#70 Patch Proposition ti Fix #1206 ticket

[Denis HUGONNARD-ROCHE]

hi Simon

here is a first past proposed .
To fix the limits i have simply test when gnucobol does not produce usable number in float long format as it is the largest limits we have
Here is the program i used

       IDENTIFICATION DIVISION.
       PROGRAM-ID . stuff .
      *>
       ENVIRONMENT DIVISION.
       DATA DIVISION .
       WORKING-STORAGE  SECTION.

       01 B               FLOAT-LONG      .

      *>
       PROCEDURE DIVISION .
      *>------------------
      *>

      *    KO means that the result in floatingg point is not meaningful
      *    ie the floating point result is 0
      *
      *

      *====================== EXP   ============================

      *    OK
           display 'EXP 709' .
           MOVE FUNCTION EXP(709) TO  B .
           display b.
           display '------------------------------------------------' .
      *    OK
           display 'EXP 710' .
           MOVE FUNCTION EXP(710) TO  B .
           display b.
           display '------------------------------------------------' .
      *    OK
           display 'EXP -744' .
           MOVE FUNCTION EXP(-744) TO  B .
           display b.
           display '------------------------------------------------' .
      *    OK
           display 'EXP -745' .
           MOVE FUNCTION EXP(-745) TO  B .
           display b.
           display '------------------------------------------------' .
      *====================== EXP10 ============================
      *    OK
           display 'EXP10 308' .
           MOVE FUNCTION EXP10(308) TO  B .
           display b.
           display '------------------------------------------------' .
      *    KO
           display 'EXP10 309' .
           MOVE FUNCTION EXP10(309) TO  B .
           display b.
           display '------------------------------------------------' .



      *    OK
           display 'EXP10 -323' .
           MOVE FUNCTION EXP10(-323) TO  B .
           display b.
           display '------------------------------------------------' .



      *    KO
           display 'EXP10 -324' .
           MOVE FUNCTION EXP10(-324) TO  B .
           display b.
           display '------------------------------------------------' .

      *====================== FACTORIAL ============================

      *    OK
           display 'FACT 170' .
           MOVE FUNCTION FACTORIAL(170) TO  B .
           display b.
           display '------------------------------------------------' .
      *    OK
           display 'FACT 171' .
           MOVE FUNCTION FACTORIAL(171) TO  B .
           display b.
           display '------------------------------------------------' .

[Simon Sobisch]

Thanks for checking the numbers and the patch.
please adjust it so:

• exception EC-ARGUMENT-FUNCTION is raised
• no NaN is set
• for "too big" set COB_DECIMAL_INF
• for "too small" numbers set the srcval/mpf to the min you've found working "+/- 1"

Can you add an entry for libcob/Changelog and tests/testsuite.src/run_misc.at with your test code into your patch as well? Then we can easily see how portable the given tests are.

[Simon Sobisch]

.... Side note: we do need to check COB_DECIMAL_INF in all places where we currently check NaN and likely return INF in most places (and/or raise an exception or return 0); you may add that in the same patch

[Denis HUGONNARD-ROCHE]

Simon
Here's the fix for exp, exp 10, and factorial. I've also included other tests for these functions, unrelated to the fix, to improve code coverage.