feat: Arabic/Farsi/Urdu bidi reordering and contextual shaping — PR 1/3 (#2541)

Co-authored-by: Uri Tauber <uritaube@gmail.com>

lib/MiniBidi/bidiclasses.t

@@ -1,15 +1,19 @@
-/* bidiclasses.t — bidi class table for CrossPoint Hebrew/English epub.
+/* bidiclasses.t — bidi class table for CrossPoint RTL (Hebrew/Arabic) epub.
  *
  * Coverage rationale:
- *   Hebrew + English is the primary target. However, CrossPoint renders
- *   Latin and Cyrillic scripts for many other languages, so these MUST be
- *   classified as L (not fall through to ON) to avoid regression when they
- *   appear adjacent to Hebrew runs.
+ *   Hebrew and Arabic-script languages (Arabic, Farsi, Urdu, Sindhi, Pashto,
+ *   Kurdish) are the RTL targets. CrossPoint also renders Latin and Cyrillic
+ *   scripts for many other languages, so these MUST be classified as L (not
+ *   fall through to ON) to avoid regression when they appear adjacent to
+ *   RTL runs.
  *
  *   Scripts NOT in this table fall through to ON — correct per UAX#9 for
- *   scripts CrossPoint's fonts don't support (CJK, Arabic, Devanagari, etc.)
+ *   scripts CrossPoint's fonts don't support (CJK, Devanagari, etc.)
  *   ON is the right class for "unknown" — it behaves neutrally.
  *
+ *   Arabic ranges are sourced from Unicode UCD extracted/DerivedBidiClass.txt
+ *   (values verified against Unicode 17.0.0).
+ *
  * Entries sorted ascending by first (binary search requirement).
  */
 
@@ -84,6 +88,36 @@
 {0x05D0, 0x05EA, R},    /* alef … tav */
 {0x05F0, 0x05F4, R},    /* alternative forms + geresh/gershayim */
 
+/* ── Arabic / Perso-Arabic (DerivedBidiClass.txt) ───────────────────── */
+/* Letters are AL (Arabic Letter), harakat/marks are NSM, Arabic-Indic
+   digits are AN, extended (Farsi/Urdu) digits are EN.
+   0x0606-0x0607, 0x060E-0x060F, 0x06DE, 0x06E9 are ON — fall through. */
+{0x0600, 0x0605, AN},   /* Arabic number signs (Cf) */
+{0x0608, 0x0608, AL},   /* Arabic ray */
+{0x0609, 0x060A, ET},   /* per mille / per ten thousand */
+{0x060B, 0x060B, AL},   /* afghani sign */
+{0x060C, 0x060C, CS},   /* Arabic comma */
+{0x060D, 0x060D, AL},   /* Arabic date separator */
+{0x0610, 0x061A, NSM},  /* honorifics + small marks */
+{0x061B, 0x061F, AL},   /* semicolon, ALM, end-of-text, hamza mark, question mark */
+{0x0620, 0x064A, AL},   /* core Arabic letters + tatweel */
+{0x064B, 0x065F, NSM},  /* harakat (fathatan … wavy hamza below) */
+{0x0660, 0x0669, AN},   /* Arabic-Indic digits ٠-٩ */
+{0x066A, 0x066A, ET},   /* Arabic percent sign */
+{0x066B, 0x066C, AN},   /* decimal / thousands separators */
+{0x066D, 0x066F, AL},   /* five-pointed star, dotless beh/qaf */
+{0x0670, 0x0670, NSM},  /* superscript alef */
+{0x0671, 0x06D5, AL},   /* extended letters: Farsi, Urdu, Sindhi, Pashto, Kurdish */
+{0x06D6, 0x06DC, NSM},  /* Quranic annotation marks */
+{0x06DD, 0x06DD, AN},   /* end of ayah */
+{0x06DF, 0x06E4, NSM},
+{0x06E5, 0x06E6, AL},   /* small waw / small yeh */
+{0x06E7, 0x06E8, NSM},
+{0x06EA, 0x06ED, NSM},
+{0x06EE, 0x06EF, AL},   /* dal/reh with inverted V */
+{0x06F0, 0x06F9, EN},   /* extended Arabic-Indic digits ۰-۹ (Farsi/Urdu) — EN per UCD */
+{0x06FA, 0x06FF, AL},
+
 /* ── Latin Extended Additional (L) ─────────────────────────────────── */
 /* Covers accented chars for Vietnamese, Welsh, Romanian, etc.
    Not currently rendered by CrossPoint fonts, but costs only 2 table rows. */
@@ -110,6 +144,14 @@
 {0x2069, 0x2069, PDI},
 {0x206A, 0x206F, BN},
 
+/* ── Arabic presentation forms (output of do_shape()) ───────────────── */
+/* Contextual/ligature forms emitted by the shaper must classify as AL so
+   a reshaped line still resolves RTL. Ranges per DerivedBidiClass.txt;
+   0xFBC3-0xFBD2 are ON — fall through. */
+{0xFB50, 0xFBC2, AL},   /* Presentation Forms-A: Perso-Arabic contextual forms */
+{0xFBD3, 0xFBFF, AL},   /* Presentation Forms-A: NG … Farsi Yeh forms */
+{0xFE70, 0xFE74, AL},   /* Presentation Forms-B: harakat isolated forms */
+{0xFE76, 0xFEFC, AL},   /* Presentation Forms-B: contextual forms + Lam-Alef ligatures */
 
 /* ── Byte Order Mark ────────────────────────────────────────────────── */
 {0xFEFF, 0xFEFF, BN},