@MathiasVP , I have some concerns in how I'm doing dominance checks, in that they are not optimized with any special annotations. I wonder if at scale this means dominance would be computed universally before joined with source/sinks.

@MathiasVP , I have some concerns in how I'm doing dominance checks, in that they are not optimized with any special annotations. I wonder if at scale this means dominance would be computed universally before joined with source/sinks.

Yes, I don't the approach you have here is how we should do it. With this approach we're:

1. Probably materializing both sinkStrictlyPostDominatesSource and sourceStrictlyDominatesSink (unless the compiler inlines them) - and they're both pretty large predicates.
2. Generating a lot of source/sink pairs before restricting it to the sinks that dominate the source. This is actually what we're seeing with the changes to the .expected file: dataflow is generating more tuples in the edges and nodes relations, which we're then excluding in the select clause with the domination/post-domination condition.

If you don't mind, I'll push a commit with some of the changes I think is necessary?

I've opened microsoft#39 for you to take a look at.

I've opened microsoft#39 for you to take a look at.

I've accepted this. Let's role with it!

Note. I added missing deallocation functions as part of this PR just now, but for some reason it is showing the entire file changed rather than a few of the definitions for deallocation were added to.