#478 float-decimal intermediate rounding and exceptions are not always correct

[Simon Sobisch]

This is a follow-up to [bugs:#470] which showed that computations with ON SIZE ERROR are handled different than computations without - but this shouldn't be the case.

In general this is what should be done in this case:

• use an intermediate decimal128 item for each part

• if the result of an intermediate item would be farther from zero than is allowed: -9 999 999 999 999 999 999 999 999 999 999 999E+6144 and +9 999 999 999 999 999 999 999 999 999 999 999E+6144- the EC-SIZE-OVERFLOW exception condition is set to exist and the content of the resultant identifier is unchanged.

• if the result of an intermediate item would be nearer to zero than is allowed: 1.0E-6176 and +1.0E-6175 - the INTERMEDIATE ROUNDING clause active should be checked:

  • PROHIBITED - EC-SIZE-TRUNCATION is set (further results undefined, I'd say: do rounding and further calculation)
  • all others or none specified - the intermediate value is set to the nearest possible zero value

• if the result of a receiving item would be nearer to zero than is allowed: 1.0E-6176 and +1.0E-6175
  --> ideally depending on a compiler configuration flag:
  --> configuration value1 (defined in COBOL 2014): the EC-SIZE-UNDERFLOW exception condition is set to exist and the content of the resultant identifier is unchanged
  --> configuration value2 (defined in COBOL 2002 which doesn't has EC-SIZE-UNDERFLOW): the receiving item is set to the nearest possible zero value

• if the receiving item is a FLOAT-DECIMAL-16 item the following additional rules exist

  • if the value is farther from zero than is allowed: -9.999 999 999 999 99E+384 and +9.9999 999 999 999 999E+384- the EC-SIZE-OVERFLOW exception condition is set to exist and the content of the resultant identifier is unchanged.
  • if the value is nearer to zero than is allowed: 1.0E-398 and +1.0E-398
    --> ideally depending on a compiler configuration flag:
    --> configuration value1 (defined in COBOL 2014): the EC-SIZE-UNDERFLOW exception condition is set to exist and the content of the resultant identifier is unchanged
    --> configuration value2 (defined in COBOL 2002 which doesn't has EC-SIZE-UNDERFLOW): the receiving item is set to the nearest possible zero value

If any of the mentioned exceptions is set the statements in the ON SIZE ERROR phrase should be executed.

[Simon Sobisch]

Considered to be fine in re-branch after Ron's changes, merged to trunk with [r2324]. We should add test cases for intermediate rounding options, but this can be done after the rc-1 (ideally before the final release).