fix(resolution): a same-name re-binding is not a local definition

[danusha2345]

Follow-up to #1759 (the landing of #1735). Fixes a regression that is on main now: object-literal-methods.test.ts and ui-steps-api.test.ts fail at cd4e65b (3 tests), because the local-binding rule reads a store-action re-binding as a local definition.

What

const { fetchUser } = useStore.getState() and const setZipUri = useStore((s) => s.setZipUri) are how a store action reaches its caller, and the store-action resolution follows exactly those shapes. Rule 2 of #1759 ("a name the file binds itself has no cross-file candidate") counted both as definitions and dropped loginFlow → fetchUser and handleZipComplete → setZipUri — edges the graph is built to hold.

A declaration whose pattern or initializer names the same member is a re-binding of something defined elsewhere, not a definition: the graph's symbol is what the bare call means. isLocallyBoundJsName now skips a destructuring pattern altogether and a plain const x = … whose initializer mentions x (s.setZipUri, options.now). const transform = makeTransform(), a parameter, a function/class declaration and an arrow bound to a fresh name still shadow.

The fixture's const now = options.now || (() => Date.now()) case is dropped from the "locally bound" test: by this rule it is a re-binding, and the graph's cross-file now is what it meant before #1759 too. The new test pins the store shapes.

Measured

vitejs/vite 8492422, wasm arm, edge rows keyed with resolvedBy. On main at 2f1a99d (pre-#1759) the two rules with this fix: LOST 709 / GAINED 137 — the removals are the #1714 shape (import(…) → runner.ts::import 264, resolve(…)/transform(…) → PluginContainer 217, a fixture's test 91, …); 86 of the gains are resolve bound by import { resolve } from 'node:path' landing on a nested const resolve in config.ts, which #1715's bare-import guard declines. Stacked on #1715: LOST 515 / GAINED 56 (41 are log(…) moving onto event.d.ts's declared global).

Tests

bare-call-no-method 6/6 (one new: a hook selector and a getState() destructure keep their edges), object-literal-methods and ui-steps-api pass again, resolution 198/198, cross-file-visibility, frameworks-integration; tsc clean. Full suite on Linux: only the viewer-less codegraph ui / ui-package files fail here, same as on main.

🤖 Generated with Claude Code

[danusha2345]

Second commit, a29d112: the exemption was too wide. On a Kotlin + JS + Go app, const now = opts.now || Date.now read as a re-binding and 24 now() calls in the renderer landed on a Kotlin test's private val now — that alias IS a local binding, nothing in the graph is what the call means. A re-binding is now only a destructuring pattern or a selector arrow that picks the member off its own parameter (useStore((s) => s.setZipUri)); () => Date.now() is not one. The options.now fixture is back in the test. That app: −509 / +13 against main, the 13 all rcjoystick.js onto stick-curve.js (real). The three store-action tests still pass.

[danusha2345]

Merged current main (3ed73bc) into this branch: head 2b126b6, no conflicts. On that merge the two ui-steps-api tests and the loginFlow → fetchUser assertion of object-literal-methods go green (the three #1759 regressions this PR is for). What is left red in that file is a different assertion — hardReset → reset through useStore.getState().reset() — and it is red on plain main since de5adba (#1790 drops the identifier-rooted member call), see #1794. This PR does not touch that path.