FWD » Core Development » Base Language

Currently, each legacy class has a synthetic 'execute' instance method which includes initialization like instance-level buffer scopes (for temp-tables), ON ROUTINE-/BLOCK-LEVEL statements@, etc.

We can get rid of the BlockManager.externalProcedure call in the converted code, and emulate this directly in ControlFlowOps.initializeLegacyObject, only once, for the entire initialization of this instance.

The overhead comes from the fact that currently, each super-class has its own BlockManager.externalProcedure Ideally, considering that the synthetic 'execute' method can't contain application converted code, and will contain just initialization state, we can cut-down on the state being tracked when executing this (I'm thinking about ProcedureManager stacks).

Contantin, just a question here: I thought we open scopes in externalProcedure due to the fact that we are aware only at conversion which buffer scopes should be opened. How can we know at run-time which buffers to open? Do we use reflection and identify the global record buffers?

What I meant is to emit all the code (like buffer scopes) directly in the Java method, and do not enclose it in BlockManager.externalProcedure. All legacy object instantiation is done via ObjectOps APIs, so we can call the push/pop scope code there. So this will perform only one scope processing per instance, not one for each super-class, as it is done now.

Oh right, this makes sense - getting rid of the BlockManager API. This way we avoid doing ProcedureManager work.

Code Review Task Branch 8084a Revisions 14870 through 14873

The changes are good.

As an additional improvement, I wonder if we should change how BLOCK-LEVEL and ROUTINE-LEVEL statements work.

Instead of emitting onBlockLevel(Condition.ERROR, Action.THROW); or onRoutineLevel(Condition.ERROR, Action.THROW); into the execute() method, I think we should just emit a class-level annotation. It is really just a boolean flag anyway. This would be the case for both procedures and OO .cls files. The result would be much cleaner (code wise). I also wonder if it might be more efficient since some additional execute() methods will now be empty.

I also don't know why we emit the condition and action parameters since they cannot vary and really have no meaning. But if we get rid of these calls anyway, it would be even better.

Greg Shah wrote:

As an additional improvement, I wonder if we should change how BLOCK-LEVEL and ROUTINE-LEVEL statements work.

I agree, but this would take a little more work.

Can 8084a be committed as is?

Greg Shah wrote:

Yes, if you are complete with testing.

I haven't converted all apps, but I think the conversion rules are good.

If you've done enough, you can merge.

Branch 8084a was merged to trunk rev 14876 and archived.