[RFC] options: convert some guarded options to new warning mechanism

[Dudemanguy]

Here's my attempt at implementing what I suggested in #14618 (comment). The mechanism works fine. I implemented both per option and for sub options.

1. I had to add a bunch of defines because of windows junk. It's not really a big deal but it's kind of ugly imo.
2. Not really sure how we want to handle suboptions that are conditional and private somewhere. They could just be moved to options.h for example, but then that makes them not-private obviously. Not a blocker for the feature but something to think about.

[github-actions]

Download the artifacts for this pull request:

Windows

• mpv-i686-w64-mingw32
• mpv-x86_64-w64-mingw32
• mpv-x86_64-windows-msvc

macOS

• mpv-macos-12-intel
• mpv-macos-13-intel
• mpv-macos-14-arm

[kasper93]

Maybe instead of defining all win32 dummy values in meson, we could try to not use them?

    {"window-corners",
#if HAVE_WIN32_DESKTOP
       OPT_CHOICE(window_corners,
        {"default", DWMWCP_DEFAULT},
        {"donotround", DWMWCP_DONOTROUND},
        {"round", DWMWCP_ROUND},
        {"roundsmall", DWMWCP_ROUNDSMALL})}
#else
       OPT_NOTAVAILABLE}
#endif

formatting can be fixed to look better, but point is to avoid using those values. I feel like those dummy definitions are polluting too much

[Dudemanguy]

We do need to be sure to keep the same valid choices to avoid it erroring out. I could do something like this.

#if HAVE_WIN32_DESKTOP
       OPT_CHOICE(window_corners,
        {"default", DWMWCP_DEFAULT},
        {"donotround", DWMWCP_DONOTROUND},
        {"round", DWMWCP_ROUND},
        {"roundsmall", DWMWCP_ROUNDSMALL})}
#else
        OPT_CHOICE_ZERO(window_corners, "default",  "donotround", "round", "roundsmall")
#endif

Only bummer with this approach is that the option names have to be typed twice but maybe not such a big deal.

Edit: Wasn't smart enough to think of a clever macro to shorthand it oh well.

[kasper93]

We do need to be sure to keep the same valid choices to avoid it erroring out.

Can't we just skip checking choices for unavailable options?

EDIT: Also this would actually be required, because we don't want to run string validators on them, those can do platform specific checks.

[Dudemanguy]

I don't see how we skip setting choices short of skipping the option altogether?

[kasper93]

I though the point was to show warning when using not compiled in option. So why it maters what args are used, if it is not available anyway?

[Dudemanguy]

Well the choices could be set to nothing but then that makes them hard errors on the cli again which is inconsistent with the other ones.

[kasper93]

Now that it will not be hard error we can remove this from test

diff --git a/test/meson.build b/test/meson.build
index 44983f18d0..1b7bf11ff6 100644
--- a/test/meson.build
+++ b/test/meson.build
@@ -119,7 +119,7 @@ if get_option('libmpv')
     exe = executable('libmpv-test', 'libmpv_test.c',
                      include_directories: incdir, link_with: libmpv)
     file = join_paths(source_root, 'etc', 'mpv-icon-8bit-16x16.png')
-    test('libmpv', exe, args: file, timeout: 60, should_fail: not features['lua'])
+    test('libmpv', exe, args: file, timeout: 60)
 
     exe = executable('libmpv-encode', 'libmpv_encode.c',
                      include_directories: incdir, link_with: libmpv)

[Dudemanguy]

No that test still fails which I'd expect because loading the osc won't work.

[na-na-hi]

However, this hurts the portability of mpv config files and means it's
possible for configs to break depending on a machine. This is not so
nice.

Is it? The affected options won't work anyway if they are unavailable, and using them only hard quits if they are from the commandline, but doesn't quit if it's in a config file. This PR only changes the logs from errors to warnings and makes no meaningful difference other than allowing commandline usage.

[Dudemanguy]

I feel like I've encountered something that errored out because of this in the past, but I guess I can't think of any examples.