DEV-SAMPLES-C-PMUniText (PMUniText)

Prototype: Unicode text (including Japanese, Chinese, Korean, Russian, Greek) in a plain Presentation Manager program - no Qt, no SDL, no external libraries. Text comes from UTF-8 language files that sit next to the executable.

How it works

Piece	Where	What it does
UTF-8 language files	lang\*.txt	KEY=value, \n = line break, # comment; added/edited without recompiling
UTF-8 text	src\lang.c utf8_sanitize()	text stays UTF-8 in memory; BOM skipped, malformed sequences become U+FFFD
Drawing	src\uni.c uni_begin/uni_draw	the UTF-8 is converted to UCS-2 for every GPI call and drawn in code page 1200: logical font with usCodePage = 1200, GpiSetCp(hps, 1200), GpiCharStringAt with the length in bytes (2 per character)
Menus	src\main.c	every item switched to MIS_OWNERDRAW; the UTF-8 label is the item handle (hItem) and is painted with the same drawing code
Plain fallback	uls_to_cp()	UTF-8 -> process code page through ULS (UCONV.DLL, loaded at run time); ? for characters that do not exist in the code page
Font choice	resolve_face()	per-language FONT= list, then built-in candidates; a face is accepted only if it really draws the language's characters (rendered into a memory bitmap and compared with the face's "missing glyph")

Language menu items are labelled with each language's own name (LANG_NAME), so every script is visible in the menu at once.

Findings (ArcaOS 5.1, tested in the VM)

• Render in code page 1200 (UCS-2), not 1208. GPI accepts UTF-8 directly (a font created with FATTRS.usCodePage = 1208 and GpiSetCp(hps, 1208) draws the same pels as the UCS-2 text in code page 1200; probe\probe1208.c, probe\probe1208.log), but 1208 is meant for reading and writing UTF-8, not for display: it is much slower to render. 1200 is also a standard Unicode encoding (UCS-2 / the fixed-width subset of UTF-16) and the GPI text functions take length-bounded byte arrays, so a UniChar string goes in as it is, without a terminator. Thanks to Alex Taylor, who tested this extensively (his own samples: https://altsan.org/os2/toolkits/uls/index.html#samples). (The first version of this sample used 1200, version 0.3 switched to 1208 after a suggestion by Dave Yeo, version 0.4 is back on 1200 for the speed.) Note that the font's code page decides: a font created for 1200 followed by GpiSetCp(1208) draws garbage.
• GPI length arguments are in bytes (2 per UniChar for code page 1200).
• Non-BMP characters (4-byte UTF-8) are sent as surrogate pairs; they do not crash, but the tested fonts have no glyphs for them (two missing-glyph boxes).
• Faces that contain Kana, Kanji/Hanzi and Hangul glyphs: Droid Sans Combined, Times New Roman MT 30, Monotype Sans Duospace WT J, Times New Roman WT J. Latin/Greek/Cyrillic: most of the installed outline fonts. Full list: probe\probe.log.
• PM sends WM_MEASUREITEM / WM_DRAWITEM of a menu to the frame (the menu's owner), not to the menu window. Subclassing only the menu window gives an empty menu bar.
• The title bar and standard controls cannot show Unicode. The title therefore uses the ASCII English language name (LANG_ENGLISH); everything else that must be localized is drawn.
• Plain (non-owner-drawn) menus work but show ? for anything outside code page 850. Switch between both modes in Options to see the difference.

Mnemonics and accelerators

• Mnemonics: put one ~ before the letter in a MENU_* value, like a normal PM menu (MENU_FILE=~File). For scripts without Latin letters use the Asian convention and add the Latin letter in parentheses (MENU_FILE=ファイル(~F)). The mnemonic is underlined in the owner-drawn item; Alt+letter opens a top-level menu and a plain letter runs an item of an open submenu (PM only does this for text items, so main.c does it for owner-drawn ones). In plain-text mode PM handles the ~ itself.
• Accelerators: keys are bound in the ACCELTABLE in src\main.rc (F5 reload, Ctrl+X exit, Ctrl+U Unicode menus, Ctrl+B larger, Ctrl+S smaller, F1 about). The key names shown right-aligned in the menu come from the accels[] table in main.c, so keep both in sync.
• PM does not set KC_CHAR for Alt+letter WM_CHAR messages; check KC_ALT and the character code.

Limits of the prototype

• Only the BMP is usable in practice: GPI accepts non-BMP UTF-8 (emoji, rare CJK ideographs) but the installed fonts have no glyphs for it. No right-to-left or complex shaping (Arabic, Hebrew, Indic) - GPI does no layout.
• No text input / IME. Entry fields, MLEs and list boxes still use the process code page.
• IPF help (wipfc) is not Unicode; a Unicode help text would need its own viewer.

Command line

pmunitext [-plain] [-lang code] [-log] (-log writes pmunitext.log with the owner-draw messages)

Build (both toolchains)

• Open Watcom: compile_wat.cmd -> bin-wat\pmunitext.exe (+ bin-wat\lang\)
• GCC / kLIBC: compile_gcc.cmd -> bin-gcc\pmunitext.exe (+ bin-gcc\lang\)

Layout

src\       main.c (window, menus, painting)  uni.c (UTF-8 helpers, ULS, fonts, drawing)  lang.c (language files)
lang\      en es ru el ja zh_CN zh_TW ko
img\       screenshots
probe\     probe.c + probe.log  (what GPI does with code page 1200 on this system)
           probe1208.c + probe1208.log  (GPI with UTF-8 / code page 1208: works, but slower than 1200)

License

BSD 3-Clause (see LICENSE), same as the PM Template.

Release notes

• 0.4 - drawing is back on code page 1200: uni.c converts the UTF-8 text to UCS-2 for each GPI call (uni_width, uni_draw and the font coverage test), the font and GpiSetCp use 1200. Code page 1208 renders correctly but is slow (feedback from Alex Taylor). The uni_* functions still take UTF-8 and byte lengths, so the callers did not change.
• 0.3.1 - build scripts: compile_gcc.cmd sets MAKESHELL=cmd.exe (make could not find a shell on some systems; fix by Dave Yeo) and both compile scripts now use setlocal/endlocal, so they no longer change the caller's environment (MAKESHELL, EMXOMFLD_*, PATH...). The wlink settings stay: the default ilink gave warnings and an executable that hung the system. Also fixed in 0.3 and worth knowing: in 0.1/0.2 the Options - Font submenu could show shortcut texts ("Ctrl+U", "Ctrl+B"...) instead of the font names, because the owner-drawn items pointed into a small rotating buffer that was overwritten by later calls; 0.3 no longer uses that buffer.
• 0.3 - text was drawn as UTF-8 through GPI code page 1208 (replaced by 0.4).
• 0.2 - mnemonics (underlined, Alt+letter and letters in open submenus) and accelerator column for owner-drawn menus.
• 0.1 - first prototype (October 2026).