Type inference: Filter away psuedo types in getTypeMentionRoot

[hvitved]

This PR fixes an issue uncovered by unified DCA nightly for the project soto-project/soto where we mistakenly matched TypeMentions that cannot resolve (in that case they are interpreted as "resolving" to UnknownType).

DCA cannot confirm that soto-project/soto now works because it already fails to analyze on the baseline, but I have verified that it works locally. It does show a significant speedup on signalapp/Signal-iOS though.

DCA for rust also shows a speedup.

[Copilot]

Copilot review overview

🟡 Changes recommended

The behavioral change lacks regression coverage, and its documentation is now inaccurate.

Review effort: Balanced
Findings: 2

Open (2)

• Document that pseudo-rooted mentions are excluded · New
• Add regression test for excluding pseudo-rooted type mentions · New

What changed in this PR

Filters pseudo types from root type mentions during shared type inference.

Changes:

• Excludes PseudoType roots.
• Limits instantiation candidates to transitive constraints.

File	Description
shared/​typeinference/​codeql/​typeinference/​internal/​TypeInference.qll	Refines root-type and constraint matching.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[hvitved]

An unrelated small piece of manual magic; I suspect this is what causes the marginal Rust speedup.

[paldepind]

we mistakenly matched TypeMentions that cannot resolve (in that case they are interpreted as "resolving" to UnknownType).

Why do we not address that directly? I.e., make sure that no type mentions resolve to pseudo types to begin with, instead of filtering them out after the fact?

[hvitved]

Why do we not address that directly? I.e., make sure that no type mentions resolve to pseudo types to begin with, instead of filtering them out after the fact?

Because we want to allow for cases where type annotated entities are allowed to have their types inferred from the context whenever the annotation cannot be resolved (for whatever reason). For example in

let x : Vec<Unresolved> = Vec::new();
x.push(foo());

we will allow for the element type to be inferred from the return type of foo (perhaps Unresolved can actually be resolved wherever foo is defined). I.e., we effectively treat unresolvable type mentions as _.

[paldepind]

Thanks, that makes sense. I think it's worth adding that explanation in a comment before the added not result instanceof PseudoType as I don't think it's immediately clear why a type mention would have pseudo types to begin with.

[hvitved]

Thanks, that makes sense. I think it's worth adding that explanation in a comment before the added not result instanceof PseudoType as I don't think it's immediately clear why a type mention would have pseudo types to begin with.

Good idea, done.

[paldepind]

LGTM and I think the final version is much clearer and explicit 👍