[ahmadmuj]

What

--frame-red-zone only works for Windows output, but ParseArgs let it through without --windows. The only check was in Conversion::Convert, so the relinker exited 2 (conversion failed) instead of 1 (invalid arguments), and usually with an unrelated error first, since sce_module discovery runs before that check:

$ relinker --frame-red-zone in.elf out.elf
FAIL: sce_module/*.prx was not found in: ...     # exit 2

--windows-diagnostics, --windows-gui and --windows-fixed-base already fail in ParseArgs with "requires --windows". This adds the same check for --frame-red-zone:

FAIL: --frame-red-zone requires --windows        # exit 1

The Conversion.cpp guard stays for library callers. Same flag as #10682, different gap.

Tested

Ubuntu 24.04 on WSL2, GCC 13.3, relinker-only build. New frameRedZoneRequiresWindows in cli_args fails on main with expected an exception for [--frame-red-zone input.elf output.elf] and passes with the change. All 265 relinker-only tests pass. check_conventions.py --base origin/main passes. Ran the CLI before and after as shown above; --windows --frame-red-zone still gets to the framing pass.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Fixes: none
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

[thepoppfish]

Reviewed exact head 1c3bb466a89e745441023a618e7144abd6440bcb. No findings in this focused change; this adds native Windows coverage to the author's Linux report.

Windows 11 build 26200, MSYS2 UCRT64 GCC 16.2.0-6, CMake 4.3.1, Ninja 1.13.2, Python 3.14.7:

• Configured the repository's actual relinker-only Release build and built relinker and cli_args_tests (93 Ninja steps), without compiler warnings in the build log. No custom replacement for parser, path handling or conversion behavior.
• The complete registered cli_args test passes three consecutive runs (0.04/0.01/0.01 s). A scratch control removing only the two-line guard, retaining the full submitted CliArgsTests.cpp and actual NativePath.cpp, exits 1 with expected an exception for [--frame-red-zone input.elf output.elf].
• Seven registered tests pass: cli_help, input_magic, input_preservation, output_collisions, secondary_output_collisions, input_file, input_nonascii (37.71 s total, slowest individual test 14.39 s).
• Direct invocations of the actual built relinker with nonexistent input and an existing output sentinel confirm four rejections: default target, --macos before/after the flag, and --to-rosetta. Each exits 1 with exactly FAIL: --frame-red-zone requires --windows, before input access; the sentinel survives. Both orders of --windows and --frame-red-zone parse successfully and reach the missing-input conversion failure (exit 2) with the sentinel intact. These accepted-argument checks are not a claim that a title was successfully framed.
• Conventions and whitespace pass against local main b2c7a388406614fb3bc6723a6717a37196d617b8; merge-tree against that same main is clean.

The check is alongside the existing Windows-only option checks, after parsing all flags, so either accepted option order works. The Conversion::Convert guard remains available for library callers. The added unit test covers default/Linux and explicit macOS rejection plus Windows acceptance, without changing existing assertions.

Scope: the actual Windows CLI and focused tests above. I did not run all 265 registered tests, a full system-library/product build, Linux/macOS execution, sanitizer, title/GPU tests or cross-PR composition. AI-assisted: yes, OpenAI Codex.

Replay commands and standalone control

Put MinGW GCC, CMake and Ninja on PATH. On the exact checkout:

cmake -S . -B ../build-relinker-10912 -G Ninja -DCMAKE_BUILD_TYPE=Release -DANYPS5_RELINKER_ONLY=ON -DBUILD_TESTING=ON '-DCMAKE_CXX_COMPILER=g++.exe' '-DPython3_EXECUTABLE=C:/path/to/python.exe'
cmake --build ../build-relinker-10912 --target relinker cli_args_tests -j 3
ctest --test-dir ../build-relinker-10912 -R '^cli_args$' --output-on-failure --repeat until-fail:3
ctest --test-dir ../build-relinker-10912 -R '^(cli_help|input_magic|input_file|input_preservation|output_collisions|secondary_output_collisions|input_nonascii)$' --output-on-failure

For the standalone negative control, place the following files in ../review-10912-controls with the exact checkout at ../review-10912 relative to that directory (or adjust REPO and the script paths). Run python make_control.py, configure/build its CMake project, and run build16/cli_args_old.exe. It should fail at the new guard assertion above. smoke_cli.py runs the six direct invocations against the real built relinker; its temporary input/output directory cleans itself up.

CMakeLists.txt

cmake_minimum_required(VERSION 3.20)
project(AnyPS5CliArgsControl LANGUAGES CXX)
set(REPO "${CMAKE_CURRENT_LIST_DIR}/../review-10912" CACHE PATH "Exact PR checkout")
add_executable(cli_args_old
    "${REPO}/core/relinker/cli/tests/CliArgsTests.cpp"
    CliArgs-old.cpp
    "${REPO}/core/relinker/io/src/NativePath.cpp")
target_compile_features(cli_args_old PRIVATE cxx_std_20)
target_include_directories(cli_args_old PRIVATE
    "${REPO}/core/relinker/cli/include"
    "${REPO}/core/relinker/io/include"
    "${REPO}/core/relinker/domain/include")
target_link_options(cli_args_old PRIVATE -static -static-libgcc -static-libstdc++)

make_control.py

from pathlib import Path

root = Path(__file__).resolve().parent
source = (root.parent / 'review-10912/core/relinker/cli/src/CliArgs.cpp').read_text()
guard = '    if (args.frameRedZone && !args.toWindows)\n        throw std::runtime_error("--frame-red-zone requires --windows");\n'
assert source.count(guard) == 1
(root / 'CliArgs-old.cpp').write_text(source.replace(guard, ''))

smoke_cli.py

from pathlib import Path
import subprocess
import tempfile

root = Path(__file__).resolve().parent
tool = root.parent / 'build-relinker-10912/core/relinker/relinker.exe'
with tempfile.TemporaryDirectory() as directory:
    source = Path(directory) / 'missing.elf'
    output = Path(directory) / 'output.exe'
    original = b'existing output must survive'
    output.write_bytes(original)
    rejected = (
        ['--frame-red-zone'],
        ['--macos', '--frame-red-zone'],
        ['--frame-red-zone', '--macos'],
        ['--to-rosetta', '--frame-red-zone']
    )
    accepted = (
        ['--windows', '--frame-red-zone', '--skip-sce-module'],
        ['--frame-red-zone', '--windows', '--skip-sce-module']
    )
    for flags in rejected:
        result = subprocess.run([str(tool), *flags, str(source), str(output)],
                                capture_output=True, text=True, timeout=30)
        assert result.returncode == 1, (flags, result.returncode, result.stdout, result.stderr)
        assert result.stderr.strip() == 'FAIL: --frame-red-zone requires --windows', (flags, result.stderr)
        assert output.read_bytes() == original and not source.exists()
        print('argument rejection:', ' '.join(flags), 'exit 1, exact diagnostic, output preserved')
    for flags in accepted:
        result = subprocess.run([str(tool), *flags, str(source), str(output)],
                                capture_output=True, text=True, timeout=30)
        assert result.returncode == 2, (flags, result.returncode, result.stdout, result.stderr)
        assert '--frame-red-zone requires --windows' not in result.stderr
        assert output.read_bytes() == original and not source.exists()
        print('accepted arguments:', ' '.join(flags), 'reaches missing-input failure, exit 2, output preserved')

[ahmadmuj]

Checked #9408 by hand since the bot couldn't compare it. In CliArgs.cpp it only removes the --trace-imports / --macos check in ParseArgs (lines 202-203 on current main). This PR only adds the --frame-red-zone / --windows check after --windows-diagnostics (lines 196-197). Applied both hunks to main acbfd7c in either order and ran git merge-file: clean, no conflicts, same result. No shared lines, so no Depends on either way.