[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Chicken-users] (abs 2147483648) evaluates to negative number
From: |
John Cowan |
Subject: |
Re: [Chicken-users] (abs 2147483648) evaluates to negative number |
Date: |
Fri, 31 Jul 2009 11:29:49 -0400 |
User-agent: |
Mutt/1.5.13 (2006-08-11) |
John Cowan scripsit:
> The bug is still present in 4.1.1, and it's drastic:
>
> #;1> (abs 2147483648)
> -2147483648
> #;2> (abs 2147483649)
> 2147483647
> #;3> (abs 2147483650)
> 2147483646
> #;4> (abs 4294967296)
> 0
Oops, dropped the following:
Almost certainly this is happening because the fixnum is being converted
to a C int, which is only 32 bits even on 64-bit architectures, at least
the ones we support (in the standard LP64 model, ints are 32-bit and
only longs and pointers are 64-bit).
A fixnum should always be converted to a C long, which will retain
full 63-bit precision on 64-bit systems and still provide 31 bits on
32-bit systems. It wouldn't surprise me if there were a fair number of
instances of this bug in various parts of Chicken's C code.
A question that remains is what to do when calling an existing C routine
that expects an int with a fixnum that is too big to fit in 32 bits.
Just truncating the fixnum doesn't seem like a Good Thing.
--
John Cowan address@hidden http://www.ccil.org/~cowan
Most languages are dramatically underdescribed, and at least one is
dramatically overdescribed. Still other languages are simultaneously
overdescribed and underdescribed. Welsh pertains to the third category.
--Alan King