Word wraps every citation in a w:sdt; the inline walk ignored the wrapper, so
its runs never reached the paragraph. a fixture fixture lost every [7]-style
reference with them, and now matches LibreOffice's page count.
Word's w:lvlJc / ODF's fo:text-align on the level: set right, the label grows
into the margin instead of over the text it labels. That is what the built-in
Roman numberings use, and a fixture's numerals collided without it.
A spacing above single puts all of its leading after the line (probed against
LibreOffice at 150%); CSS splits it. The block is shifted back up by that half,
which leaves the flow — and pagination — where it was.
Both anchor offsets now render: the x moves the rotor inside the full-width
float, the y the float's top. Chromium puts every line after such a float
below it, so a frame set below the paragraph top is sunk behind the text at
import and the node view measures what those lines already cover.
A formula node stores only its LaTeX source; the MathML on screen, the ODF
formula object and Word's OMML are derived from it, and both importers parse
foreign markup back into that source, so an imported formula stays editable.
Word documents were losing their formulas outright: m:oMath sits beside the w:r
runs in its own namespace, past convertInline's `namespaceURI !== W` guard.
An as-character image is now a line box in getLineRects, so a paragraph of
stacked images breaks between them instead of overflowing the page. Only a
floated image keeps its paragraph atomic, and the leaf's height reaches down
to the lowest float — a float hangs out of the paragraph's box, so
offsetHeight alone let the image walk off the page foot. prevPlacementsKey
damps the ping-pong that overhang causes between two layouts.
Only the top level's own w:ind w:left (ODF: the label-alignment margin)
was read; every level below it was forced to the editor's 1.27cm step.
Both formats store a level's indent absolutely while the editor nests one
step per level, so import now subtracts the level above and the ODF
export adds the running sum back per level (DOCX already accumulated).
Measured: a 0.5cm/-0.6cm pair puts LibreOffice's two levels at 1.77cm and
2.44cm, and the editor draws them at 1.769 and 2.44. Parity corpus
unchanged.
w:tcMar and an ODF cell style's own fo:padding were both dropped: the
table's value (in ODF the first cell's) stood for every cell.
cellPadding is now an attr on tableCell/tableHeader too, measured against
the table's rather than the editor default, so only a real disagreement
carries one. --cell-pad inherits, so the cell's own declaration wins for
its own td and the CSS is unchanged.
Checked against LibreOffice: a 1.2cm cell in a 0.3cm table puts its text
at 332.2pt where the grid column starts at 297.6.
The zones already imported, rendered and exported per section, but only
section 1 was editable state — the rest were marked read-only.
Editing now targets the section of the page it starts on: a double-click
edits the zone under the pointer, the Layout-panel buttons the current
page's. zoneKey/writeZone route the single live editor's read and its
write-back to either the six per-zone props or the extraHfSections entry.
Measured against LibreOffice on a purpose-built fixture the rule is
right, and on the corpus it is a loss: one fixture gains 22 issues.
The strut is also what holds up a line of no runs and, through the
reserved ::after, the paragraph mark on the last line.
LibreOffice restarts the tab grid on the continuation line; Chromium
hangs the leftover tabs and starts the next line at the margin, 160mm
away in the repro. The run is walked from the pen before it — the one
thing the fix can't move, so the answer holds whether or not the break
is already in place — and a break widget goes where the line runs out.
Measuring this turned up why the two disagreed in the first place: a
file that declares no tab interval falls back to ODF's own 2cm, not to
LibreOffice's UI default of 1.25, so leaving ours implied on export
rendered every tab elsewhere. Both exports declare it now, and both
imports fall back to their format's value rather than to ours.
Only runs of two or more are walked: reading a block's geometry forces
a reflow, and doing it per tab across 48 pages slowed pagination enough
that the parity harness compared a layout still settling.
The left⇥centre⇥right layout every letterhead uses rides on them, and a
zone had none: Word keeps them on the Header style, ODF on the zone
paragraph's, and neither was read or written.
An inactive zone is static generateHTML output no plugin reaches, so it
is measured straight on the DOM. Two bugs surfaced doing it: a segment
its own advance had pushed onto the next line measured as wrapped, and
the alignment then had no way back — it takes the sum of the segment's
client rects now. And the scale came from offsetHeight, which rounds
half a percent away on a zone one line tall.
The default tab interval was fixed at 1.25cm, so a file's own
w:defaultTabStop / style:tab-stop-distance was ignored and the DOCX
export wrote 1.25 back over Word's 1.27. It is a document setting now.
A stop can also carry the character it repeats across its gap. odf-kit
has no leader option and mints one automatic style per option set, so
the leader rides style:type behind a sentinel — matching on position
and alignment alone gave every stop at 6cm the same leader, which
LibreOffice drew back plain as day.
A document had one set for all of it, taken from whichever section
declared one last — so a fixture lost its letterhead entirely
and page 1 got the empty header its final section asked for.
A body block now carries `sectionBreak`, ordinal rather than indexed so
editing cannot desync it. Import resolves Word's "Link to Previous" per
reference type; export writes a set per sectPr, and on the ODF side each
section past the first gets its own master page, which the block switches
to. The layer picks a page's set from where pageBreaks reports the
sections starting, and each section's zone reaches ride
`--pb-section-reach` so a later section's letterhead makes room for
itself: page 10 of a fixture was 24.9mm out, now 1.1mm.
Editing still reaches section 1 only; the rest render read-only.
The root file is read at every session start and had grown to 87 KB, so every
session paid for knowledge it did not use. Move the per-directory detail into
nested CLAUDE.md files, which load when a file in that directory is touched,
and the five deep-dive topics into docs/. The root keeps the commands, the hard
rules and an index of where the rest lives.
Content moved verbatim — the probed Word/LibreOffice behaviour, sentinel order
and shared constants are what the code cannot say. CI fails the root over 120
lines, since a prose budget alone is what let it grow.
CSS gives ::marker no position, so Chromium set it flush against the
text — about 4mm right of where Word and LibreOffice draw it. The marker
is now the item block's ::before, placed at the hanging indent both
exports already write (LIST_HANGING_CM 0.635cm): measured against LO's
own render, the three bullet levels land within 0.04mm.
Out of flow, because in flow it grew every list line by a pixel and cost
a fixture fixture a page. `list-style: none` drops only the marker box,
so the list-item counter and <ol start> keep numbering; a bullet's own
char rides an inherited property, since attr() in a pseudo would read the
block rather than the list.
Three of them described something the reference does not do. The trailing
space is counted by LibreOffice as well; neither engine breaks after a
slash or inside a run of dots; and the hard-break line was real but came
from the paragraph mark, not the break.
Each was checked against soffice on a one-page fixture, so the record now
says what was measured rather than what was inferred.
A long URL or a German compound ran straight past the right margin, where
Word and LibreOffice break it — measured against LO in a 2cm column, the
break now falls on the same character.
`break-word` rather than `anywhere`, so a table column's minimum width is
still its longest word.
The reserved paragraph mark is inline content past the block's last BR,
so it opened a line of its own: LibreOffice set the next paragraph 4.7mm
higher than we did, and a fixture paid for every one of them.
The empty-block rule already pulled the mark out of flow for exactly this
reason — it now covers any block that ends in a break, seating the glyph
on the last line rather than the first.
The cell padding was a constant in three places that disagreed with each
other, so a narrow column lost 0.38cm it did not have: a bibliography
marker column of 0.71cm wrapped "[2] " onto three lines and every entry
grew with it, costing a fixture a whole page.
The table now carries its own margins (w:tblCellMar, direct or from the
table style; fo:padding on the ODF cell style) and both exports write
them back. A side within a twentieth of a millimetre of the default is
the default, or Word's 108 twips would tag every table.
The parity runner pinned one build number, so a fresh
`playwright-core install chromium` left it pointing at a directory that
is not there. PARITY_CHROME still overrides.
keepLines (w:keepLines, fo:keep-together) is keepNext's sibling and was the
one of the pair nobody read: a block that declares it now paginates whole,
unless it is taller than a page, where Word splits it too.
The vertical frame offset is data only from here on. A line box avoids a
float's whole margin box, so applying it made dead space where Word flows
text — measured: a fixture's figure page went from 131mm off to 187mm.
Text boxes and shapes had no offsets at all, and for images the wrap side
overrode the x: a file that names the side its text flows on lost the
coordinate, our own export included. Side and place are read separately now,
and one helper turns both into the float's margins for every frame kind.
The re-save leg is the only place that proves what we write is what LO
understands. A raised run is stored as a percentage of the font size, so the
fixture picks one that stays whole — LO rounds the percentage on save.
Letter case, the underline's shape and colour, a double strikethrough and a
freely raised run all had no home in the model, so every one of them was
dropped on import. They ride the marks now, render in CSS, and round-trip.
odf-kit's run formatting has a field for none of them, so a sentinel folds
them into the run's own automatic style — the route the character styles
already take, and it reaches every path odf-kit serializes.
Its leader is real dots now, not a border, so it scales with the font and
reaches the PDF as text. Title and entries follow the document's Heading 1
and default style, which is what the index styles of Word and LO are based on.
Word's positionV and ODF's svg:y say how far below its paragraph a frame
sits; ignoring it put every figure at the anchor itself. It is the float's
top margin, so lines that fit above the frame still flow there, as in Word.
It compares where text sits, so a dropped highlight or fill reads as green;
the visual check that catches those, and its two traps, are written down.
Also records the list marker's ~4mm offset, which the tolerance hides.
Word's highlighter is w:highlight, a palette name; only w:shd's hex fill was
read, so 121 yellow runs of a fixture imported unmarked. 'none' maps to the
same 'auto' sentinel an empty w:shd gives, so it still cancels a style's pen.
Everything else in tests/ compares documents at the model level; nothing checked
what a page actually looks like. run.mjs renders the same file through soffice
and through the real app in Chromium, reads both back as word boxes, and diffs
page count, line count, line text and line position.
The corpus stays out of the repo: make-fixtures.mjs authors the baseline .docx
files, and real-world documents are whatever you drop into the gitignored
fixtures/ yourself.
A drawing the editor cannot render was dropped outright, and everything below
it moved up: a thesis with five charts and four EMF figures lost 102cm of
layout, five whole pages of it. They now come in as a labelled box of the
drawing's own size, so the pages hold what they hold in Word.
A German document's "Inhalt" came back as "Table of Contents" — the node
carried no title of its own, so the generated one always won. Now the
imported heading rides on the node and goes back out with it.
A letterhead header is usually a couple of empty lines, the logo, then one
more — the empty leading ones were dropped on import, and the body's content
area counted every line as one text line, logo included. Page 1 of a contract
form started 19mm too high.
A heading whose successor no longer fits below it was left stranded at the
page foot; Word and LibreOffice move it down with them, which is why their
heading styles all carry w:keepNext / fo:keep-with-next. Headings do it by
default here, so the attr only marks the other blocks that ask for it.
One level deep: a run of kept-together blocks would need the pagination pass
to reconsider leaves it has already placed.
Word's Title style tightens by -0.5pt, so a title came out ~3mm wide of
LibreOffice's. letterSpacingPt joins TextProps and rides both formats
(w:spacing, fo:letter-spacing).
Line spacing needed the same care: it is proportional to the font's natural
line height, so the factor and that height are now separate CSS variables and
editor.css multiplies them in one place. Before, whichever of the two was
declared nearer simply overwrote the other — a block carrying its own font
silently dropped the style's spacing.
The block's CSS strut is the floor under every line, so a paragraph whose runs
are smaller or of another family than the style rendered too tall — 03-fonts
drifted 4mm over a page. The mark's family now rides the block like its size
already did (Word's w:pPr/w:rPr/w:rFonts, ODF the paragraph's own text
properties), and supplies the natural line height single spacing means.
A mark that declares nothing but whose runs all agree takes their font: that
is the paragraph's font, and it is what a word processor struts with.
Word anchors a wrapped frame at an offset; the model only knew a side, so a
frame the file puts 91mm in snapped to the margin and the text beside it came
out 20mm too narrow. wrapOffset carries that x (cm from the column's left
edge, Word's posOffset / ODF svg:x) and the float's outer margin renders it —
against the column vars, so an indented anchor paragraph cannot skew it.
Picking a wrap side in the toolbar clears the offset: that gesture means
"put it there".
An inline atom wider than the column is pushed onto a line of its own and
leaves the empty line above it in the document — 5mm per image. A picture the
file sizes to the column lands there anyway once its cm/EMU width is rounded
to whole px, which is how a full-width Word picture triggered it.
A paragraph holding an image rendered a line taller than it does anywhere
else: under white-space: pre-wrap the browser already shows the last line, so
ProseMirror's hack BR adds an empty one. The hack is kept where it earns its
keep — after a real break it carries the caret onto the empty line, and in an
empty block it is the only line box there is.
Word's inserted page number sits in a w:sdt, so the paragraph is not a child
of w:ftr and the whole zone came back empty. Mirrors what convertBlocks
already does for the body.
wrapText="bothSides" says nothing about the side, so every such frame floated
left — including the ones Word puts in the right half of the column, which
then pushed the text off the page's right edge instead of flowing beside it.
Word's Table Grid — the style nearly every real table names — sets space
after 0 and single line spacing in its own w:pPr, which is why a table's rows
are line-tight. The cell paragraphs were taking the document default instead.
An explicit single spacing is now written out as well: it only used to be
dropped as redundant, but the block's own style may set another one.
Word's List Paragraph carries it, so consecutive list points sit line-tight.
Without it every one of them took the document default's 8pt after, and a
document that is mostly bullets drifted more than a page.
A run with no font already fell back to the document's theme; the style did
not, so the registry held Standard and the headings without a family. Single
line spacing is measured against that family, so it resolved to the wrong one.
Steffen Becker