doclayout-0.5.0.2: changelog.md

# doclayout

## 0.5.0.2

  * Performance optimizations:

    + Fuse double modify calls in outp, newline, and FCookedText handler.

    + Cache polymorphic prefix in RenderState. Add prefixA, prefixTrimmedA,
      and prefixLen fields to RenderState that cache the polymorphic
      conversions of the Text prefix. This eliminates
      the per-line fromString/T.unpack/realLength conversions in outp,
      newline, and the FCookedText handler. Add a setPrefix helper that
      updates all cached fields atomically. The block handler computes the
      space-appended prefix incrementally to avoid expensive recomputation
      on deeply nested blocks.

    + Avoid redundant realLength calls in mergeBlocks padding.
      Thread precomputed line widths through the merge process by changing
      the internal representation from (Int, [a]) to (Int, [(Int, a)]) where
      the inner Int is each line's rendered width. This avoids calling
      realLength on every line during every merge pass; widths are computed
      once in toBlockSpec and carried through.

    + Remove redundant normalize in flatten for Prefixed. renderList
      already normalizes when processing FPrefixed. This removes the
      double traversal.

    + Use strict foldl' in Attributed foldlChar instance.

    + Add shortcuts for Thai, emoji modifiers, CJK, and fullwidth chars.
      Add new shortcuts to updateMatchStateNarrow and updateMatchStateWide:

      - Thai script (U+0E00-U+0E7F) with combining marks for vowels/tones
      - Zero-width joiner (U+200D) with inline emoji-joining logic
      - Emoji presentation variation selector (U+FE0F)
      - Emoji skin tone modifiers (U+1F3FB-U+1F3FF)
      - CJK Compatibility Ideographs (U+F900-U+FAFF)
      - Fullwidth forms (U+FF01-U+FF60, U+FFE0-U+FFE6)
      - Supplementary CJK Extension B+ (U+20000-U+3FFFF)

  * Use withNumTests instead of deprecated withMaxSuccess in test suite.

## 0.5.0.1

  * Have `nestle` remove leading CarriageReturn.

## 0.5

  * Extract existing HasChars definition into a module
    Text.DocLayout.HasChars (Evan Silberman).

  * Add a `build` method to the signature of HasChars [API change]
    (Evan Silberman). This has a default definition, and is only needed with
    the ANSI renderer, so existing users should not need to add anything to
    their HasChars instances.

  * Introduce support for ANSI-styled output (Evan Silberman) [API change].
    `renderPlain` and `renderANSI` are now exported; the old `render`
    is a synonym of `renderPlain`.  In addition, various functions are
    exported to add ANSI formatting (including bold, italics,
    underline, strikeout, links, and colors) to a Doc. The Attributed
    type is also now exported.

  * Change type of Block constructor, replacing `[a]` with `[Attributed a]`,
    which carries a Font along with an inner string type [API change]
    (Evan Silberman).

  * Introduce FlatDocs and use them for rendering (Evan Silberman).
    This is an internal concept, not part of the public API.
    FlatDoc is an "intermediate representation" for the Doc "interpreter".
    The general design is that any Doc can be turned into a list of FlatDocs
    that carry equivalent information. The main point of doing this
    is to replace the nested Styled and Linked Docs, which form a more
    complicated tree structure than previously existed in DocLayout.

  * Deprecate `unfoldD`.

## 0.4.0.1

  * Add clause for Empty to renderList (#22).
  * Remove upper bound for criterion

## 0.4

  * Expose `unfoldD` [API change].
  * Remove `realLengthNoShortcut`, `isEmojiModifier`, and
    `isEmojiJoiner` [API change] (Stephen Morgan).
  * Add new exported functions `realLengthNarrowContext`,
    `realLengthWideContext`, `realLengthNarrowContextNoShortcut`,
    `realLengthWideContextNoShortcut`, `isSkinToneModifier`, `isZWJ`
    [API change] (Stephen Morgan).
  * Compute `realLength` strictly.
  * Make `getOffset` stricter.
  * Drop support for ghc <= 8.4, add test for ghc 9.2.
  * Don't collapse the CarriageReturn + Newline combination (#20).
    We want to ensure that a literal starting with a Newline
    doesn't lose the newline if it occurs after a CarriageReturn.
    This affects code blocks in pandoc that begin with newlines.
  * Improve performance of NoShortcut code (Stephen Morgan).
  * Simplify emoji processing (Stephen Morgan).
  * Add benchmarking for code with no shortcuts (Stephen Morgan).
  * Add unicodeWidth.inc to cabal file.
  * Fix `offset`, `minOffset`, `updateColumn` so they don't re-render.
  * Get unicode block widths directly from the Unicode specification, rather
    than writing it out ourselves (Stephen Morgan).
  * Resolve the width of ambiguous characters based on their context
    (Stephen Morgan).
  * Spacing marks should have nonzero width, even though they are combining
    characters (Stephen Morgan).
  * Add shortcuts for extended Latin, Arabic, Cyrillic, Greek,
    Devangari, Bengali, Korean, Telugu, and Tamil (Stephen Morgan).
  * Fix location of extra-source-files in cabal.
  * update.hs: require text package
  * Handle emoji variation modifiers specially, so the keypad emoji can be
    ignored (Stephen Morgan).  This results in a 16% speedup of realLength
    on ascii text.
  * Add benchmarks for all scripts used by more than 50 million people, plus
    a couple more. (#9, Stephen Morgan).

## 0.3.1.1

  * Fix the end of the block of zero width characters which contains
    the zero-width joiners and directional markings (Stephen Morgan, #5).
    This fixes a regression introduced in 0.3.1, affecting code
    points 0x2010 to 0x2030.

## 0.3.1

  * Improved handling of emojis.  Emojis are double-wide, but
    previously this library did not treat them as such.  We now
    have comprehensive support of emojis, including variation
    modifiers and zero-width joiners, verified by a test suite.
    Performance has been confirmed to be no worse for text without emojis.
    (Stephen Morgan, #1).  API changes: export `realLengthNoShortcut`,
    `isEmojiModifier`, `isEmojiVariation`, `isEmojiJoiner`.

## 0.3.0.2

 * NOINLINE `literal` instead of `fromString` (#2, sjakobi).
   This produces a further reduction in allocations and
   pandoc compile time.

## 0.3.0.1

 * NOINLINE `fromString` (#1).
   @sjakobi reports that this change reduced total allocations
   for building pandoc-2.12 with GHC 8.10.4 by 8.5% and reduced
   peak allocations are reduced from 3854MB to 3389MB.

## 0.3

  * Add foldlChar to signature of HasChars [API change].
  * Use foldlChar in realLength. This avoids a stack overflow
    we were getting with long strings in the previous version
    (with foldrChar).  See jgm/pandoc#6031.
  * Replace isBlank with isBreakable and improved startsWithBlank.
    Previously isBlank was used in the layout algorithm where
    what we really wanted was isBreakable.
  * Avoid unnecessary calculation in updateColumns.
  * Replace a right fold with a strict left fold.
  * Add strictness annotations in realLength and updateColumn.

## 0.2.0.1

  * Made `realLength` smarter about combining characters.
    If a string starts with a combining character, that character
    takes up a width of 1; if the combining character occurs after
    another character, it takes 0.  See jgm/pandoc#5863.
  * Improve `isBlank`, re-use in rendering code `for BreakingSpace`.
  * Fixed incorrect `Text` width in renderig blocks.

## 0.2

  * Add instances for `Doc`: `Data`, `Typeable`, `Ord`, `Read`, `Generic`.
  * Add `literal` (like `text`, but polymorphic).
  * Change some `IsString` constraints to `HasChars`.
  * Add some default definitions for methods in `HasChars`.
  * Change `offset` and `minOffset` to be more efficient (in
    simple cases they no longer render and count line lengths).
  * Add `updateColumn`.
  * Fix problem with `lblock`/`cblock`/`rblock` when `chop` is
    invoked. This caused very strange behavior in which text
    got reversed in certain circumstances.

## 0.1

  * Initial release.