Merge pull request #22991 from hrydgard/rtl-text

Lay out right-to-left text consistently across text drawers, fix fa_IR/ar_AE strings

Common/Data/Encoding/Utf8.cpp

@@ -169,6 +169,24 @@ bool UTF8StringHasNonASCII(std::string_view utf8string) {
 	return UTF8StringNonASCIICount(utf8string) > 0;
 }
 
+static bool CodepointIsRTL(uint32_t c) {
+	return (c >= 0x0590 && c <= 0x08FF) ||  // Hebrew, Arabic, Syriac, Thaana, NKo, ...
+		(c >= 0xFB1D && c <= 0xFDFF) ||  // Hebrew and Arabic presentation forms
+		(c >= 0xFE70 && c <= 0xFEFF) ||
+		(c >= 0x10800 && c <= 0x10FFF) ||
+		(c >= 0x1E800 && c <= 0x1EFFF);
+}
+
+bool UTF8StringHasRTL(std::string_view utf8string) {
+	UTF8 utf(utf8string);
+	while (!utf.end()) {
+		if (CodepointIsRTL(utf.next())) {
+			return true;
+		}
+	}
+	return false;
+}
+
 #ifdef _WIN32
 
 std::string ConvertWStringToUTF8(const wchar_t *wstr) {

Common/Data/Encoding/Utf8.h

@@ -92,6 +92,8 @@ private:
 int UTF8StringNonASCIICount(std::string_view utf8string);
 
 bool UTF8StringHasNonASCII(std::string_view utf8string);
+// True if the string contains a character from a right-to-left script (Hebrew, Arabic, ...).
+bool UTF8StringHasRTL(std::string_view utf8string);
 
 
 // Removes overlong encodings and similar.

Common/Data/Text/I18n.cpp

@@ -50,6 +50,10 @@ std::string I18NRepo::LanguageID() {
 	return languageID_;
 }
 
+bool I18NRepo::IsRTL() const {
+	return languageID_ == "he_IL" || languageID_ == "ar_AE" || languageID_ == "fa_IR" || languageID_ == "ku_SO";
+}
+
 I18NRepo::I18NRepo() {
 	Clear();
 }

Common/Data/Text/I18n.h

@@ -108,6 +108,8 @@ public:
 	bool LoadIni(const std::string &languageID, const Path &overridePath = Path()); // NOT the filename!
 
 	std::string LanguageID();
+	// The loaded language is written right to left.
+	bool IsRTL() const;
 
 	std::shared_ptr<I18NCategory> GetCategory(I18NCat category);
 

Common/Render/Text/draw_text.cpp

@@ -17,6 +17,7 @@
 #include "Common/System/Display.h"
 #include "Common/GPU/thin3d.h"
 #include "Common/Data/Hash/Hash.h"
+#include "Common/Data/Text/I18n.h"
 #include "Common/Data/Text/WrapText.h"
 #include "Common/Data/Encoding/Utf8.h"
 
@@ -90,6 +91,33 @@ const std::string &TextDrawer::WrapString(const FontStyle &style, float scale, s
 	return entry.wrapped;
 }
 
+// The backends disagree on a line's base direction: GDI and DirectWrite always start left to right, the others
+// go by the first strong character, which is wrong for a Hebrew sentence that starts with "PPSSPP". An explicit
+// embedding (RLE ... PDF) means the same to all of them.
+std::string_view TextDrawer::ApplyBaseDirection(std::string_view str, std::string *storage) {
+	if (!g_i18nrepo.IsRTL() || !UTF8StringHasRTL(str)) {
+		return str;
+	}
+	static constexpr std::string_view RLE = "\xE2\x80\xAB";  // U+202B
+	static constexpr std::string_view PDF = "\xE2\x80\xAC";  // U+202C
+	storage->clear();
+	std::vector<std::string_view> lines;
+	SplitString(str, '\n', lines);
+	for (size_t i = 0; i < lines.size(); i++) {
+		if (i > 0) {
+			storage->push_back('\n');
+		}
+		if (UTF8StringHasRTL(lines[i])) {
+			storage->append(RLE);
+			storage->append(lines[i]);
+			storage->append(PDF);
+		} else {
+			storage->append(lines[i]);
+		}
+	}
+	return *storage;
+}
+
 float TextDrawer::CalculateDPIScale() const {
 	if (ignoreGlobalDpi_)
 		return dpiScale_;
@@ -143,7 +171,8 @@ void TextDrawer::DrawString(DrawBuffer &target, const FontStyle &style, float sc
 		TextureDesc desc{};
 		std::vector<uint8_t> bitmapData;
 		EnsureFont(style);
-		if (!DrawStringBitmap(style, bitmapData, *entry, bitmapFormat, str, align, emoji)) {
+		std::string bidiStorage;
+		if (!DrawStringBitmap(style, bitmapData, *entry, bitmapFormat, ApplyBaseDirection(str, &bidiStorage), align, emoji)) {
 			// Nothing drawn. The entry without a texture records that fact.
 			return;
 		}
@@ -199,7 +228,8 @@ void TextDrawer::MeasureString(const FontStyle &style, float scaleX, float scale
 	if (inserted) {
 		EnsureFont(style);
 		float extW = 0.0f, extH = 0.0f;
-		MeasureStringInternal(style, str, &extW, &extH);
+		std::string bidiStorage;
+		MeasureStringInternal(style, ApplyBaseDirection(str, &bidiStorage), &extW, &extH);
 		entry.width = extW;
 		entry.height = extH;
 		// Hm, use the old calculation?
@@ -264,7 +294,9 @@ bool TextDrawer::DrawStringBitmapRect(const FontStyle &style, float scale, std::
 
 	const int wrap = align & (FLAG_WRAP_TEXT | FLAG_ELLIPSIZE_TEXT);
 	EnsureFont(style);
-	return DrawStringBitmap(style, bitmapData, entry, texFormat, wrap ? std::string_view(WrapString(style, scale, str, bounds.w, wrap)) : str, align, fullColor);
+	std::string bidiStorage;
+	const std::string_view text = wrap ? std::string_view(WrapString(style, scale, str, bounds.w, wrap)) : str;
+	return DrawStringBitmap(style, bitmapData, entry, texFormat, ApplyBaseDirection(text, &bidiStorage), align, fullColor);
 }
 
 void TextDrawer::AddDropShadow(std::vector<uint8_t> &bitmapData, TextStringEntry &entry, Draw::DataFormat srcFormat) const {

Common/Render/Text/draw_text.h

@@ -114,6 +114,10 @@ protected:
 	// Wraps and/or ellipsizes, caching the result. The reference stays valid until the next OncePerFrame.
 	const std::string &WrapString(const FontStyle &style, float scale, std::string_view str, float maxWidth, int flags);
 
+	// When the UI language is right to left, puts each line containing right-to-left text in an RTL embedding,
+	// so every backend lays it out right to left, whatever its first character. Otherwise returns str as is.
+	static std::string_view ApplyBaseDirection(std::string_view str, std::string *storage);
+
 	// v has the 8-bit alpha.
 	static u16 AlphaToPremul4444(u32 v) {
 		v = (v >> 4) & 0x0F;

Common/Render/Text/draw_text_sdl.cpp

@@ -58,6 +58,15 @@ static bool ReorderBidi(std::string_view str, std::string *out) {
 			result.push_back('\n');
 		}
 		std::string_view line = lines[i];
+		FriBidiParType baseDir = FRIBIDI_PAR_ON;
+		// TextDrawer::ApplyBaseDirection wraps RTL lines in RLE ... PDF. Take the direction from that instead,
+		// so the marks don't count as a second level and send pure Arabic lines down the reordering path.
+		const std::string_view rle = "\xE2\x80\xAB";
+		const std::string_view pdf = "\xE2\x80\xAC";
+		if (line.size() >= rle.size() + pdf.size() && startsWith(line, rle) && endsWith(line, pdf)) {
+			line = line.substr(rle.size(), line.size() - rle.size() - pdf.size());
+			baseDir = FRIBIDI_PAR_RTL;
+		}
 		logical.resize(line.size() + 1);
 		FriBidiStrIndex len = fribidi_charset_to_unicode(FRIBIDI_CHAR_SET_UTF8, line.data(), (FriBidiStrIndex)line.size(), logical.data());
 		if (len <= 0) {
@@ -66,7 +75,6 @@ static bool ReorderBidi(std::string_view str, std::string *out) {
 		}
 		visual.resize(len + 1);
 		levels.resize(len);
-		FriBidiParType baseDir = FRIBIDI_PAR_ON;
 		if (!fribidi_log2vis(logical.data(), len, &baseDir, visual.data(), nullptr, nullptr, levels.data())) {
 			return false;
 		}

UI/MiscScreens.cpp

@@ -304,9 +304,8 @@ NewLanguageScreen::NewLanguageScreen(std::string_view title) : ListPopupScreen(t
 			continue;
 		}
 
-		// We only support Arabic on platforms where we have support for the native text rendering
-		// APIs, as proper Arabic support is way too difficult to implement ourselves.
-#if !(PPSSPP_PLATFORM(WINDOWS) || PPSSPP_PLATFORM(ANDROID))
+		// Arabic script needs a text drawer that does shaping (see TextDrawer::Create), the atlas font can't.
+#if defined(__LIBRETRO__) || !(PPSSPP_PLATFORM(WINDOWS) || PPSSPP_PLATFORM(ANDROID) || PPSSPP_PLATFORM(MAC) || PPSSPP_PLATFORM(IOS) || defined(USE_SDL3_TTF))
 		if (tempLangs[i].name.find("ar_AE") != std::string::npos) {
 			continue;
 		}