Shared: Prepare model generation for C++ adoption

[MathiasVP]

This PR slightly generalizes the model generation input signature to allow for a smooth C++ adoption:

• 732fcbf moves the asParameter out of the NodeExtended signature. This is necessary because the return type of C++'s Node.asParameter is different from the actual parameter nodes we use inside dataflow.
• Likewise, c484945 moves the getEnclosingCallable out of the NodeExtended for similar reasons. I also moved getAsExprEnclosingCallable for symmetry.
• ea3bb8c adds a hook to customize the ReturnValue string for ValueReturnKinds. This necessary because C++ has ReturnValue, ReturnValue[*], ReturnValue[**], etc, for returning p, *p, **p from a function.

Additionally, I fixed a potential cartesian product in a1dc874 that I was hitting when testing MaD on a large C++ repository internally at Microsoft.

Commit-by-commit review recommended.

@michaelnebel does this need additional testing other than CI? And if so, how do I go about doing that?

[michaelnebel]

does this need additional testing other than CI? And if so, how do I go about doing that?

Yes, we should probably use the mad-queries suites in DCA for Java and C# to check performance (and stability of constructed models). These also print some model diff output as well as measure the performance. For technical reasons the DCA experiment needs the query suite both as the query suite being executed but also as "uninterpreted" queries.

If you want to add a similar DCA suite for C++ we might need to make some changes to DCA to support the model generator related summaries for C++?

I will start some experiments for C# (for .NET Runtime) and Java (SDK). 😄

[MathiasVP]

If you want to add a similar DCA suite for C++ we might need to make some changes to DCA to support the model generator related summaries for C++?

Uh, yeah that sounds good. We can maybe discuss that once we've got this merged and #19274 (at which point I can actually open a PR that implements ModelGeneratorInputSig.

[michaelnebel]

Fantastic!

[michaelnebel]

Uh, nice catch!

[michaelnebel]

@MathiasVP : I ran the modelgenerator on dotnet/runtime for C# and on jdk17 for Java and it works fine; Don't let the DCA timeout block merging this PR.

[MathiasVP]

@MathiasVP : I ran the modelgenerator on dotnet/runtime for C# and on jdk17 for Java and it works fine; Don't let the DCA timeout block merging this PR.

Thanks for checking 🙇 I'll go ahead and hit the Big Green Button then 🚀