Unsigned integers

[jdebord]

It seems that neither the DOS version nor the Linux version of FB can handle the unsigned long integers correctly.

See for instance the example program litnum_ops:

Code: Select all

	if( 4294967295UL + 10UL = 9UL ) then
		print 4294967295UL + 10UL; " = "; 9UL
	end if
		
	if( 4294967295UL > 2147483648UL ) then
		print 4294967295UL; " > "; 2147483648UL
	end if
	
	if( (1! or 2!) = 3 ) then
		print 1! or 2!; " ="; 1! or 2!
	end if	
		
	if( (1! shl 2!) = 4 ) then
		print 1! shl 2!; " ="; 1! shl 2!
	end if
	
	sleep

This bug existed in some previous Windows version but it has been corrected. Perhaps the DOS/Linux versions were not updated accordingly?

[DrV]

I don't see the bug; with fbc 0.14, I get these results:

Code: Select all

C:\>fbc math.bas -target win32

C:\>math
9 = 9
4294967295 > 2147483648
 3 = 3
 4 = 4

C:\>fbc math.bas -target dos

C:\>math
9 = 9
4294967295 > 2147483648
 3 = 3
 4 = 4

Could you explain what the buggy behavior is?

[DrV]

Hmm, compiling with the DOS compiler (rather than cross-compiling from Win32 fbc) results in this:

Code: Select all

C:\>fbc math.bas

C:\>math
 3 = 3
 4 = 4

Which is wrong, I suppose. :) So it's a bug in the compiler somewhere, not in the generated code...

[jdebord]

Yes, the bug appears when using the compiler from the DOS or Linux versions. (I did not try to cross-compile with the Windows compiler).

It's a saturation problem: integers > 2147483648 are given the value 2147483648.

This problem was identified and solved in the Windows compiler. So it explains why the cross-compilation with the Windows compiler gives the correct results.

[v1ctor]

Fixed, changes are in CVS, atoi() won't handle unsigned int literals above 2^31 in DOS or Linux, that seems to be a M$-only feature.

[jdebord]

Thanks!

[jdebord]

Thanks!