Virama conjunct+Mc width fix

bin/update-tables.py

@@ -798,7 +798,10 @@ def parse_indic_conjunct_breaks(fname: str) -> dict[str, TableDef]:
     }
 
 
-ISC_VALUES = ('Consonant',)
+# ISC_CONSONANT is generated for the public API but not used by the runtime.
+# Virama and Invisible_Stacker are treated as zero-width combining marks (Mn),
+# matching Ghostty's uucode; the runtime no longer consults the ISC categories.
+ISC_VALUES = ('Consonant', 'Virama', 'Invisible_Stacker')
 
 
 def parse_indic_syllabic_category(fname: str) -> dict[str, TableDef]:

docs/intro.rst

@@ -506,10 +506,15 @@ This library is used in:
 Other Languages
 ===============
 
-There are similar implementations of the `wcwidth()`_ and `wcswidth()`_ functions in other
-languages.
+The following libraries provide grapheme and emoji support and closely align with our
+specification_:
+
+- `jacobsandlund/uucode`_ Zig
+- `contour-terminal/libunicode`_ C++20
+
+There are similar implementations of at least the `wcwidth()`_ and `wcswidth()`_ functions in other
+languages:
 
-- `contour-terminal/libunicode`_: C++20
 - `ridiculousfish/widecharwidth`_: Python
 - `termux/wcwidth`_: C
 - `powerman/wcwidth-icons`_: C
@@ -534,8 +539,10 @@ History
 =======
 
 0.8.0 *(unreleased)*
-  * **Improved** memory resource usage and import time for Python 3.15 using lazy imports `PR #221`_.
+  * **Improved** memory usage and import time for Python 3.15 using lazy imports `PR #221`_.
   * **Improved** performance on Python 3.15 using standard library iter_graphemes() `PR #206`_.
+  * **Bugfix** Invisible_Stacker viramas now form conjuncts (Burmese, Khmer, etc.) and
+    change some Virama width calculations to match `jacobsandlund/uucode`_ (ghostty) `PR #XXX`_.
 
 0.7.0 *2026-05-02*
   * **New** support for `kitty text sizing protocol`_ (OSC 66) in `width()`_ and `clip()`_.
@@ -799,6 +806,7 @@ https://www.cl.cam.ac.uk/~mgk25/ucs/wcwidth.c::
 .. _`joshuarubin/wcwidth9`: https://github.com/joshuarubin/wcwidth9
 .. _`spectreconsole/wcwidth`: https://github.com/spectreconsole/wcwidth
 .. _`contour-terminal/libunicode`: https://github.com/contour-terminal/libunicode
+.. _`jacobsandlund/uucode`: https://github.com/jacobsandlund/uucode
 .. _`ridiculousfish/widecharwidth`: https://github.com/ridiculousfish/widecharwidth
 .. _`termux/wcwidth`: https://github.com/termux/wcwidth
 .. _`powerman/wcwidth-icons`: https://github.com/powerman/wcwidth-icons

docs/specs.rst

@@ -114,15 +114,19 @@ by a Nukta (``Mn``) and then a vowel sign (``Mc``) is measured as base + 1.
 Virama Conjunct Formation
 -------------------------
 
-In `Brahmic scripts`_, a `Virama`_ (``Indic_Syllabic_Category=Virama`` in
-`IndicSyllabicCategory.txt`_) between two consonants triggers `conjunct`_
-formation: the consonants are merged into a single ligature glyph.
+In `Brahmic scripts`_, `IndicSyllabicCategory.txt`_ defines two categories
+that trigger `conjunct`_ formation between consonants: `Virama`_ ("may act
+as a Pure_Killer or Invisible_Stacker depending on context") and
+``Invisible_Stacker``_ ("not visible by itself; causes conjunct formation
+or consonant stacking", the "only as consonant stackers" category
+described in the Virama section header).
 
 - A ``Consonant`` immediately following a ``Virama`` contributes 0 width.
 - The conjunct still occupies cells and the next visible advance settles it:
 
-  - A following ``Mc`` (`Spacing Combining Mark`_, e.g. a vowel sign) counts as
-    1 cell and closes the conjunct.
+  - A following ``Mc`` (`Spacing Combining Mark`_, e.g. a vowel sign) closes
+    the conjunct and accounts for its 1-cell contribution.  The ``Mc`` itself
+    adds no independent width when following a conjunct.
   - A following character with positive width (or end of string) adds 1 cell
     for the conjunct before counting its own width.
 
@@ -176,6 +180,7 @@ See also: `L2/2023/23107`_ "Proper Complex Script Support in Text Terminals".
 .. _`Nonspacing Mark`: https://www.unicode.org/versions/latest/core-spec/chapter-4/#G134153
 .. _`IndicSyllabicCategory.txt`: https://www.unicode.org/Public/UCD/latest/ucd/IndicSyllabicCategory.txt
 .. _`Indic_Syllabic_Category`: https://www.unicode.org/reports/tr44/#Indic_Syllabic_Category
+.. _`Invisible_Stacker`: https://www.unicode.org/Public/UCD/latest/ucd/IndicSyllabicCategory.txt
 .. _`Brahmic scripts`: https://en.wikipedia.org/wiki/Brahmic_scripts
 .. _`Virama`: https://www.unicode.org/glossary/#virama
 .. _`conjunct`: https://www.unicode.org/glossary/#consonant_conjunct

tests/test_core.py

@@ -308,7 +308,7 @@ def test_devanagari_script():
               "\u093F")   # MatraL, Category 'Mc', East Asian Width property 'N' -- DEVANAGARI VOWEL SIGN I
     # 23107-terminal-suppt.pdf suggests wcwidth.wcwidth should return (2, 0, 0, 1)
     expect_length_each = (1, 0, 1, 0)
-    # virama conjunct collapses KA+virama+SSA into one cell, Mc adds +1
+    # virama conjunct collapses KA+virama+SSA into one cell, Mc vowel sign is part of the conjunct
     expect_length_phrase = 2
 
     # exercise,
@@ -330,7 +330,7 @@ def test_tamil_script():
     # 23107-terminal-suppt.pdf suggests wcwidth.wcwidth should return (3, 0, 0, 4)
     expect_length_each = (1, 0, 1, 0)
 
-    # virama conjunct collapses KA+virama+SSA into one cell, Mc adds +1
+    # virama conjunct collapses KA+virama+SSA into one cell, Mc vowel sign is part of the conjunct
     expect_length_phrase = 2
 
     # exercise,
@@ -353,7 +353,7 @@ def test_kannada_script():
               "\u0cc8")   # MatraUR, Category 'Mc', East Asian Width property 'N' -- KANNADA VOWEL SIGN AI
     # 23107-terminal-suppt.pdf suggests should be (2, 0, 3, 1)
     expect_length_each = (1, 0, 1, 0)
-    # virama conjunct collapses RA+virama+JHA into one cell, Mc adds +1
+    # virama conjunct collapses RA+virama+JHA into one cell, Mc vowel sign is part of the conjunct
     expect_length_phrase = 2
 
     # exercise,
@@ -441,7 +441,7 @@ def test_mc_width_consistency(repeat):
     ("\u0915\u094D\u0924\u093F", 2),
     ("\u0915\u094D\u0930\u093F", 2),
     ("\u0A95\u0ACD\u0A95\u0ACB", 2),
-    ("\u0938\u094D\u0924\u094D\u0930", 2),
+    ("\u0938\u094D\u0924\u094D\u0930", 3),
     ("\u0938\u094D\u0924", 2),
     ("\u0915\u094D\u0020", 2),
     ("\u09A4\u09CD\u200D\u09AA", 2),
@@ -454,6 +454,30 @@ def test_virama_conjunct(phrase, expected):
     assert wcwidth.width(phrase) == expected
 
 
+@pytest.mark.parametrize("phrase,expected", [
+    ("\u0995\u09CD\u09A4\u09BF", 2),       # Bengali C+V+C+Mc: ক্তি
+    ("\u0915\u094D\u0924\u093F", 2),       # Devanagari C+V+C+Mc: क्ति
+    ("\u0995\u09CD\u09A4", 2),             # C+V+C (no Mc), unchanged
+    ("\u0995\u09BF", 2),                   # C+Mc (no virama), unchanged
+])
+def test_virama_conjunct_mc_vowel(phrase, expected):
+    """Vowel sign (Mc) following a virama conjunct does not add independent width."""
+    assert wcwidth.wcswidth(phrase) == expected
+    assert wcwidth.width(phrase) == expected
+
+
+@pytest.mark.parametrize("phrase,expected", [
+    ("\u1000\u1039\u1000", 2),             # Burmese KA+VIRAMA+KA
+    ("\u1000\u1039\u1000\u1039\u1002", 3),  # Burmese KA+V+KA+V+GA
+    ("\u1000\u1039\u200D\u1000", 2),       # Burmese KA+V+ZWJ+KA
+    ("\u1782\u17D2\u1782\u17C1", 2),       # Khmer KO+COENG+KO+VOWEL_E (Mc)
+    ("\u1780\u17D2\u1780", 2),             # Khmer KA+COENG+KA
+])
+def test_virama_conjunct_invisible_stacker(phrase, expected):
+    assert wcwidth.wcswidth(phrase) == expected
+    assert wcwidth.width(phrase) == expected
+
+
 def test_zwj_at_end_of_string():
     """ZWJ at end of string (not after virama) is consumed with zero width."""
     assert wcwidth.wcswidth('a\u200D') == 1

tox.ini

@@ -208,7 +208,7 @@ warn_unused_ignores = true
 [testenv:codespell]
 basepython = python3.14
 deps = codespell
-commands = codespell --skip="*.pyc,htmlcov,_build,build,*.egg-info,.tox,data,./tests/*.txt,*.csv,*.ods,table_*.py,docs/specs.rst,*.isorted" \
+commands = codespell --skip="*.pyc,htmlcov,_build,build,*.egg-info,.tox,data,./tests/*.txt,*.csv,*.ods,table_*.py,docs/specs.rst,*.isorted,ucs-detect" \
                      --ignore-words-list="thirdparty,claus,oclock,womens,aprox" \
                      --uri-ignore-words-list '*' \
                      --summary --count

wcwidth/_constants.py

@@ -4,7 +4,10 @@
 from .table_mc import CATEGORY_MC
 from .table_wide import WIDE_EASTASIAN
 from .table_zero import ZERO_WIDTH
-from .table_grapheme import EXTENDED_PICTOGRAPHIC, GRAPHEME_REGIONAL_INDICATOR
+from .table_grapheme import (ISC_VIRAMA,
+                             EXTENDED_PICTOGRAPHIC,
+                             ISC_INVISIBLE_STACKER,
+                             GRAPHEME_REGIONAL_INDICATOR)
 from .table_ambiguous import AMBIGUOUS_EASTASIAN
 from .unicode_versions import list_versions
 
@@ -23,35 +26,10 @@
 _REGIONAL_INDICATOR_SET = frozenset(
     range(GRAPHEME_REGIONAL_INDICATOR[0][0], GRAPHEME_REGIONAL_INDICATOR[0][1] + 1)
 )
-_ISC_VIRAMA_SET = frozenset((
-    0x094D,   # DEVANAGARI SIGN VIRAMA
-    0x09CD,   # BENGALI SIGN VIRAMA
-    0x0A4D,   # GURMUKHI SIGN VIRAMA
-    0x0ACD,   # GUJARATI SIGN VIRAMA
-    0x0B4D,   # ORIYA SIGN VIRAMA
-    0x0BCD,   # TAMIL SIGN VIRAMA
-    0x0C4D,   # TELUGU SIGN VIRAMA
-    0x0CCD,   # KANNADA SIGN VIRAMA
-    0x0D4D,   # MALAYALAM SIGN VIRAMA
-    0x0DCA,   # SINHALA SIGN AL-LAKUNA
-    0x1B44,   # BALINESE ADEG ADEG
-    0xA806,   # SYLOTI NAGRI SIGN HASANTA
-    0xA8C4,   # SAURASHTRA SIGN VIRAMA
-    0xA9C0,   # JAVANESE PANGKON
-    0x11046,  # BRAHMI VIRAMA
-    0x110B9,  # KAITHI SIGN VIRAMA
-    0x111C0,  # SHARADA SIGN VIRAMA
-    0x11235,  # KHOJKI SIGN VIRAMA
-    0x1134D,  # GRANTHA SIGN VIRAMA
-    0x11442,  # NEWA SIGN VIRAMA
-    0x114C2,  # TIRHUTA SIGN VIRAMA
-    0x115BF,  # SIDDHAM SIGN VIRAMA
-    0x1163F,  # MODI SIGN VIRAMA
-    0x116B6,  # TAKRI SIGN VIRAMA
-    0x11839,  # DOGRA SIGN VIRAMA
-    0x119E0,  # NANDINAGARI SIGN VIRAMA
-    0x11C3F,  # BHAIKSUKI SIGN VIRAMA
-))
+_ISC_VIRAMA_SET = frozenset(
+    cp for lo, hi in (*ISC_VIRAMA, *ISC_INVISIBLE_STACKER)
+    for cp in range(lo, hi + 1)
+)
 # pylint: disable=invalid-name
 _LATEST_VERSION = list_versions()[-1]
 _CATEGORY_MC_TABLE = CATEGORY_MC[_LATEST_VERSION]

wcwidth/_wcswidth.py

@@ -8,7 +8,6 @@
     "wcwidth._constants",
     "wcwidth._wcwidth",
     "wcwidth.bisearch",
-    "wcwidth.table_grapheme",
     "wcwidth.table_vs16",
 ]
 # local
@@ -74,26 +73,30 @@ def wcswidth(
     # grapheme-clustering state
     last_measured_idx = -2
     last_measured_ucs = -1
-    last_was_virama = False
+    prev_was_virama = False
     conjunct_pending = False
 
     while idx < end:
         char = pwcs[idx]
         ucs = ord(char)
-
-        # ZWJ (U+200D)
+        #
+        # Much of the logic below matches the logic in width(), but is repeated for improved
+        # performance, they are given matching index reference numbers (starting at #6).
+        #
+        # 5. ZWJ (U+200D): consumed without contributing width.
+        # Virama codepoints are treated as zero-width combining marks (Mn), matching
+        # Ghostty's uucode library.  Whether a virama+consonant sequence should
+        # collapse to 1 or 2 cells is undefined by Unicode; we follow uucode's (1).
         if ucs == 0x200D:
-            if last_was_virama:
+            if prev_was_virama:
                 idx += 1
             elif idx + 1 < end:
-                last_was_virama = False
                 idx += 2
             else:
-                last_was_virama = False
                 idx += 1
             continue
 
-        # VS16 (U+FE0F): converts preceding narrow character to wide.
+        # 6. VS16 (U+FE0F): converts preceding narrow character to wide.
         if ucs == 0xFE0F and last_measured_idx >= 0:
             total_width += bisearch(
                 ord(pwcs[last_measured_idx]),
@@ -103,7 +106,7 @@ def wcswidth(
             idx += 1
             continue
 
-        # Regional Indicator & Fitzpatrick (both above BMP)
+        # 7. Regional Indicator & Fitzpatrick (both above BMP)
         if ucs > 0xFFFF:
             if ucs in _REGIONAL_INDICATOR_SET:
                 ri_before = 0
@@ -121,15 +124,17 @@ def wcswidth(
                 continue
 
         # Virama conjunct formation
-        if last_was_virama and bisearch(ucs, ISC_CONSONANT):
+        if prev_was_virama and bisearch(ucs, ISC_CONSONANT):
+            if conjunct_pending:
+                total_width += 1
             last_measured_idx = idx
             last_measured_ucs = ucs
-            last_was_virama = False
+            prev_was_virama = False
             conjunct_pending = True
             idx += 1
             continue
 
-        # Normal character: measure with wcwidth
+        # 8. Normal character: measure with wcwidth
         w = _wcwidth(char)
         if w < 0:
             # C0/C1 control character
@@ -141,15 +146,18 @@ def wcswidth(
             total_width += w
             last_measured_idx = idx
             last_measured_ucs = ucs
-            last_was_virama = False
+            prev_was_virama = False
         elif last_measured_idx >= 0 and bisearch(ucs, _CATEGORY_MC_TABLE):
-            # Spacing Combining Mark (Mc) following a base character adds 1
-            total_width += 1
+            # Spacing Combining Mark (Mc) following a base character
+            if conjunct_pending:
+                total_width += 1
+                conjunct_pending = False
+            else:
+                total_width += 1
             last_measured_idx = -2
-            last_was_virama = False
-            conjunct_pending = False
+            prev_was_virama = False
         else:
-            last_was_virama = ucs in _ISC_VIRAMA_SET
+            prev_was_virama = ucs in _ISC_VIRAMA_SET
         idx += 1
 
     if conjunct_pending: