refactor: clear every deprecation warning moon check reports

[mizchi]

Summary

moon check reported 1795 warnings on native and 10 on js, all deprecations that #179 left as a follow-up ("warning-only and left as is", ~1.5k sites). This clears all of them: every target is now 0 errors, 0 warnings.

Six commits, one per deprecation family, so each can be read on its own:

commit	what	sites
670dd27	StringBuilder::new() → StringBuilder()	304 in 144 files
a1aed6f	Array::new(capacity=..) → Array(capacity=..), and @fs.read_dir(..).to_array() → to_owned()	18 + 3
f15b5d8	@sys.{get,set,unset}_env_var / get_env_vars → the same four in moonbitlang/core/env	1552 in 117 files
6ddd9da	cmd/git-bit's implicitly promoted OsFs trait methods → two extend declarations	30
e026dcf	JS entry points: @sys.get_cli_args() → @env.args()	2
7d15d30	JS FFI signatures: Array → FixedArray	8

The three that were not plain renames

• @env.args() is not a drop-in for @sys.get_cli_args(). On the JavaScript backend it returns process.argv unchanged, while the deprecated function dropped argv[0], the runtime executable. Both JS entry points index from the script path, so they drop it at the source rather than shifting every index below.

• extend only lists the methods that are actually called. It makes a promoted method look like a function the package defines, so listing one no caller reaches (readdir, mtime, remove_file, remove_dir) trades the deprecation for an unused_value warning. is_dir has exactly one such caller, in a whitebox test, and is qualified there instead. cmd/bit holds the same OsFs but needs nothing — it reaches the filesystem through &RepoFileSystem everywhere.

• The JS FFI bodies are unchanged, because the representation they see is compatible. Measured with a throwaway module on this toolchain: FixedArray[String] arrives as the same plain JS array Array[String] did, and FixedArray[Byte] arrives as a Uint8Array — array-like, with length, and exactly what Uint8Array.from(bytes ?? []) already produced from the plain array. The MoonBit side fills a FixedArray::make buffer instead of pushing onto a growable array.

moonbitlang/core/env gained an import in bitx_bitconfig and in bit_vfs (its for "test" block, where the one call site lives); the other seven packages already had it. moonbitlang/x/sys then has no user left in seven packages, so that import goes too — cmd/bit and cmd/git-bit keep it, since @sys.exit and @sys.get_cli_args are not deprecated and are still called there.

Verification

• moon check for native, js and wasm: 0 errors, 0 warnings each (from 1795 / 10 / 0).
• moon test --target js -p mizchi/bit_lib: 361/363. The two failures are the ssh-ed25519 signing tests that shell out to ssh-keygen, which is not installed in this sandbox — the same two build: follow moonbit 0.10.13 and moonbitlang/async 0.22, update all deps #179 saw.
• node --test tools/npm-lib.test.mjs tools/npm-cli.test.mjs tools/js-build.test.mjs: 5/5, against freshly built payloads, so the JS host's byte-writing path and the relay transport's header path both run through the changed FFI. The tree-shaken bundle is 174,389 raw / 40,246 gzip bytes, just under its 174,458 / 40,265 baseline.
• moon test --target native -p mizchi/bit_runtime -p mizchi/bitx_bitconfig: 26/26 — the two modules that exercise get/set/unset round-trips most.
• moon build --target native --release modules/bit, then a smoke run of the binary: init, add, commit, status, log all behave, GIT_AUTHOR_* / GIT_COMMITTER_* / GIT_*_DATE land in the commit (read back with real git log, which is the end-to-end proof that @env.get_env_var is equivalent), git fsck is clean, and bit grep --column -n -o still reports the absolute columns.
• node tools/check-layers.mjs and tools/check-layer-guards.sh pass.
• The six commits were re-derived from a clean tree and verified byte-identical to the state all of the above ran against.

Not done / follow-ups

• moon fmt was deliberately not run. On this toolchain it reformats 123 files this change never touches, so the tree is not fmt-clean independently of this work; worth a separate pass.
• npm/hub.raw.js is a checked-in generated bundle that already differs from a fresh build by ~16.8k lines on main (older compiler codegen), so regenerating the npm payloads is its own change, not this one.
• pkf run check still passes --warn-list=-27-20-82, i.e. deprecations stay suppressed there. With the tree clean, dropping -20 would keep it that way; the cost is that a toolchain bump turns new deprecations into a red CI, so that call is left to you.
• bitx_rebase_ai imports moonbitlang/x/sys without using it — pre-existing, not caused by this change, left alone.

🤖 Generated with Claude Code

https://claude.ai/code/session_0134zgvUroAxAAggLS9yufPX

---

Generated by Claude Code