009 - 12:46
Among others, there was a bug printing numbers very close to 1, like (1.0d - (1.0d timesTwoPower: -52)) or the equivalent FloatQ.
I fixed this by performing the floorLog after converting to a fraction, but there are still bugs reading.
#2 submitted by Paolo Bonzini on Sat, 07/11/2009 - 11:03
| Status: | active | » fixed |
should be fixed now.