Data flow: Compute local big step relation as stage output

[hvitved]

This PR moves the LocalFlowBigStep module from between stage 2 and stage 3 into the stage module, calculated as an output of the stage. The LocalFlowBigStep module is parameterized by an existing step relation, and we invoke the module in Stage3Param (as before this PR), and in Stage6Param, using the big step relation from Stage3Param as input.

The benefit of doing this is that we can sometimes skip intermediate nodes, resulting in smaller edges relations (as seen in the changes to the expected test output).

[geoffw0]

CPP DCA LGTM; two results lost, but I've seen these two wobble a few times lately. 👍

Swift DCA LGTM; just some extractor wobble. Analysis time change likely to be wobble as well, I can't rule out a real slowdown, but I trust you've done more accurate performance checks elsewhere. 👍

[aschackmull]

What's going on here? To no longer parameterise over LocalCc in the stages seems like a non-trivial change that's unrelated to the PR description. Also, getLocalCc is expected to be functional, which now no longer holds.

[hvitved]

I have reverted this change.

[aschackmull]

A change like this would be nice to perf-check on its own.

[hvitved]

I have reverted this change.

[aschackmull]

This here is why it's problematic to remove the parameterisation on LocalCc. It's possible that we want to always use call-context sensitivity + local call context sensitivity, but if so, then that should be investigated on its own.

[hvitved]

I have reverted this change.

[aschackmull]

The benefit of doing this is that we can sometimes skip intermediate nodes, resulting in smaller edges relations (as seen in the changes to the expected test output).

Is there a motivating example where this results in a significant reduction? Otherwise, my gut inclination is to not do this. What's the DIL size cost?

[hvitved]

I have updated this PR so we now only recompute the big-step relation in stage 6 (using the big step relation from stage 3 as input).

[aschackmull]

A few variable renamings that I think makes the code a bit clearer.

Suggested change

	exists(NodeEx node0, boolean preservesValue0, DataFlowType t0, string label0, Ap ap |
	Input::localStep(node0, state, node2, state, preservesValue0, t0, cc, label0) and
	revFlow(node2, pragma[only_bind_into](state), pragma[only_bind_into](ap)) and
	not outBarrier(node0, state) and
	(preservesValue = true or ap instanceof ApNil)
	|
	node1 = node0 and
	localFlowEntry(node1, pragma[only_bind_into](state), pragma[only_bind_into](ap)) and
	preservesValue = preservesValue0 and
	label = label0 and
	t = t0 and
	node1 != node2
	or
	exists(boolean preservesValue1, DataFlowType t1, string label1 |
	localFlowStepPlus(node1, pragma[only_bind_into](state), node0, preservesValue1, t1,
	cc, label1) and
	not node0 instanceof FlowCheckNode and
	preservesValue = preservesValue0.booleanAnd(preservesValue1) and
	label = mergeLabels(label1, label0) and
	if preservesValue0 = true then t = t1 else t = t0
	exists(NodeEx mid, boolean preservesValue2, DataFlowType t2, string label2, Ap ap |
	Input::localStep(mid, state, node2, state, preservesValue2, t2, cc, label2) and
	revFlow(node2, pragma[only_bind_into](state), pragma[only_bind_into](ap)) and
	not outBarrier(mid, state) and
	(preservesValue = true or ap instanceof ApNil)
	|
	node1 = mid and
	localFlowEntry(node1, pragma[only_bind_into](state), pragma[only_bind_into](ap)) and
	preservesValue = preservesValue2 and
	label = label2 and
	t = t2 and
	node1 != node2
	or
	exists(boolean preservesValue1, DataFlowType t1, string label1 |
	localFlowStepPlus(node1, pragma[only_bind_into](state), mid, preservesValue1, t1,
	cc, label1) and
	not mid instanceof FlowCheckNode and
	preservesValue = preservesValue2.booleanAnd(preservesValue1) and
	label = mergeLabels(label1, label2) and
	if preservesValue2 = true then t = t1 else t = t2

[aschackmull]

One minor comment, otherwise LGTM.