Shared: Use sourceBoundedFastTC in TypeTracking

[MathiasVP]

We've seen quite high iteration counts for this predicate in C/C++ on some projects. I don't see why we shouldn't use a HOP here 🤷

In particular, this significantly reduces the analysis time on https://github.com/openvinotoolkit/oneDNN when running with --ram=6000 (which we saw in the 2.23.0 upgrade)

[Copilot]

Pull Request Overview

This PR optimizes the TypeTracking implementation by using sourceBoundedFastTC for transitive closure computation instead of manual recursion. The change aims to reduce high iteration counts observed in C/C++ projects by leveraging a Higher-Order Predicate (HOP) for better performance.

Key Changes

• Replaced manual recursion in standardFlowsTo with sourceBoundedFastTC transitive closure
• Introduced simpleLocalSmallStepPlus predicate using the optimized transitive closure
• Modified isLocalSourceNode predicate to support the new approach

[hvitved]

Isn't it better to cache simpleLocalSmallStepPlus, which will persist the underlying FastTC structure instead of the materialized relation? standardFlowsTo then needs to be non-cached and pragma[inline]ed. See also

codeql/ruby/ql/lib/codeql/ruby/typetracking/ApiGraphShared.qll

Line 120 in b4e6d30

predicate epsilonStar(ApiNode pred, ApiNode succ) = fastTC(epsilonEdge/2)(pred, succ)

for another example of this.

[MathiasVP]

Uh, that's a good idea. I didn't actually know that caching a fastTC didn't just cache the materialized relation. I've fixed this in bd7e20d

[aschackmull]

This is a small helper for standardFlowsTo. I think it should be moved down, so it appears directly before standardFlowsTo. Also, check whether the RA is affected by manually inlining this - if not, then that simplifies things. Gah, never mind, it's of course also input to the TC.

[MathiasVP]

:til: when you fix a Code Scanning warning the conversation just ... disappears? I've made standardFlowsTo private in f8bdf92 now. Thanks for explaining how it should be fixed, @aschackmull! I was planning on asking either you or the Foundations team about it at some point so I appreciate that you just told me an easy fix (which I guess I could've also spotted by looking at the ql-for-ql query 😅)