applib: render bidirectional text with the Unicode bidi algorithm

src/fw/applib/fonts/codepoint.c

@@ -218,8 +218,3 @@ bool codepoint_is_special(const Codepoint codepoint) {
   return (codepoint >= MIN_SPECIAL_CODEPOINT &&
           codepoint <= MAX_SPECIAL_CODEPOINT);
 }
-
-bool codepoint_is_rtl(const Codepoint codepoint) {
-  return (codepoint >= MIN_ARABIC_CODEPOINT && codepoint <= MAX_ARABIC_CODEPOINT) ||
-         (codepoint >= MIN_HEBREW_CODEPOINT && codepoint <= MAX_HEBREW_CODEPOINT);
-}

src/fw/applib/fonts/codepoint.h

@@ -70,6 +70,3 @@ Codepoint emoji_shape_pair(const Codepoint curr_cp, const Codepoint next_cp,
 // This is a least dirty hack to enable special rendering when a special codepoint is hit in the
 // text being rendered
 bool codepoint_is_special(const Codepoint codepoint);
-
-// Check if a codepoint is from a right-to-left script (Arabic, Hebrew)
-bool codepoint_is_rtl(const Codepoint codepoint);

src/fw/applib/graphics/arabic_shaping.c

@@ -5,10 +5,6 @@
 
 #include <string.h>
 
-// Caps codepoints per shaping call. Sized to match walk_line's 128-byte
-// shape buffer (~64 Arabic codepoints at 2 UTF-8 bytes each).
-#define MAX_SHAPE_CODEPOINTS 64
-
 // Connectivity flags for Arabic letters
 #define JOIN_NONE       0x00  // Does not connect (space, punctuation)
 #define JOIN_RIGHT_ONLY 0x01  // Connects only to the right (non-connecting letters)
@@ -304,13 +300,13 @@ bool arabic_is_transparent(Codepoint cp) {
 }
 
 size_t arabic_shape_text(const utf8_t *src, size_t src_len,
-                         utf8_t *dest, size_t dest_size) {
-  if (src == NULL || dest == NULL || src_len == 0 || dest_size == 0) {
+                         utf8_t *dest, size_t dest_size, Codepoint *cp_scratch) {
+  if (src == NULL || dest == NULL || cp_scratch == NULL || src_len == 0 || dest_size == 0) {
     return 0;
   }
 
-  // First pass: collect all codepoints into an array
-  Codepoint codepoints[MAX_SHAPE_CODEPOINTS];
+  // First pass: collect all codepoints into the caller's scratch array
+  Codepoint *codepoints = cp_scratch;
   size_t num_codepoints = 0;
 
   utf8_t *ptr = (utf8_t *)src;
@@ -355,7 +351,9 @@ size_t arabic_shape_text(const utf8_t *src, size_t src_len,
       }
     }
 
-    // Lam-Alef collapses two letters into one ligature glyph.
+    // Lam-Alef collapses two letters into one ligature glyph. Emoji pairs are
+    // NOT folded here: regional indicators must reach the bidi engine as their
+    // UCD class L, so the draw pass folds them after reordering.
     bool consumed_next = false;
     Codepoint shaped_cp = arabic_shape_pair(prev_cp, curr_cp, next_cp, &consumed_next);
 

src/fw/applib/graphics/arabic_shaping.h

@@ -48,6 +48,11 @@ Codepoint arabic_shape_codepoint(Codepoint prev_cp, Codepoint curr_cp, Codepoint
 Codepoint arabic_shape_pair(Codepoint prev_cp, Codepoint curr_cp, Codepoint next_cp,
                             bool *consumed_next);
 
+//! Caps codepoints per arabic_shape_text() call. The bidi render path shapes
+//! per codepoint with paragraph context and no longer uses this helper; the
+//! cap covers its remaining (test and utility) callers.
+#define MAX_SHAPE_CODEPOINTS 96
+
 //! Shape Arabic text by converting basic Arabic letters to their
 //! contextual presentation forms based on position in words.
 //!
@@ -59,7 +64,10 @@ Codepoint arabic_shape_pair(Codepoint prev_cp, Codepoint curr_cp, Codepoint next
 //! @param src_len Length of source string in bytes
 //! @param dest Destination buffer for shaped text
 //! @param dest_size Size of destination buffer in bytes
+//! @param cp_scratch Working array for at least MAX_SHAPE_CODEPOINTS
+//!        codepoints. Caller-provided so the render path can keep it off the
+//!        2-4 KiB task stacks (it reuses the bidi heap scratch).
 //! @return Number of bytes written to dest (excluding null terminator),
 //!         or 0 on failure
 size_t arabic_shape_text(const utf8_t *src, size_t src_len,
-                         utf8_t *dest, size_t dest_size);
+                         utf8_t *dest, size_t dest_size, Codepoint *cp_scratch);