DFA improvements #1807
Description
Following #1783, this issue regroups all identified improvements for DFA.
- The
WithDfamode for CFG building is the same as theExtendedmode, the only difference is on the generic arguments for the block type. We could decouple CFG building from DFA by removing the need for typing the block data (i.e. stored asobject, DFA builder would have to cast when reading) - The current implementation does not handle child variable writes correctly. In Cobol writing into a child variable means that all parent variables are modified. This is partially handled in
ZCallPgmReportfor levels 88 in CALLs but it should be generalized.- The class
DefaultDataFlowGraphBuilderis responsible for implementing DFA building specialized for TypeCobol/Cobol language so this should be handled there.
- The class
- Extend
ValueOriginclass to handle more instructions. - general remarks
- restrict visibilities where possible to avoid misuses
- remove unused code, factorize where possible
- Some reports collect results during the Node phase, others may work on a
CompilationUnitafter parsing has completed so it is confusing on how to use theIReportinterface. Maybe create 2 sorts of reports (one is tied to the parsing process, the other directly uses the results) ?- In fact this is related to the concept of analyzer, we need to have at least 2 kinds of analyzer : one for the node building phase, one other after the CrossCheck. Each report would have to choose which kind of analyzer it uses.
Improve DFA results with correct CSV-compliant escaping for instruction text