runtime: throw TypeError for new Symbol and new BigInt

[andrewtdiz]

Summary

Symbol and BigInt are callable factory/coercion functions in Node, but neither is constructable. Perry implements call lowering for both, yet new Symbol(...) and new BigInt(...) appear to fall through the generic built-in new placeholder path instead of throwing.

Node behavior

Observed with local Node v25.9.0:

console.log(typeof Symbol("x")); // "symbol"
console.log(typeof BigInt("42"), BigInt("42").toString()); // "bigint", "42"

try { new Symbol("x"); } catch (e) { console.log(e.name, e.message); }
// TypeError Symbol is not a constructor

try { new BigInt("1"); } catch (e) { console.log(e.name, e.message); }
// TypeError BigInt is not a constructor

Perry evidence

• crates/perry-hir/src/lower/expr_call/globals.rs lowers Symbol() / Symbol(description) to Expr::SymbolNew(...) and BigInt(value) to Expr::BigIntCoerce(...), so ordinary calls are covered.
• crates/perry-runtime/src/symbol.rs implements js_symbol_new_empty, js_symbol_new, js_symbol_for, and related runtime support. crates/perry-runtime/src/bigint.rs implements the BigInt allocation/coercion runtime.
• crates/perry-runtime/src/object/global_this.rs::GLOBAL_THIS_BUILTIN_CONSTRUCTORS and crates/perry-codegen/src/expr/helpers.rs::is_global_this_builtin_name() include Symbol and BigInt as function-valued globals.
• crates/perry-hir/src/lower/expr_new.rs has no special class_name == "Symbol" or class_name == "BigInt" rejection path.
• crates/perry-codegen/src/lower_call/new.rs::lower_new() calls lower_builtin_new() first; there is no Symbol / BigInt arm there. For built-in/native class names without a dedicated handler and no user class, it lowers args for side effects and allocates an empty object placeholder via js_object_alloc(0, 0). That means new Symbol("x") / new BigInt("1") can produce an ordinary object instead of throwing.

Expected fix direction

Add a new-expression rejection path for these non-constructable built-ins so both direct and obvious rebound constructor forms throw a TypeError rather than allocating placeholders. Ordinary Symbol(...), Symbol.for(...), Symbol.keyFor(...), and BigInt(...) call behavior should remain unchanged.

Related but not duplicate: #2852 tracks missing well-known Symbol.* constants, not new Symbol construction semantics.

[andrewtdiz]

I’ll take this one. I checked the current open PR list and don’t see overlapping new Symbol / new BigInt constructor-rejection work. I’ll add focused Node parity coverage for direct, globalThis, and obvious alias constructor forms, then route both static and dynamic new paths to a TypeError while leaving callable Symbol(...) and BigInt(...) behavior unchanged.

[andrewtdiz]

Opened PR: #2903

Added test-parity/node-suite/globals/symbol-bigint-new-throws.ts covering callable Symbol(...) / BigInt(...) plus direct, globalThis, and alias new forms.

Evidence:

• Pre-fix focused parity failed and the alias path crashed after direct/global forms returned placeholder objects: test-parity/reports/parity_report_20260530_015442.json
• Post-fix focused parity passes: test-parity/reports/parity_report_20260530_020504.json

The fix routes static new Symbol / new BigInt and globalThis member constructor forms through TypeError-throwing runtime helpers after argument evaluation, and rejects alias constructor forms in js_new_function_construct.

Validation run:

• PATH="/root/.npm/_npx/8758e404b5eed2f3/node_modules/.bin:$PATH" node --experimental-strip-types test-parity/node-suite/globals/symbol-bigint-new-throws.ts
• env RUSTC_WRAPPER= PATH="/root/.npm/_npx/8758e404b5eed2f3/node_modules/.bin:$PATH" ./run_parity_tests.sh --suite node-suite --module globals --filter symbol-bigint-new-throws
• env RUSTC_WRAPPER= cargo check -p perry-hir -p perry-codegen -p perry-runtime
• cargo fmt --all -- --check
• git diff --check
• git diff --cached --check
• jq empty test-parity/known_failures.json test-parity/threshold.json

Known unrelated repo check: ./scripts/check_file_size.sh still fails on crates/perry-runtime/src/node_stream_readwrite.rs at 2001 lines; that file is unchanged here and is already 2001 lines on origin/main.