Fix Overriding Immutable Variables

[olabusayoT]

• since we can no longer override immutable variables in Scala 3, we change Ustate's dataOutputStream field to be a private mutable variable with a setter and getter method

DAFFODIL-2975

[stevedlawrence]

+1, just some comment about a different approach. I have no strong feeling about which approach to take so this is fine with me

[stevedlawrence]

This approach with getters/setters actually makes for a possible enhancement. I think that UStateForSuspension should never change the dataOutputStream. If that's true, you could change _dataOutputStream to protected, and then this this could be changed to something like

_dataOutputStream  = dataOutputSTream

And UStateForSuspension could overwrite setDataOutputStream to be something like

override def setDataOutputStream(...) = {
  Assert.invariantFailed("Should never change dataOutputStream on a suspension")
}

This effecively forces the dataOutputStream field to be unchanging for suspensions.

I don't think we necessarily need to do this as part of this PR (I'm not even sure it suspension DOSs never change), but this is a potential advantage for these kinds of setters/getters.

[olabusayoT]

@mbeckerle , can you confirm that UStateForSuspension shouldn't change the dataOutputStream?

[mbeckerle]

I believe this is correct - once you have created a UStateForSuspension, lots of things are frozen there.

That said, the interplay and way output streams do this gradual double-buffering and collapsing of the buffers is quite tricky. It's the most complex part of the Daffodil runtime, so .... do I really know that there's no exception to this rule? No. We should try it.

[stevedlawrence]

Note that we don't necessarily need these getters/setters. We could make it so dataOutputStream is still public and initialize it to _, and then instead of overriding it, UStateForMain constructor would look like

dataOutputStream = DirectOrBufferedDataOutputStream(...)

and UStateForSuspension would look like

class UStateForSuspension(
  ...,
  suspensionDataOutputStream: DirectOrBufferedDataOutputStream,
  ...) {

  dataOutputStream = suspensionDataOutputSTream
}

That would avoid the needs for getters/setters and would make this a much smaller change. Though it does make it a but unsafer since it requires implementations to initialize it rather than the compiler forcing it to be overridden and initialized, but it's not a big deal for us since we only have these two classes.

[mbeckerle]

+1
I'm ok with this change. But we should definitely do the override of the setter behavior in the IO streams model to see if this no-writing the output data stream from UStateForSuspension in fact works.