Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!mips!apple!agate!ucbvax!DOSUNI1.BITNET!ULAMMERS
From: ULAMMERS@DOSUNI1.BITNET
Newsgroups: comp.sys.atari.st
Subject: (none)
Message-ID: <9003030804.AA24624@ucbvax.Berkeley.EDU>
Date: 2 Mar 90 18:29:31 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 65

Date: 02 March 1990, 18:25:30 GMT
From: ULAMMERS at DOSUNI1
To:   INFO-ATARI16@WSMR-SIMTEL20.ARMY.MIL
Subject: Re: Turbo C v1.0 bug? -- NO!

In article <870@tnosoes.UUCP> mcsun!hp4nl!tnosoes!joep@uunet.uu.net
(Joep Mathijssen) writes:

>||   #include <stdio.h>
>||
>||   long tab[9000];
>||
>||   main()
>||   {
>||       int i;
>||       long j;
>||
>||       i = 0x2000;      /* 0x2000 = 8192 */
>||       j = 0x2000L;
>||       fprintf(stderr, "%lx, %lx\n", &(tab[i]), &(tab[j]));
>||   }
>||
>||
>
>RESULT:
>
>||   561b2, 661b2
>
>Can anybody explain the difference in the result when using an 'int' or
>'long int' as array-index?
>
>Joep,
>
>PS: don't tell me this is a feature!

Ok, I'm not going to tell you this is a feature but it isn't neither
a bug. The keypoint is, that in order to access an element of a vector
larger than 32K (this is true for tab) you  m u s t  always use
a long index. For the program above Turbo-C generates something like this:
...
  lea       tab,a2      ; a2 contains vector base address
  move.w    #$2000,d3   ; i=0x2000;
  move.l    #$2000,d4   ; j=0x2000L;
  move.l    a2,a0
  move.l    d4,d0
  lsl.l     #2,d0       ; muliply d0 by 4 == sizeof(long)
  add.l     d0,a0       ; a0=&tab[j]
  move.l    a0,-(sp)    ; push a0 for printf
  move.l    a2,a0
  move.w    d3,d0
  lsl.w     #2,d0       ; some as above, but now: bits 14 and 15 are lost
                        ; this means d0.w will be 0 now.
  add.w     a0,d0       ; a0=&tab[0] !!!!!
  move.l    a0,-(sp)    ; push a0 for printf
...
So, the calculation of &tab[i] failed because 0x2000*4 > 32K and this is
too large for an int.
BTW, the information to use long indices in conjunction with large vectors
is already included in the README-file that comes along with the compiler.
+++++++++++++++++++++++++++++++++++++++++++++++++++
+ Uwe Lammers                                     +
+ Univerity of Osnabrueck, Departement of Physics +
+ Email: ULAMMERS@DOSUNI1.BITNET                  +
+ Send flames to /dev/null                        +
+++++++++++++++++++++++++++++++++++++++++++++++++++