v0.5.2

Fixed

• A target inside a scaled ancestor (a hero fitted to a phone with scale: 0.4) split into
  words or chars came out with every unit boxed at the scale of its glyphs, since a client rect
  is measured in painted pixels while a unit's width is written in the target's own. The split
  now reads the scale off the target itself, its painted box against its laid-out one, and
  restates every extent, gap and hyphen in the target's own pixels. No ancestor is read and no
  transform is touched, so the two phases stay as they were.

v0.5.1

Fixed

• iOS Safari reads every range in the text that follows a styled ::first-letter one glyph
  late, so a drop cap paragraph split into words or chars came out with every unit sized for
  the character after it. The split now notices the shift, since the node's whole range comes
  back without the glyph's own box, and reads that node one glyph back. A first letter that
  does not float is measured from the line's start edge, where it always sits.
• iOS Safari also reports a zero-width rect at the end of the line before a wrap for a range
  that starts at that wrap, which read as a word broken across two lines: a Japanese paragraph
  split into words lost a line break and ran off the page. A rect with no extent along the line
  is no text and is ignored. And a fragment's room comes back a pixel wide of its glyph there,
  which read as a hyphen; a pixel or less of leftover is rounding now.

v0.5.0

Fixed

• A page translator (Chrome's, Google Translate's) used to rewrite a split target unit by
  unit, from the page's declared language whatever the text had become, into words the boxes
  were not measured for: words overlapped and gaps opened (#1). Every line the split writes
  now carries translate="no", so a translator leaves a split target as it is. Translating it
  is a text change like any other: the source markup goes back, the translator rewrites it,
  the split is made again. The README shows how, and the demo does it.

Added

• The demo's install command has a button that copies it, with a check for a moment
  after; where there is no clipboard to write to, the command is selected instead.

v0.4.0

Added

• splitText takes an array, or any iterable, of targets and returns one TextSplit per
  target. Every target is planned before any is written, so a page of blocks costs the one
  layout a single block does, where a loop of single splits forced one per block: a word or
  character split of an article takes about a third less time, most of all in Safari. The
  hyphen a broken word is restated with is now measured once every cut is made, for the same
  reason, in a single split as well.

v0.3.0

Added

• mask takes an object naming, per level, how far the clip reaches past the box across
  the line: mask: { lines: ".25em" }. Display type set tighter than its glyph boxes
  leaves descenders and accents outside the line box, and a clip at the box cuts them
  until it is dropped; a reach keeps them in the window with no change to layout, since it
  is a negative inset on the clip itself. Read the clip back off the mask's inline style
  to restore it after clearing it.

v0.2.1

Fixed

• Lines set tighter than their glyph boxes (a display heading at line-height: 0.85,
  or any leading below the font's content area) no longer merge into one line. A rect
  joins the row its centre falls in, where it used to join any row it overlapped, and
  consecutive rows of tight type overlap by a few pixels.

Added

• The demo outlines every line, word, character and mask on request, from a
  Boxes control in the panel, so what the split produced is visible on the
  page.
• The demo's header shows the CSS-only reveal next to the scripted one: the
  stylesheet that staggers off the index the split writes, and the two lines of
  script it needs.

Changed

• The demo page reads as a specimen book: the cases sit in named groups with a
  table of contents that stays in view, every case shows its id and what it asks
  of the split, and the controls sit in one panel that keeps out of the text.
  The page is set in Geist and Geist Mono, self-hosted as variable faces. It
  reverts and splits every case again when its column changes width, with no
  reveal a second time, the way a consumer of the library has to, and its
  header shows the code for the reveal and for the resize, in plain browser
  APIs. The README now says so too: a split is a snapshot of one layout, and
  nothing animates it or watches for the next one.

• Releases publish from GitHub Actions. npm run release now ends by pushing
  the tag, and the Release workflow publishes to npm through trusted publishing,
  with provenance, and lists the GitHub release; no npm login or one-time
  password is needed on the machine that cuts the release.

v0.2.0

Added

• npm run release -- patch|minor|major: one command that rolls this changelog,
  restates the size in the README and the demo, bumps, tags, pushes, publishes to
  npm and opens a GitHub release. AGENTS.md and .agents/skills/ document how to
  develop, test and release the library.

Changed

• Only text is split. A block-level piece that is not running text (a media
  tile, a button row, a table, a box-like custom element, a summary, a block
  matching ignore) used to be one unmasked line, and was revealed as if it
  were text; it is now left exactly where it is and is not a unit at all, so
  lines, words and chars hold text units only. An inline-level piece
  still rides along inside its line and still counts as one word.
• A float is no longer left inside the line block the cut dropped it in, where
  the line's animation carried it along. It is put back in front of the block
  of the line it floated beside, at the same top, so the lines still flow
  around it as painted and nothing animates it.
• A floated drop cap (a ::first-letter with float) is a float like any
  other: its restated glyph now sits in front of the first line block instead
  of inside it, where the line's mask clipped it to one line's height during
  the reveal and the line's animation moved it.
• A word or char split of a paragraph with a floated drop cap no longer widens
  the first word by the glyph's box: the word is measured without the glyph,
  the glyph has no extent among the graphemes, and the char unit it leaves is
  dropped rather than kept as an empty unit.
• A run with nothing to split (a lone float, an ignored element, a <br>) is
  left exactly as it is instead of becoming an empty line.

v0.1.0

Added

• splitText(target, options): splits a target into painted lines, and words and
  graphemes inside them, without changing how the text wraps.
• Line detection off the browser's own line boxes, so text-wrap: balance,
  pre-line, <br>, right-to-left text, scripts without spaces, hyphenation,
  overflow-wrap, text-indent, floats, multi-column layout and vertical
  writing modes all come out as painted.
• Content classification by layout: block containers are split inside
  themselves, pieces that are not running text (media, button rows, tables,
  box-like custom elements, ignored elements) ride along whole, hidden content
  is left alone.
• Text decorations carried onto word and character units, so underlined links
  keep their rule and follow hover.
• A styled ::first-letter restated on the glyph and a ::first-line on the
  first line block, so big initials, floated drop caps and first-line styles
  survive the split in every browser.
• Word and character units, and the gaps between them, sized to their painted
  extents, so boxing the text moves nothing.
• A hyphen the browser drew at a break restated in a box of the same width.
• Verified in Chromium, Firefox and WebKit through the demo's own check.
• Selection hooks on every unit (data-line, data-word, data-char,
  data-mask, --line, --word, --char) and on the target (data-split,
  --lines, --words, --chars), plus optional class names.
• A framework-free demo page with a check that compares every split with the
  lines the browser painted, text and geometry.