wasm-gc backend: ReadOnlyArray::at fails to compile with packed-i8 array.get error

[bobzhang]

Running moon test on a clean checkout of main (or any branch — reproduces independently of source changes) fails on the wasm-gc backend during execution of the bool blackbox test driver:

CompileError: WebAssembly.Module(): Compiling function #248:"_M0MPC15array13ReadOnlyArray2atGjE" failed: array.get: Immediate array type 71 has packed type i8. Use array.get_s or array.get_u instead. @+38042
Error: Failed to run the test: /Users/dii/git/core/_build/wasm-gc/debug/test/bool/bool.blackbox_test.wasm
The test executable exited with exit status: 1

The mangled symbol _M0MPC15array13ReadOnlyArray2atGjE resolves to ReadOnlyArray::at (builtin/readonlyarray.mbt:25-41), which delegates to FixedArray::at. The emitted wasm uses the unsigned array.get opcode against an array type whose element is declared as packed i8, which the wasm-gc spec only allows via array.get_s / array.get_u. Node's V8 engine therefore rejects the module at instantiation.

The compile-time output is fine; the failure happens when V8 validates the module, so the problem is in moonc's wasm-gc codegen for ReadOnlyArray[T] instantiated at a type whose backing FixedArray uses a packed representation.

Reproduction

git clone git@github.com:moonbitlang/core.git
cd core
moon test                        # fails
moon test -p moonbitlang/core/bool   # smaller repro
moon test --target js            # passes (6367 tests)
moon test --target native        # passes (6321 tests)

Environment

• macOS (Darwin 25.4.0, arm64)
• moon 0.1.20260429 (26f35db 2026-04-29)
• moonc v0.9.1+dd1e997c9-nightly (2026-04-30)
• moonrun 0.1.20260429
• Node.js v25.8.2 (Homebrew)

Notes

• JS and native backends are unaffected.
• CI presumably uses an older Node.js (or a moonrun that bundles its own runtime), since the wasm-gc job passes there. This may be a regression triggered by stricter validation in newer V8 versions, or by recent codegen changes for ReadOnlyArray / packed FixedArray.

Originally surfaced while validating an unrelated PR (#3505); filing here so it can be tracked independently.