[hvitved]

This PR starts treating constructors like ordinary instance methods without return types, in order to be able to handle inheritance in the same way as for ordinary methods.

class A<Ta> {
  init(x : Ta) { ... }
}

class B<Tb> : A<Tb?> { }

A.init(0);        // (1) has type `A<Int>`
B.init(.some(0)); // (2) has type `B<Int>`
B<Int>(.none);    // (3) has type `B<Int>`

Because the constructor calls are treated as normal instance method calls, we use the type qualifier to represent the receiver, and then allow for contextual type information to flow into the receiver:

• (1) The receiver has type A, Ta is inferred to be instantiated to Int, which is propagated directly back into the receiver.
• (2) The receiver has type B, Ta is inferred to be instantiated to Int? which in turn means Tb is inferred to be instantiated to Int, which is then propagated back into the receiver.
• (3) The receiver has type B<Int>, and no further type inference is needed.

DCA shows a small increase in resolution rates.

Commit-by-commit review is encouraged (third commit is best reviewed ignoring whitespace changes).

[Copilot]

🟡 Changes recommended

Constructor inference loses contextual argument types and can produce incorrect inherited type mappings and superclass results.

4 open issues

• Restore reverse contextual type propagation from calls to receivers · New
• Fix subtype ordering in root type satisfaction lookup · New
• Resolve non-identifier qualifier roots before receiver inference · New
• Disambiguate inherited constructor names in test expectations · New

What changed in this PR

Improves constructor-call type inference in Unified, particularly for inherited Swift initializers.

Changes:

• Models constructors using receiver types.
• Extends shared contextual inference support.
• Adds inherited-constructor tests and updates expectations.

File	Description
unified/​ql/​test/​library-tests/​type-inference/​type-inference.expected	Updates generated inference expectations.
unified/​ql/​test/​library-tests/​type-inference/​lub.swift	Updates an inline expected initializer target.
unified/​ql/​test/​library-tests/​type-inference/​constructors.swift	Adds inherited and generic constructor cases.
unified/​ql/​lib/​codeql/​unified/​internal/​typeinference/​TypeMention.qll	Excludes enum cases from resolved types.
unified/​ql/​lib/​codeql/​unified/​internal/​typeinference/​TypeInference.qll	Adds receiver-based constructor inference.
unified/​ql/​lib/​codeql/​unified/​internal/​ExprPositions.qll	Excludes enum cases from type contexts.
shared/​typeinference/​codeql/​typeinference/​internal/​TypeInference.qll	Extracts reusable contextual typing logic and adds an inference hook.

_{Review effort: Max}

[Copilot]

🟡 Changes recommended

Unresolved type-resolution and contextual-inference defects affect correctness.

5 open issues

• Preserve enum type for unqualified enum-case construction · New
• Infer contextual holes from constructed receiver types · New
• Recognize enum constructors in generic contextual inference · New
• Resolve associated-type variants by accessing class · New
• Swap incorrect initializer annotation targets · New

4 resolved since last review

• Disambiguate inherited constructor names in test expectations
• Resolve non-identifier qualifier roots before receiver inference
• Fix subtype ordering in root type satisfaction lookup
• Restore reverse contextual type propagation from calls to receivers

_{Review effort: Max}

[Copilot]

🔵 Needs a closer look

Shared recursive inference changes have unresolved correctness concerns and need cross-language validation.

3 open issues

• Exclude explicitly typed paths from inherited constructor inference · New
• Preserve contextual inference for unresolved generic results · New
• Fix missing contextual result path to T10b · New

5 resolved since last review

• Swap incorrect initializer annotation targets
• Resolve associated-type variants by accessing class
• Recognize enum constructors in generic contextual inference
• Infer contextual holes from constructed receiver types
• Preserve enum type for unqualified enum-case construction

_{Review effort: Max}

[asgerf]

The changes in unified look reasonable to me. I'll defer to @paldepind to approve after reviewing the type inference library changes.

[paldepind]

LGTM 🎉