contact-red · redvers · May 28, 2026 · May 28, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -6,6 +6,8 @@ All notable changes to this project will be documented in this file. This projec
 
 ### Fixed
 
+- Grapheme cursor now implements UAX #29 GB9c (Indic_Conjunct_Break, Unicode 15.1). Sequences like Devanagari KA + VIRAMA + TA are now correctly treated as a single cluster. Caught by the new GraphemeBreakTest.txt conformance runner.
+
 
 ### Added
 
@@ -34,8 +36,11 @@ All notable changes to this project will be documented in this file. This projec
 - `Trim` primitive — `trim`, `trim_start`, `trim_end` over the White_Space property
 - `Replace` primitive — `all`, `first`
 - Error types: `InvalidUtf8(offset)`, `OutOfRange(index, size)`, `InvalidScalar(value)`
+- `IndicConjunctBreak` closed union (`InCBNone` / `InCBConsonant` / `InCBLinker` / `InCBExtend`) and `_UcdIndicConjunctBreak` lookup table; surfaced via `Codepoints.indic_conjunct_break`
+- `make conform`: NormalizationTest.txt Part 2 — for every assigned cp not in @Part1 of the test file, verify X == NFC(X) == NFD(X) == NFKC(X) == NFKD(X)
+- `make conform-grapheme`: UAX #29 GraphemeBreakTest.txt conformance (1,093 cases including GB9c)
 - 146 PonyCheck unit tests
-- GitHub Actions CI: pr workflow with lint, changelog verify, and full UAX #15 conformance
+- GitHub Actions CI: pr workflow with lint, changelog verify, and full normalization + grapheme conformance
 
 ### Changed
 
diff --git a/Makefile b/Makefile
@@ -46,7 +46,7 @@ SOURCE_FILES := $(shell find $(SRC_DIR) -name '*.pony')
 
 all: ci
 
-ci: unit-tests conform
+ci: unit-tests conform conform-grapheme
 
 test: unit-tests
 
@@ -76,7 +76,7 @@ $(BUILD_DIR):
 dependencies: corral.json
 	$(GET_DEPENDENCIES_WITH)
 
-.PHONY: all ci test unit-tests clean realclean dependencies docs ucd-build ucd-generate ucd-download conform conform-build
+.PHONY: all ci test unit-tests clean realclean dependencies docs ucd-build ucd-generate ucd-download conform conform-build conform-grapheme conform-grapheme-build
 
 .DEFAULT_GOAL := all
 
@@ -145,3 +145,13 @@ $(conform_binary): $(SOURCE_FILES) $(shell find unicode_conform_main -name '*.po
 
 conform: $(conform_binary)
 	$(conform_binary) $(UCD_DIR)/NormalizationTest.txt
+
+conform_grapheme_binary := $(BUILD_DIR)/unicode_grapheme_conform_main
+
+conform-grapheme-build: $(conform_grapheme_binary)
+
+$(conform_grapheme_binary): $(SOURCE_FILES) $(shell find unicode_grapheme_conform_main -name '*.pony' 2>/dev/null) | $(BUILD_DIR) dependencies
+	$(PONYC) -o $(BUILD_DIR) unicode_grapheme_conform_main -b unicode_grapheme_conform_main
+
+conform-grapheme: $(conform_grapheme_binary)
+	$(conform_grapheme_binary) $(UCD_DIR)/auxiliary/GraphemeBreakTest.txt
diff --git a/unicode/_grapheme_cursor.pony b/unicode/_grapheme_cursor.pony
@@ -15,16 +15,15 @@
 //   GB9b      — Prepend ×
 //   GB11      — emoji ZWJ sequence: Extended_Pictographic Extend* ZWJ ×
 //               Extended_Pictographic
+//   GB9c      — Indic_Conjunct_Break (Unicode 15.1):
+//                 InCB=Consonant [InCB=Extend InCB=Linker]* InCB=Linker
+//                   [InCB=Extend InCB=Linker]* × InCB=Consonant
+//               Suppresses the break before a Consonant when the
+//               history since the last cluster start contains a
+//               Consonant followed by at least one Linker, with only
+//               Extend/Linker codepoints between.
 //   GB12, GB13 — Regional_Indicator pairs
 //   GB999     — otherwise: break
-//
-// Not yet implemented:
-//
-//   GB9c      — Indic conjunct breaks (Unicode 15.1+). Requires the
-//               InCB property table, which lands in a later M1
-//               generator. Until then, text using ZWJ-linked Indic
-//               conjuncts (Devanagari, Bengali, etc.) may produce
-//               extra break points within an intended visual unit.
 
 class ref _GraphemeCursor
   let _bytes: String box
@@ -36,6 +35,8 @@ class ref _GraphemeCursor
                                // (Extend|ZWJ)* on the current cluster?
   var _saw_zwj: Bool           // last codepoint in the cluster was ZWJ
                                // (set after E_Pict Extend*)
+  var _incb_consonant_seen: Bool  // GB9c: seen InCB=Consonant in cluster
+  var _incb_linker_seen: Bool     // GB9c: seen InCB=Linker after consonant
   var _emitted_first: Bool     // has the first cluster been emitted yet?
 
   new ref create(bytes: String box) =>
@@ -46,6 +47,8 @@ class ref _GraphemeCursor
     _ri_run = 0
     _in_emoji_seq = false
     _saw_zwj = false
+    _incb_consonant_seen = false
+    _incb_linker_seen = false
     _emitted_first = false
 
   fun ref next_range(): ((USize, USize) | None) =>
@@ -67,14 +70,24 @@ class ref _GraphemeCursor
     _start = _pos
     _pos = _pos + cp0_len
     let g0 = _UcdGraphemeBreak.of(cp0)
+    let incb0 = _UcdIndicConjunctBreak.of(cp0)
     _reset_state(g0)
+    _reset_incb(incb0)
 
     // Extend the cluster as long as GB rules say "no break."
     while _pos < size do
       (let cp1, let cp1_len) =
         try _decode(_pos)? else (U32(0xFFFD), USize(1)) end
       let g1 = _UcdGraphemeBreak.of(cp1)
-      if _break_between(_prev_break, g1) then
+      let incb1 = _UcdIndicConjunctBreak.of(cp1)
+      var should_break = _break_between(_prev_break, g1)
+      // GB9c override: suppress an otherwise-required break when the
+      // current cluster contains a Consonant + Linker history and the
+      // next codepoint is also a Consonant.
+      if should_break and _gb9c_applies(incb1) then
+        should_break = false
+      end
+      if should_break then
         // Boundary before cp1 — leave it for the next call.
         let range = (_start, _pos)
         _prev_break = g1
@@ -84,6 +97,7 @@ class ref _GraphemeCursor
       end
       // No break: extend the cluster.
       _advance_state(g1)
+      _advance_incb(incb1)
       _pos = _pos + cp1_len
     end
     // Reached end of input — emit the final cluster.
@@ -106,6 +120,55 @@ class ref _GraphemeCursor
       else 0
       end
 
+  fun ref _reset_incb(incb: IndicConjunctBreak) =>
+    """
+    Initialize InCB tracking when a new cluster's first codepoint is
+    `incb`. Only a Consonant primes the GB9c state machine; any other
+    starting value leaves both flags off.
+    """
+    match incb
+    | InCBConsonant =>
+      _incb_consonant_seen = true
+      _incb_linker_seen = false
+    else
+      _incb_consonant_seen = false
+      _incb_linker_seen = false
+    end
+
+  fun ref _advance_incb(incb: IndicConjunctBreak) =>
+    """
+    Update InCB tracking when a codepoint joins the current cluster.
+    Transitions (state = (consonant_seen, linker_seen)):
+      Consonant: → (true, false)   — a new Consonant restarts the search
+      Linker   : if consonant_seen, → (true, true)
+      Extend   : keep state
+      None     : break the chain → (false, false)
+    """
+    match incb
+    | InCBConsonant =>
+      _incb_consonant_seen = true
+      _incb_linker_seen = false
+    | InCBLinker =>
+      if _incb_consonant_seen then _incb_linker_seen = true end
+    | InCBExtend =>
+      None
+    | InCBNone =>
+      _incb_consonant_seen = false
+      _incb_linker_seen = false
+    end
+
+  fun _gb9c_applies(next: IndicConjunctBreak): Bool =>
+    """
+    True iff GB9c says to suppress the break before `next`. Active
+    only when next is Consonant and the current cluster's history
+    includes Consonant + ... + Linker (Extends and additional
+    Linkers allowed between).
+    """
+    match next
+    | InCBConsonant => _incb_consonant_seen and _incb_linker_seen
+    else false
+    end
+
   fun ref _advance_state(g: GraphemeBreak) =>
     """
     Update tracking state after extending the current cluster with a

diff --git a/unicode/_ucd_indic_conjunct_break.pony b/unicode/_ucd_indic_conjunct_break.pony
@@ -0,0 +1,39 @@
+// Auto-generated by unicode-build. Do not edit by hand.
+// Regenerate via:
+//   make ucd-generate
+//
+// Source: DerivedCoreProperties.txt InCB entries.
+// Codepoints outside any range default to InCBNone.
+
+primitive _UcdIndicConjunctBreak
+  fun of(cp: U32): IndicConjunctBreak =>
+    let t = _table()
+    var lo: USize = 0
+    var hi: USize = t.size() / 18
+    while lo < hi do
+      let mid = lo + ((hi - lo) / 2)
+      let base = mid * 18
+      try
+        let range_lo: U32 =
+          _UcdHex.byte(t, base)?
+            or (_UcdHex.byte(t, base + 2)? << 8)
+            or (_UcdHex.byte(t, base + 4)? << 16)
+            or (_UcdHex.byte(t, base + 6)? << 24)
+        let range_hi: U32 =
+          _UcdHex.byte(t, base + 8)?
+            or (_UcdHex.byte(t, base + 10)? << 8)
+            or (_UcdHex.byte(t, base + 12)? << 16)
+            or (_UcdHex.byte(t, base + 14)? << 24)
+        if cp < range_lo then hi = mid
+        elseif cp > range_hi then lo = mid + 1
+        else
+          return IndicConjunctBreaks._from_byte(
+            U8.from[U32](_UcdHex.byte(t, base + 16)?))
+        end
+      else return InCBNone
+      end
+    end
+    InCBNone
+
+  fun _table(): String val =>
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
diff --git a/unicode/indic_conjunct_break.pony b/unicode/indic_conjunct_break.pony
@@ -0,0 +1,61 @@
+// `Indic_Conjunct_Break` property values (UAX #44, added Unicode 15.1).
+//
+// Used by UAX #29 rule GB9c to suppress cluster breaks within Indic
+// conjuncts of the form:
+//
+//   Consonant [Extend|Linker]* Linker [Extend|Linker]* × Consonant
+//
+// The four values match DerivedCoreProperties.txt's `InCB` property:
+//
+//   None       — default; codepoint isn't involved in Indic conjunct
+//                linking.
+//   Consonant  — a Linking Consonant (Devanagari KA, Bengali KA, ...).
+//   Linker     — a Conjunct Linker (DEVANAGARI SIGN VIRAMA, etc.).
+//   Extend     — Extend-class cp valid inside a conjunct (NUKTA, ZWJ).
+//
+// Like the other closed unions in this package, exhaustive `match`
+// over `IndicConjunctBreak` is compile-checked.
+
+primitive InCBNone
+  fun code(): String val => "None"
+  fun string(): String val => "None"
+
+primitive InCBConsonant
+  fun code(): String val => "Consonant"
+  fun string(): String val => "Consonant"
+
+primitive InCBLinker
+  fun code(): String val => "Linker"
+  fun string(): String val => "Linker"
+
+primitive InCBExtend
+  fun code(): String val => "Extend"
+  fun string(): String val => "Extend"
+
+type IndicConjunctBreak is (InCBNone | InCBConsonant | InCBLinker | InCBExtend)
+
+primitive IndicConjunctBreaks
+  fun from_iso(s: String box): (IndicConjunctBreak | None) =>
+    match s
+    | "None" => InCBNone
+    | "Consonant" => InCBConsonant
+    | "Linker" => InCBLinker
+    | "Extend" => InCBExtend
+    else None
+    end
+
+  fun _to_byte(b: IndicConjunctBreak): U8 =>
+    match b
+    | InCBNone      => 0
+    | InCBConsonant => 1
+    | InCBLinker    => 2
+    | InCBExtend    => 3
+    end
+
+  fun _from_byte(b: U8): IndicConjunctBreak =>
+    match b
+    | 1 => InCBConsonant
+    | 2 => InCBLinker
+    | 3 => InCBExtend
+    else InCBNone
+    end