Re: [Chicken-users] Re: Problem w/ Utf8 egg: couldn't load case-map-1.dat

[Alex Shinn]

At Wed, 7 Sep 2005 10:06:51 -0700, Kon Lovett wrote:
> 
> The exception is raised in  (read-block! ...): (u32vector-set! vec i 
> (read-binary-uint32 port))
> (read-binary-uint32 ...) is returning a negative #, (u32vector-set! 
> ...) only accepts positive #

The bug is in arithmetic-shift on 32-bit platforms:

  (arithmetic-shift 181 24)
  => -1258291200.0

On my x86-64 machine it gives the correct answer:

  (arithmetic-shift 181 24)
  => 3036676096

It seems if both arguments are fixnums and the second is positive,
arithmetic-shift translates to C_fixnum_shift_left, which is

#define C_fixnum_shift_left(n1, n2)     (C_fix(C_unfix(n1) << C_unfix(n2)))

This causes the overflow into negative.

----

Note the case-map library was originally written for Gauche, and it's
a headache for me to maintain it under two separate platforms.  I'm
considering leaving it out of the next utf8 egg and moving it to the
common-scheme network.  Is anyone else currently using Chicken's
utf8 case-map or char-set?

I'm also considering breaking the utf8 module into 3 modules:

  utf8
  utf8-srfi-13
  utf8-srfi-14

which will speed compile and load times (it's a pretty huge module).
This is also because I plan to write a string-cursor based alternative
to srfi-13 - string indexes are not a good way to work with utf8
strings.  Does anyone have any thoughts on this?

-- 
Alex