runner: --describe prints a core's options and input labels, no ROM needed - #1

What

retro-core-runner --describe --core <lib> [--package <shim>] prints what a core declares — its options and input descriptors — without a ROM and without running a session. --version gains describe 1 so a host can check for support first, and runner_probe exposes it as RunnerVersion::describe.

This is what retro-hub's new settings page is built from (Retro-Launcher feat/direct-home): labels, choices and button names come from the core, not from a list compiled into the hub.

Behaviour

• The core is loaded and its .rcore.toml checked exactly as for a run; a mismatch still exits 2.
• options() and input_descriptors() are read, and nothing else: no init, load or deinit, and --out is not created. rcore.h declares both "read before load".
• --package is validated as usual: refused on a core without game_package, and the file must be readable.
• Failure: the message goes to stderr, stdout stays empty, and the exit code is 2.

Format (docs/CORE_RUNNER.md)

One record per line, TAB-separated, with \, TAB, LF and CR escaped:

describe	1
core	<id>	<version>	<platforms>
option	<key>	<enum|bool|int|string>	<restart,netplay,developer|->	<has_default>	<default>	<min>	<max>	<label>	<description>
value	<key>	<enum value>
input	<button bit>	<axis+1>	<axis_direction>	<label>

Tests

• ctest 16/16: the 11 existing tests plus 5 new runner_describe_* cases (plain, package, package_none, unwanted, no_core). They compare stdout byte for byte, and the fake cores now declare fields carrying TAB, LF, CR and backslash.
• To confirm the tests can fail, I broke the expected output (wrong labels, a wrong escape); they failed.
• Real n64lle generic core 0.374.0: exit 0, 76 options, 16 values and 16 inputs, all with the right field counts.

Open

The ABI never says how an RCORE_OPT_BOOL value is spelled. --describe prints the core's default as given; the hub only draws a checkbox when that default shows the spelling.

🤖 Generated with Claude Code