v1.2.0

1.2.0 (2026-11-05)

Added

• muniSegmentIterator_FindBreaksAround gives the boundaries around a
  byte offset, the last at or before it and the first after it, without
  reading the text from its start: what Backspace, the arrow keys and a
  click on a word need. It answers as the iterator would, for grapheme
  clusters, words, sentences and lines at every setting, class
  overrides and a word segmenter included, and leaves the iterator as
  it was. In ordinary text it reads a few code points, so on a long
  paragraph of English prose it takes under half a microsecond for
  clusters, words and lines, where stepping a fresh iterator from the
  start takes milliseconds; sentences back up to the last terminator.
  It needs an iterator given its whole text at init
  (#278).

No results change: the Unicode data is still 18.0.0 and
MUNI_RESULTS_VERSION stays 2, so stored normalized keys, skeletons
and verdicts from 1.1.2 stay valid.

v1.1.2

1.1.2 (2026-11-05)

Fixed

• Every internal function that returns bool now carries
  MUNI_NODISCARD, so a compiler warns where the library would drop
  one of their answers, and tools/check_nodiscard.py keeps new ones
  marked. No call dropped one, so nothing a caller sees changes: same
  API, same behavior. The header tests now also check that
  MUNI_NODISCARD is not empty in C23
  (#276).

No results change: the Unicode data is still 18.0.0 and
MUNI_RESULTS_VERSION stays 2, so stored normalized keys, skeletons
and verdicts from 1.1.1 stay valid.

v1.1.1

1.1.1 (2026-11-04)

Fixed

• MUNI_NODISCARD was empty under GCC 14 in C23, since GCC 14 reports
  its C23 as __STDC_VERSION__ 202000L and the macro waited for
  202311L, so a program built with GCC 14 could drop any status the
  library returns without a warning. In C the macro now takes
  [[nodiscard]] in any dialect after C17 whose compiler has the
  attribute. A program that drops a status now gets
  -Wunused-result from GCC 14 too, which -Werror makes an error
  (#271).

No results change: the Unicode data is still 18.0.0 and
MUNI_RESULTS_VERSION stays 2, so stored normalized keys, skeletons
and verdicts from 1.1.0 stay valid.

v1.1.0

1.1.0 (2026-11-04)

Added

• Every call that takes a muniSegmentIterator or a
  muniScriptIterator back now checks it first, and refuses one no
  init call set up with muni_errorInvalid, leaving it as it was: a
  zeroed iterator, one of stray bytes, or one whose counts, flags or
  states are out of range. Such an iterator was undefined behavior
  before, and a forged script iterator could overflow a buffer on the
  stack. muniSegmentIterator_GetLineBreakStrictness and
  muniSegmentIterator_GetWritingSystem give their defaults for it, as
  for NULL. A copy of an initialized iterator is accepted. The
  pointers an iterator keeps (the text, the overrides, the breaker)
  cannot be checked, so an iterator changed there by hand stays outside
  the contract
  (#264,
  #267).

Changed

• The check costs each iterator call a few nanoseconds: line breaking
  over English prose takes 1 to 2% longer, and up to 20% where every
  call reports a boundary two code points after the last. Neither
  iterator changes size
  (#264).

No results change: the Unicode data is still 18.0.0 and
MUNI_RESULTS_VERSION stays 2, so stored normalized keys, skeletons
and verdicts from 1.0.0 stay valid.

v1.0.0

1.0.0 (2026-11-03)

The API is settled. 1.0.0 has the API and the results of 0.13.0; what
changes is what a caller may rely on from here. The promise is in
What 1.0.0 promises.

Changed

• Within 1.x no function, type, enum value, field or macro of the
  eleven public headers, harfbuzz.h included, is removed or renamed,
  and no release is breaking: a change that would break 1.x is made
  additive or does not ship. There will be no 2.0.0
  (#257).
• The numbers of the property enumerations in properties.h are part
  of the API, so data may store them; a value a new Unicode version
  adds goes after the last
  (#257).
• The CMake packages maul-unicode and maul-unicode-harfbuzz accept
  a request for any earlier 1.y, where 0.x accepted only its own
  version; the shared library's SOVERSION is 1.0, so a program
  built against 1.0 loads any 1.0.z and is built again for 1.1
  (#248).

No results change: the Unicode data is still 18.0.0 and
MUNI_RESULTS_VERSION stays 2, so stored normalized keys, skeletons
and verdicts from 0.13.0 stay valid.

v0.13.0

0.13.0 (2026-11-02)

Added

• muniMeasureUtf8 finds the longest prefix of UTF-8 text that ends at
  a boundary of a muniTextUnit and holds at most maxBytes bytes and
  at most maxUnits units, and gives its length in bytes and in units;
  either output may be NULL. It turns a byte offset into a count of
  units, turns a count of units into a byte offset, cuts text to fit a
  buffer, and counts the grapheme clusters of a whole text or finds
  where a count of them ends. A limit that falls inside a code point or
  a cluster ends the prefix at the boundary before it, and a count of
  UTF-16 units that falls inside a surrogate pair ends it before the
  pair. A call that fits whole and asks for no count does not walk the
  text
  (#243).
• muni_unitUtf16, a new muniTextUnit, counts UTF-16 units: one for
  a code point up to U+FFFF, two for one past it, and one for a maximal
  ill-formed subpart, as its U+FFFD. With muniMeasureUtf8 it maps
  UTF-8 byte offsets to the UTF-16 offsets that AppKit, UIKit and
  Android take, and back
  (#243).

Changed

• Breaking: on muni_errorCapacity, muniConvertUtf8ToUtf16,
  muniConvertUtf16ToUtf8, muniConvertUtf8ToUtf32 and
  muniConvertUtf32ToUtf8 give the input offset of the first code
  point not written, instead of the input length. The output holds the
  conversion of the input before that offset, always whole code points,
  and neededOut still counts the whole conversion. A caller can keep
  what was written, or convert the rest from the offset, instead of
  converting again with less input until it fits. A strict-mode error
  after the output filled up still returns that error and its offset.
  Code that read the input length from the offset on
  muni_errorCapacity uses its own length
  (#241).

Removed

• Breaking: muniCutUtf8 is removed. muniMeasureUtf8 with
  maxUnits set to SIZE_MAX gives the same cut; the unit now comes
  before the byte limit. No consumer called it
  (#243).

  0.12.0	0.13.0
  muniCutUtf8(text, length, budget, unit, &cut)	muniMeasureUtf8(text, length, unit, budget, SIZE_MAX, &cut, NULL)

v0.12.0

0.12.0 (2026-11-02)

Added

• muniSegmentIterator_GetWordKind tells what the segment ending at the
  boundary a word iterator last reported holds, as a muniWordKind: no
  word (spaces, punctuation, symbols), a number, letters, kana or
  ideographs, the highest kind among its code points. Word movement and
  double-click selection skip segments of kind muni_wordNone. Letters
  with digits are letters, and a superscript digit such as "²" makes no
  number. ICU's word rule statuses agree on segments of letters or
  digits alone; on a mixed segment ICU reports the last rule it matched,
  so it calls "ab12" a number
  (#231).

Changed

• Word breaking executes 4.5% to 6.7% more instructions, to keep the
  kind of each segment
  (#231).

v0.11.0

0.11.0 (2026-11-02)

Added

• muniGetDecomposition writes the full decomposition of one code
  point, canonical (muni_nfd) or compatibility (muni_nfkd), every
  mapping applied, in canonical order; Hangul syllables give their jamo
  and a code point that does not decompose gives itself. It reports the
  count it needs, with muni_errorCapacity when the output is too
  small. MUNI_MAX_DECOMPOSITION (18, for U+FDFA in NFKD), now public
  in normalize.h, always suffices for this Unicode version; a release
  with a new Unicode version may raise it
  (#227).
• muniDecodeUtf16, the counterpart of muniDecodeUtf8, decodes the
  code point at the start of UTF-16 units: a surrogate pair takes two
  units, anything else one. An unpaired surrogate gives
  muni_errorUtf16Surrogate with U+FFFD and one unit, as
  muniConvertUtf16ToUtf8 replaces it, so a decoding loop always makes
  progress
  (#228).
• muniCutUtf8 gives the length of the longest prefix of UTF-8 text
  within a byte budget that ends at a boundary of a unit:
  muni_unitCodePoint, where each maximal ill-formed subpart counts as
  one, or muni_unitGraphemeCluster, which keeps a letter with its
  accents, an emoji ZWJ sequence or a flag whole, for text fields and
  preedit strings. The unit is an enumeration so that a later release
  can add words
  (#230).

v0.10.0

0.10.0 (2026-11-02)

⚠ BREAKING CHANGES

• Line iterators take a strictness level and a writing system in place
  of the CJ switch. muniSegmentIterator_EnableCjAsIdeographic(iterator, true) becomes muniSegmentIterator_SetLineBreakStrictness(iterator, muni_lineBreakLoose), which also adds loose's other breaks.
  muniSegmentIterator_IsCjAsIdeographicEnabled is removed;
  muniSegmentIterator_GetLineBreakStrictness reports the level. A new
  iterator breaks as before
  (#221).

Added

• Line iterators break at the levels of CSS's line-break property:
  strict, normal (the default), loose and anywhere, set with
  muniSegmentIterator_SetLineBreakStrictness before the first break.
  muniSegmentIterator_SetWritingSystem with
  muni_writingChineseJapanese turns on the rules CSS keeps for Chinese
  and Japanese. Loose breaks before small kana and the prolonged sound
  mark (Line_Break=CJ), before iteration marks and between characters
  such as two ellipses; anywhere breaks between any two grapheme
  clusters. Normal and strict keep CJ with what precedes it, as the CSS
  Text 3 Candidate Recommendation Draft of 14 August 2026 says; ICU4C's
  normal rules and earlier CSS drafts break before CJ in normal. The
  default level breaks as before, at nearly the same cost (0.15% more
  instructions on the bench)
  (#221).
• muniConvertCase takes two more languages. muni_caseGreek
  uppercases Greek as it is written in capitals: accents and breathings
  go, a dialytika stays and appears where a dropped accent kept two
  vowels apart ("Μάιος" to "ΜΑΪΟΣ"), the disjunctive eta "ή" keeps its
  tonos when it stands alone, and an iota subscript becomes a capital
  iota; it follows ICU's Greek rules. muni_caseArmenian uppercases
  the ligature "և" to "ԵՎ" and titlecases it to "Եվ". Lowercasing,
  titlecasing (but the ligature) and folding follow the default rules
  (#223).

Changed

• The results version is 2: calls to muniConvertCase with
  muni_caseGreek or muni_caseArmenian, which were refused, now give
  results. No result of the earlier languages changes, so stored keys
  and skeletons stay valid; a caller that stores the results version
  sees it move once
  (#223).

v0.9.1

0.9.1 (2026-11-01)

Changed

• The headers now say what two more refusals leave.
  muniReorderBidiLine leaves runsOut and countOut as they were on
  muni_errorInvalid, and muniGetResolvedScripts leaves scriptsOut
  and countOut as they were on muni_errorInvalid and on a UTF-8
  error. Both calls always behaved this way
  (#204).
• base.h now says what the results version tracks: the status and
  offset of the calls it covers, and their output on success, but not
  the bytes and counts a refusal leaves. This is why the change to
  refused skeletons below keeps it at 1
  (#206).

Fixed

• muniGetBidiClass returns Left_To_Right for every value past
  U+10FFFF, as properties.h says. It returned Boundary_Neutral, the
  class of the noncharacters U+10FFFE and U+10FFFF, which the table
  carried past them. Only a caller that passes a value that is no code
  point sees the change: bidi and skeletons read the class only of
  code points decoded from valid text, so their results and the results
  version are unchanged
  (#209).
• muniGetSkeleton now stops at the segment that hit the limit when it
  refuses text with muni_errorLimit (a run of combining marks past
  MUNI_MAX_NORMALIZATION_SEGMENT). It writes the skeleton of the text
  shown before that segment, as far as textOut holds it, and
  neededOut gives its length in bytes; nothing from that segment on is
  written. Before, it went on past the run, dropped the marks that did
  not fit, wrote the skeleton of the rest of the text and set
  neededOut to 0. muniNormalize and muniGetNfkcCasefold already
  stopped this way. A skeleton that succeeds is unchanged, and so is
  the results version
  (#203).