Re: [Mingw-users] sizeof long

[Joerg B. <jo...@sq...>]

Hi !

Lloyd Dupont wrote:
> 
> ok, if i well understand char, short, long have a size independant
> of thet processor
> whire are, respectively, 1, 2, 4.
> and int is the size of a register. 4 on 486 machine an even 8 on a
> 64 bits machine (so sinzeof(int) > sizeof(long) for such a machine)

I do not know the exact location(s) in the C standard(s), 
but I am quite sure this is wrong:

On byte-addressable machines, "char" is 1 byte. 
  (AFAIK, it is even defined that on _all_ machines 
  sizeof ( char ) == 1  .)

On all machines, the C standard(s) require 
  sizeof ( short ) <= sizeof ( int ) <= sizeof ( long ) 
and corresponding (take "(=" to read "is a subset of" / "is contained
in") 
  (set of all "short" values) (= (set of all "int" values) 
     (= (set of all "long" values) 
(the same for the "unsigned" variants).

Now these two properties would be violated by your assumption. 
If   sizeof(long) < sizeof(int)   ever were true, you could have 
"int" values which do not fit into a "long" variable, this might 
break several programs IMHO.

AFAIK, typically                 sizeof ( short ) == 2 
and (on 16- or 32-bit hardware)  sizeof ( long )  == 4 
    on 16-bit-hardware           sizeof ( int )   == 2 
    whereas on 32-bit-hardware   sizeof ( int )   == 4 

On 64-bit-hardware, there may be several cases: 
a)  sizeof ( int ) == 4       sizeof ( long ) == 8 
    if the system was designed as 64-bit from the start 
    (like the Alpha by Digital, now Compaq) 

b)  sizeof ( int ) == 4     sizeof ( long long ) == 8 
       (at least IBM's POWER had that new type "long long" 
       in the C compilers for AIX versions 4.1 and 4.2, 
       to overcome the 2 GB file size limit even on 32-bit HW) 
    sizeof ( long ) == 4  ??? 
    if binary compatibility with older hw versions is an issue 
    (IBM's POWER going from 32-bit to 64-bit, same may hold for HP)
(Disclaimer: I am not sure about "long" here, please check!)

Regards, 
Joerg Bruehe

-- 
Joerg Bruehe, SQL Datenbanksysteme GmbH, Berlin, Germany
     (speaking only for himself)
mailto: jo...@sq...