[proggeramlug]

Summary

infer_type_from_expr mis-typed a logical A && B / A || B as A's type whenever B inferred to Any. But && yields B's value when A is truthy, so the result can be the (unknown) right value — A's type is wrong. This advances the #3527 Express compile past object-inspect.

Root cause

A textbook feature-detect idiom:

var hasMap = typeof Map === "function" && Map.prototype;  // inferred Boolean (!)

hasMap holds an object at runtime but was typed Boolean (the left operand's type). Codegen then treated it as a boolean in object-inspect's later chain:

var mapSize = hasMap && d && typeof d.get === "function" ? d.get : null;

miscompiling the short-circuit and dereferencing null → TypeError: Cannot read properties of null (reading 'get') during init while compiling Express. (Minimal repro: var hm = typeof Map === "function" && Map.prototype; hm && null && (null).get throws; the same chain with hm typed any does not.)

Fix

In lower_types.rs, for LogicalAnd | LogicalOr: when the right operand is Any, the result is Any (the value could be that right operand), not the left operand's type. The concrete-right heuristic (x && knownTyped → knownTyped) is unchanged. This only makes the inferred type more dynamic — always codegen-safe, never less.

Verification

• object-inspect's init now works; the Express tree advances past it.
• New gap test test_gap_logical_and_value_type.ts (the hasMap idiom, &&/|| returning the right value, chained property guards) — byte-identical to Node.
• cargo test -p perry-hir green. Pre-existing failures (bind_replace, async_hooks/perf_hooks timestamp nondeterminism) verified identical on baseline — not caused by this change. cargo fmt --check clean.

Part of the #3527 chain (prior: #3537/#3543/#3546, #3551, #3733, #3751, #3764). No version bump / changelog — folded in at merge per the worktree-PR convention.