[colbymchenry]

Summary

A call or new that a strategy resolved by name or through an import could land on a TypeScript interface or type alias. Nothing can call one, or make an instance of one.

• memfs's fs-core result.ts declares interface Ok<T> beside function Ok<T>(), and the same for Err. Ok(…) through an import reached the interface.
• A new of a global that a project only declares a type for was linked to that type by name: new Response() to insomnia's Postman interface Response, new Worker() to vite's .d.ts interface, new Event() to jest-circus's type Event. jest's write(), a parameter, was linked to a type write.

This is the half of #2800 that doesn't overlap #2754, split out as asked. #2800 stays open with the default-export half.

Root cause

The resolver has no rule on what a call may target. Whichever strategy found the name (an import, then name matching) picked among same-named nodes by order, and nothing stopped it on a type.

Fix

gateTargetKind (src/resolution/index.ts) now handles JS-family calls and instantiates results whose target is an interface or a type_alias:

• The result moves to the value the target's file declares under the same name: a function, class, component, constant or variable, the first by position. That covers function Ok beside interface Ok, and a class merged with an interface.
• With no such value, the result is dropped and the reference stays unresolved, so new Response() links nothing.

Type positions (references) are untouched: r: Ok<number> still names the interface.

Validation

Tests: __tests__/call-value-targets.test.ts indexes real temp repositories (3 tests).

• memfs's Ok/Err beside their interfaces, called directly and through a barrel; a type position still names the interface.
• new Response(), new Event() and new Worker() with only project types of those names, plus a write() parameter call beside a type write: nothing links.
• new Store() reaches the class an interface merges with.

Results:

• RED on main: all 3 fail.
• GREEN on the kernel. The change is resolver-only; the kernel only extracts.
• Mutations: 5 rules removed one at a time, and each removal fails at least one test. The rules are: the gate, the value beside the type, new gated, a class beside an interface, and type aliases gated.
• 91 related test files pass (2,178 tests): import, default, sync, resolution, React Router, JSX, frameworks and others.

Sync equals a fresh index (CLI codegraph sync against a fresh init) in four cases:

• the value beside a type is added;
• the value beside a type is removed;
• a project type for a global appears;
• a type alias named like a parameter appears.

A/B: fresh init, both arms with the native kernel, site-paired edge diff, main e8ddc4e.

repo	result
memfs adae41a5	11 Ok(…)/Err(…) calls move from interface Ok/Err to the functions. new ReadableStream() no longer lands on a project interface.
insomnia 1a5a33aa	47 new Response(), new Request(), new Headers() and new Event() dropped; they had linked to Postman importer types
nuxt d296167e	57 new Response() dropped; it had linked to a type Response in a build stub
vite 4260c756	27 new Worker()/new Event() dropped; they had linked to .d.ts interfaces
webdriverio c6f450ee	26 new of globals dropped (Event, Response, Request, mocha's Context and Suite)
jest 61050e93	7 new Event(), new Module() and a new Reporter() of a local variable dropped, and write() on type write
vben bd7b04d2, outline 45bf9e29, social-app 00ff89e3, t3-turbo 8f945b7b, nitro da176d5f, vue3-antdv-admin 64c03db7, pathe bc7477a0, c12 c42af029	byte-identical
flask d086db85 (Python), gin 0f09c3a9 (Go): controls	byte-identical

I grouped every dropped new (165) by name and target. Each names a global (Worker, Event, Response, Request, Headers, ReadableStream, Node's Module, mocha's Context/Suite) or a local variable, never a class of the project. Each repo's removed set is the same as in #2800's A/B. #2800's other changes are absent here: jest's 23 Resolver edges, outline's and insomnia's guessed-function calls, webdriverio's allure calls. They stay as on main and belong to the held default-export half.

Cost: fresh index, median of 3, run back to back, alternating which arm goes first. All within noise:

repo	main	this PR
vite	9.07 s	8.74 s
nuxt	11.71 s	10.93 s
insomnia	15.26 s	15.11 s

The added work is one same-file lookup when a call's target is a type.

Full suite on fa70442 (CODEGRAPH_KERNEL_EXPECT=1, 4 workers): 725 files (719 passed, 6 skipped), 8,688 tests: 8,606 passed, 0 failed, 82 skipped. No file needed a rerun.

Not in this PR

• The default-export half (fix(js,ts): default imports follow a module's imported default; calls never land on types #2800), held for fix(js,ts): an import reaches what its module hands on, not a namesake or the first export #2754.
• function_ref and references in value positions (items.map(Ok)) aren't gated. None landed on an interface or type alias in these repos.

🤖 Generated with Claude Code