Re: [open-cobol-list] Internal compiler error

[Frank Swarbrick]

Are you sure wrt to floating-point arithmetic?

       >>SOURCE FORMAT IS FREE
program-id. fract.
data division.
local-storage section.
77  double                              float-long.
77  decimal                             packed-decimal   pic s9(3)v99.
77  bin-decimal                         binary           pic s9(4)v99.
77  zoned-decimal                       display          pic 9(3)v99.
procedure division.
    move 1 to double decimal bin-decimal zoned-decimal
    divide 10 into double decimal bin-decimal zoned-decimal
    display double ' ' decimal ' ' bin-decimal ' ' zoned-decimal
    goback.

C:\Users\Frank\src\cobol>fract
0.10000000000000001 +000.10 +0000.10 000.10

(float-long and comp-2 having the same meaning)
Looks like floating point to me...  (I didn't analyze the generated C code).

As for 64-bit, you can push 64 bits on the stack.  You must do call BY VALUE, ie:

       >>SOURCE FORMAT IS FREE
program-id. pointer64.
data division.
working-storage section.
77  int64  comp-5 pic s9(18).
procedure division.
    display int64
    call static 'test64' using value int64
    goback.

This resolves to the following C code:

#define    cob_s64_t        __int64
#define    cob_s64_ptr        cob_s64_t *

static cob_u8_t b_5[8];    /* int64 */

      (*(int *) (b_1)) = test64 ((unsigned cob_s64_t)((*(cob_s64_ptr)(b_5))));

Resolving the #defines we have (I believe):
      (*(int *) (b_1)) = test64 ((unsigned __int64)((*(__int64 *)(b_5))));

But as for returning 64 bits, I believe you are correct that it can't be done with OC 1.1, even the version that Sergey posted recently.  The above probably needs to be:

      (*(__int64 *) (b_1)) = test64 ((unsigned cob_s64_t)((*(cob_s64_ptr)(b_5))));
or
      (*(cob_s64_t *) (b_1)) = test64 ((unsigned cob_s64_t)((*(cob_s64_ptr)(b_5))));

Frank


On 6/30/2013 9:23 PM, Charles Anthony wrote:

I am pretty sure that the issues that I couldn't solve in COBOL were related to passing doubles around. C wants to pass doubles by value pushing 64-bits on the stack, and allow 64-bit return values. I was unable to convince OpenCOBOL to push 64-bit values on the stack or deal with 65-bit return-codes. Also, I note that OpenCOBOL always does decimal math, even when the operands are comp-2. I don't know if this is a COBOL thing, or just something not handled because if you want to do that kind of math, go do some FORTRAN. Anyway, I had fun.