stdlib doesn't throw Node arg-validation errors (ERR_INVALID_ARG_TYPE/EBADF/...) on bad input — ~85 tests across fs/buffer/net/http/crypto/zlib/process

[proggeramlug]

Summary

Cross-cutting parity theme, surfaced when assert.throws got fixed (#1924): Perry's APIs don't throw Node's argument-validation errors on invalid input. ~85 tests across nearly every API now correctly fail with "Missing expected exception" (the call didn't throw) or wrong error code.

Confirmed repro (perry 0.5.1032):

import * as fs from "node:fs";
fs.readFileSync(123 as any);   // Perry: returns/no throw   Node: throws EBADF
fs.accessSync(true as any);    // Perry: throws code=undefined   Node: ERR_INVALID_ARG_TYPE

Breadth (assert.throws-style failures that were false-passing before #1924)

API	failures	API	failures
fs	40	crypto	7
buffer	10	process	8
net	10	url	2
zlib	10	timers	1
http	9

What to do

Implement Node's input validation + error codes (ERR_INVALID_ARG_TYPE, ERR_OUT_OF_RANGE, EBADF, ...) on the stdlib surface — at minimum throw with the right .code/.name on bad arg types/values. This is the single highest-leverage parity theme now: it spans 7+ APIs and ~85 tests. Suggest starting with fs (40) — the validation helper there is reusable across the others.

Note: this is an honest-accounting gain, not a regression — #1924 stopped assert.throws from passing when nothing was thrown, revealing these were never actually validating. Real parity is ~28% (was 32.6% inflated).

Context: #800 --auto-optimize sweep, perry 0.5.1032.

[proggeramlug]

Update after #2035: the validation pattern is proven — fs.accessSync(true) now throws ERR_INVALID_ARG_TYPE and fs.readFileSync(123) throws EBADF, matching Node (verified). But this was a narrow slice; the radar parity is flat because the remaining tests check other functions still not validating:

fs (27 tests still 'Missing expected exception'): writeSync/writevSync, copyFile(Sync), exists(Sync), fchown(Sync), fsync/fdatasync, lchmod/lchown, callback-type validation (make-callback), and encoding validation. The validator added in #2035 should extend across these.

Other APIs untouched (~45 tests): buffer (10), net (10), zlib (10), http (9), process (8), crypto (7), url (2) — all still skip Node's arg-validation/error-codes.

Rolling the #2035 validator across the rest of fs + these APIs is what flips the ~80 remaining tests. (Full-test-file parity is conjunctive — each function's validation only flips the tests that depend solely on it, which is why per-slice gains look small.)