Enterprise Cobol V6 - Still problems after PH35652 / PH36690 and trace program compiled with OFFSET

After applying fixes PH35652 (COBOL 6.3) and PH36690 (COBOL 6.2), we notice that there are still problems to trace the execution of programs compiled with the option OFFSET / OPTIMIZE (1|2).

Despite the use of the OPTIMIZE (1) option, it happens that a COBOL source statement produces several blocks of binary instructions:
- a first block which corresponds to the start of the binary code of the COBOL instruction,
- another block (perhaps be several) rejected after the translation of an other COBOL statement.
Each block is identified in the OFFSET table.

Our trace tool works from the compilation listing, and uses either the table of OFFSETs, or all of the binary code produced by the LIST option.
When a program is compiled with the LIST option (and OPTIMIZE (1)), the step-by-step execution works fine (at this time we have not encountered any problems).
When a program is compiled with the OFFSET option (and OPTIMIZE (1)), in some cases the instructions are executed in an unexpected order, with back to instructions already executed.

The OFFSET table should differentiate the binary instructions blocks between those which correspond to the start of a COBOL statement and those which do not correspond to the start of a COBOL statement, for example by adding an annotation in the form of a specific character either on the source line number or on the block memory address.
Thus the debugging tools responsible for making step-by-step traces or setting breakpoints could select the blocks of binary instructions corresponding to the start of a COBOL statement.

If a COBOL statement is duplicated due to the activation of the INLINE option, this will result in several blocks associated with the start of the COBOL statement and several unassociated blocks at the start of the COBOL statement.

The OFFSET table must keep the identification of the blocks of instructions which do not correspond to a start of a COBOL statement because this is particularly useful in the analysis of memory dumps.