[siracozmen01]

Problem

mwinMessageButtons in include/maul-window/services.h is the only plural typedef in the API that is a choice rather than a set of flags. It is an int32_t with mwin_buttonsOk, mwin_buttonsOkCancel and mwin_buttonsYesNo, held by mwinMessageBoxDef.buttons. Every other plural (mwinModifiers, mwinMouseButtons, mwinGamepadButtons, mwinDragContents, mwinPenFlags) is a uint32_t set of flags, so def.buttons = mwin_buttonsOk reads as a set of buttons.

Change

• The type becomes mwinMessageChoice.
• Its constants become mwin_choiceOk, mwin_choiceOkCancel and mwin_choiceYesNo. They take the type's last word, as mwin_purpose* does for mwinInputPurpose. That keeps them apart from mwinMessageKind's mwin_message* constants: both types are int32_t, so one passed for the other would compile.
• The mwinMessageBoxDef field becomes choice.
• Every backend's message box and its tests read the new field. docs/user/services.md now says a message box has "a choice of buttons".
• No name in the API started with mwin_choice or was mwinMessageChoice before.

Verification

• Built with no errors or warnings: the gcc-14 tree, the clang-20 ASan, minimal and fuzz trees, MinGW, wasm32-wasi and Android.
• The macOS and iOS files (src/macos_message_box.m, src/ios_message_box.m, test/test_macos_dialog.m, test/test_ios_services.m) change only in the names, and clang-format has nothing to change in them. Their CI cells build and run them.
• tools/check.sh changed passes, the kinds and message_box suites included.
• In the CI image, x11_message_box passes on the build's own Xvfb, and web_services passes in headless Chrome on the web-wasi build.
• test_win32_message_box.exe passes under Wine. Only the Windows CI runner runs it on Windows itself.

BREAKING CHANGE: mwinMessageButtons is now mwinMessageChoice, its constants mwin_buttonsOk, mwin_buttonsOkCancel and mwin_buttonsYesNo are now mwin_choiceOk, mwin_choiceOkCancel and mwin_choiceYesNo, and mwinMessageBoxDef.buttons is now mwinMessageBoxDef.choice.
Closes #487

[siracozmen01]

Review at 7d9ea5e: ready, with one point not verified. No findings. Read, not run.

• include/maul-window/services.h at the head has the type mwinMessageChoice (line 106), the constants mwin_choiceOk/OkCancel/YesNo with the same values 0, 1 and 2, the field choice (130), and the inline default def.choice = mwin_choiceOk (155). Nothing in the header still says mwinMessageButtons, mwin_buttons or .buttons for the box.
• docs/user/services.md says "a choice of buttons" and names no old identifier.
• The 16 files change only the names. The backends (message_box.c, posix, x11, web, win32, and the macOS and iOS .m files) read def->choice and compare with mwin_choice*. The values are unchanged, so no switch moves.
• Type: the second read of #371 filed #483 as a rename for clarity, and it landed as fix(input)! (#485). This one comes from the same read's questions, so fix(services)! follows that precedent. The first read's renames (#384 to #394) were refactor!. In 0.x either one bumps the minor through !, and fix puts the entry under Bug Fixes, where it is visible, which suits a rename users must act on.
• The BREAKING CHANGE footer names all three renames.

Not verified: leftovers outside the 16 files. I can only read the files the diff touches and the ones I fetch, not search the tree. A search over the whole repository at this head for mwinMessageButtons, mwin_buttonsOk, mwin_buttonsOkCancel, mwin_buttonsYesNo and \.buttons\b near a message box would settle it, including samples/, any bindings, java/ and docs/. The build covers C and the .m files the CI cells compile, but not docs, samples outside the build, or Java strings.