Two traps cost an afternoon: an uncompressed mimetype entry, or LibreOffice
produces no PDF for the file at all, and a font name with no font-face
declaration, which it then resolves differently from the editor — inventing a
0.4mm-per-line difference that no real document has. Plus the note that a
LibreOffice shipping its own Liberation/Carlito copies needs no font install.
LibreOffice's compatibility option (ODF AddParaTableSpacingAtStart, Word's
w:suppressSpBfAfterPgBrk read the other way round). Off, the block that opens a
page loses its space above however the page broke — probed by flipping the flag
in one 63-page document, whose chapter heading then moved from 25.4mm to 20.0mm,
which is where the editor had been drawing every one of them. On is the default
both products write, and both formats carry the flag back out. Fixture 244 → 160.
Probed: a header of one 10pt line with a 12mm bottom margin puts the body at the
band's min-height, not 12mm lower, and a footer's trailing margin is dropped the
same way — while a margin between two of the zone's paragraphs counts in full
(two lines 12mm apart give a 20.16mm band). The collapsed zone paragraph carries
only the margins between its source paragraphs now. One 16-page fixture went
183 reports → 5, its body having sat 10.3mm low on every page.
The edge→zone distance was document-wide, so an index whose header starts 4cm
down had its body open 14.5mm too high and the page under it followed. HfSet
carries the section's own pair (and its first page's, where the page style hands
over): ODF's page-layout margin is that distance, Word's is w:pgMar
w:header/w:footer, and both exports write it back.
Both importers capped every side at 10cm, which is the Layout panel's limit and
no file's: a back cover pushing its last lines to the page foot with a 21cm top
margin rendered 11cm high. fitMargins caps only what would leave the page no
strip of text at all, and the zone distances follow the page's own height.
A chapter number set far larger than its title shares the title's baseline but
starts a long way above it, so banding by the top alone read one line as two
and stopped the comparison at the first chapter opening. Words carry their
height now, and a word joins the line it overlaps by half the shorter box.
Corpus unchanged at 9.
The outline level's character style, its indent and the stop its label's tab
runs to were never read, so a 96pt grey chapter number rendered at the
heading's own size and everything under it sat ~30mm high. The label is an
inline block now — one set larger than its heading grows the line as both
products grow it — hung out of the paragraph's indent with the level's stop as
its minimum width. text-indent inherits into that block and would shrink it by
exactly what it hangs out, so the label zeroes it. Both formats carry all of it
back out (ODF's label alignment, Word's w:suff / w:ind / w:rPr).
Corrects two probed claims that measurement contradicted: a hard page break
does *not* swallow the block's space above (only an automatic one does), and
the header band is max(declared height, content + gap) rather than a line
count. Adds the chapter-numbering model, the mirrored master pair, and the
harness ceiling a CSS-drawn label is.
A numbered thesis lost every "1.", "2.1.4" and "A." on import: the outline
definition was read as a list wrapping headings and dropped. It is now a
document-level definition on the style sheet (format, prefix, suffix,
display-levels, start per level), drawn by CSS counters that reset down the
levels, and carried by both formats — ODF's text:outline-style, Word's
multilevel numbering bound to the heading styles. The contents rows count the
same way, since no text walk can read a CSS counter.
LibreOffice's Left Page / Right Page chain into each other through
style:next-style-name, which was read as the different-first-page idiom: one
master's geometry and zones governed the whole section, so a book-style body
lost its running head on every second page and its inner/outer margins with
it. A next-style cycle now becomes one mirrored setup — the right page's
geometry, the left page's zones as the even variant — and style:page-usage
left/right marks the margins as mirrored.
A roman front matter before a decimal body is one page style before another
in ODF and one w:pgNumType per section in Word, not a document setting — read
as one, its pages printed 1, 3 where the reference prints i, iii. HfSet grows
pageNumberFormat, both exports write it back, and the DOCX importer takes the
document's own format from the first section, as it already does the paper.
A style's fo:text-transform / w:caps is a CSS transform in the browser, so
Range.getClientRects' text stayed lower case where the reference PDF has real
uppercase — three headings of one fixture reported a lineBreak that was not
one, and comparison stopped at each. The computed textTransform is applied to
the extracted word now. Reports rise where they were hidden behind such a line.
ODF's table:align="margins" means the table spans between its own margins, and
style:width is then the width the producing page had. Read as a left-aligned
table, the undeclared right margin was derived from that width against ours and
the table came out narrow — measured 12.4cm on a 14.7cm text width, its column
rule 10.4mm left of where LibreOffice draws it.
Five legs in tests/lo-password.test.ts, driven through the repo's first UNO
helper because --convert-to cannot pass a password; self-skips without soffice
or python3-uno and runs with npm run test:lo. The probed format details live in
docs/architecture/encryption.md.
Writes agile encryption with Word's own parameters (AES-256-CBC in 4096-byte
segments, 100000 SHA-512 rounds, HMAC integrity) into a compound file built
here — cfb.ts, MiniFAT and DIFAT included. Reads that and the older standard
encryption LibreOffice writes, whose package is AES-ECB. Verified both ways
against LibreOffice.
Writes LibreOffice's whole-package form: AES-256-GCM keyed by Argon2id over
SHA-256 of the password, the deflated package as one encrypted-package entry.
Reads that plus the per-entry AES-CBC/GCM form with PBKDF2 that every earlier
version writes; the nonce sits ahead of the ciphertext, which no manifest says.
Blowfish and unknown algorithms report an unsupported-encryption error.
ReviewMarginLayer draws the cards in a strip right of the page, at the height
of the line each belongs to, with a dashed leader running through the gap
under it. A balloon is pushed down by the one above (balloonStack, measured
per card, unit-tested) and the column belongs to its page, so it never runs
past the sheet. The strip is only reserved where the margin has something to
show; the smoke test guards a card's content against its box.
storage/markup holds the display mode (all, simple, none, original) plus, per
kind, whether it shows and where its cards go; markupView folds that into what
the page acts on. Bars, leader and both panes ask reviewItems instead of the
document, the page takes the hiding attributes, and trackChanges drops its
author decoration where the mode shows a finished text.
The thread rides the comment mark's attrs, so two runs of one comment still
merge back together and no second store is needed. ODF writes each answer as
its own annotation over the same range, naming its comment in
loext:parent-name; Word writes its own w:comment tied by w15:paraIdParent in
commentsExtended.xml (both probed). Either importer folds the answers back
into the comment they belong to, and the printed list carries them indented.
A TOC field regenerates its rows from the styles named `toc 1`…`toc 9`, and its
cached rows are skipped on import — so those styles were never seen and the index
fell back to the editor's 0.5cm per level, dropping the file's space-after with it.
Measured on a 60-row index: a 5.1mm row pitch against LibreOffice's 6.9mm.
They now land as Contents 1…10, and the DOCX export writes the rows back under
them instead of the fallback indent.
Word caches a BIBLIOGRAPHY field's rows as a w:tbl inside the field, and
only the field's paragraphs were skipped — so the list arrived once as the
regenerated index node and once as that table. Measured on a 49-page
document: two pages of duplicate. StyleName IEEE now reads as the numbered
style it is, so its citations show [1] instead of the source key.
The clipboard carries HTML, and in an HTML string any block tag inside a
<p> closes it — so a copied box arrived as loose paragraphs. It is now
serialized as an inline span carrying its blocks as JSON, read back by a
matching parse rule; its plain text stays in the span for readers that
cannot take the attribute back.
Entries were stored as an HTML string, where any block tag inside a <p>
closes it — so a copied text box, which holds blocks, lost its place in
the line. They now keep the slice's nodes; an entry written earlier keeps
its HTML string, which insertContent takes just as well.
A box that starts its paragraph was a trap: there is no text position beside
it, so the browser pulled the caret meant for the box's own place in the line
into the frame and everything typed there landed inside it.
A frame nobody is editing is now an atom to the browser — a contenteditable
node decoration, dropped while the caret is in its text, the frame placing the
caret itself on the click that enters it. ArrowLeft/ArrowRight at the box's
first and last text position step out to it. The decoration walk enters blocks,
without which it never reached an inline box at all.
A box in the line is a character its paragraph places, as a picture is, so the
three frame-toolbar buttons were the paragraph's own alignment under a frame
icon — a second place to look for the same thing. They now show for the band
wrap alone, where nothing else can express the position.
The ordinary command needed one correction to carry that: a selected box is one
node whose range covers its own paragraphs, so aligning it moved the frame and
re-set its text in the same keystroke.
textBox becomes an inline node carrying block content, like the frame both
formats write, so "in line with text" really is in the line and a box reaches
a table cell or a list item through their paragraphs. Both exports write it
where it sits (an as-char draw:frame / a wp:inline drawing) and both imports
leave it there, which drops the hoisting machinery on either side.
Two guards the schema cannot express: no box inside a box, and a selection
reaching into a box grows to the box's bounds rather than dissolving it.
The shape body wrote spAutoFit, which LibreOffice reads as min-height 0
and then draws the box's text detached from the shape, over whatever
follows it — measured on a 90px box: the frame collapsed to one line and
its text landed 22pt below it, on the next paragraph. noAutofit is what
LibreOffice writes for the same frame; with it the text sits inside the
box at its declared height. Content taller than that is clipped, DOCX
having no fo:min-height for the ODF leg's auto-grow.
wrapAlign only ever came from a file, so a box inserted here stayed at
the left in both modes that leave it in flow. The floating toolbar now
offers left/centre/right for `inline` and `topBottom` (a side wrap names
the side itself) and drops a coordinate placement when one is picked.
DOCX carried neither: `topBottom` now writes wp:align and an as-char box
the w:jc of its rebuilt marker paragraph, both read back on import. ODF
already wrote both. Measured centred at 297px of a 76…718px column, and
rendered through LibreOffice from .odt and .docx alike.
A break between spanned rows stretched the cell across the gap: the
merged table drew over the page gap and the next page's header. A
group too tall for a page still breaks per row. An in-flow text box
took the same treatment for floats — CSS lays a block box over one
instead of beside it, which had the vertical frame covering its
neighbour. Both match LibreOffice's render of the same document.
corpus/word/ holds each document saved again by Word — rsids,
proofErr, separator notes, theme docDefaults — asserted to read like
the docx-lib original and survive our round trip. The formula fixture
is reauthored: a hand-wrapped MathNumerator doubled m:num/m:den, which
Word refuses to open, and the radical gets a visible degree.
Four new themed documents (11–14), authored by the docx lib with
LibreOffice ODT twins like the rest; make-fixtures takes a regex to
regenerate a subset without rewriting the committed files. The twins
need libreoffice-math — without it soffice drops every formula
silently.
Renders the same document with and without the border through
LibreOffice and compares every word box — pins the fo:margin carve-out
at render level, where the round trip is blind (both sides symmetric).
The generator assigns a builtin (node species following the style's
level kinds, override-, start- and marker-format-free — each of those
drops the name by design) and no longer pairs textAlign right with an
rtl paragraph, where it is the suppressed default. Found the w:jc
suppression fix the commit before this one.
The DOCX import dropped left as the default and kept right regardless
of w:bidi; an rtl paragraph's default edge is right. start/end read as
physical left/right — LibreOffice spells an rtl paragraph's physical
left as w:jc="start" (probed via its own DOCX export).
utils/chapterField owns the per-zone reading (header = the chapter at
the page top, footer = the last begun) so the probed LibreOffice
semantics is unit-tested; the DOM measurement feeding atTop stays in
Editor.svelte.
The margin carve-out under a page border, the bare note anchor and a
Writer frame keeping a list and a nested image through an soffice
re-save each get an asymmetric assertion — a symmetric export/import
pair had hidden all three.
Without a parent chain to the Frame style LibreOffice read the box as a
drawing object, whose text loses lists, nested images and borders. The
frame takes fo:background-color/fo:border and its writing mode in the
graphic properties; an inline box's centring rides the anchor paragraph.
The import reads the frame spellings back and drops LO's from-top/0
respelling of the default alignment.
Word reads a \# picture with the reader's regional separators, so a
German Word took "0.00"'s dot as grouping and printed 84 as "084".
The importer now reads a picture back by its shape, whichever locale
wrote it.
The three index kinds exported as fields with an empty result; Word for
Mac never updates fields on open, so every index rendered as nothing.
They are now complex fields whose result is the editor's cached entries
(indent per level, leader tab, pages), the form Word itself writes; the
importer recognizes the complex INDEX/BIBLIOGRAPHY form and keeps the
attrs the fldSimple path set.
Word resolves the quoted style name against the localized built-in
name, so 'STYLEREF "Heading 1"' shows a field error in any
non-English Word. The numeric argument means outline level in Word
and LibreOffice alike (probed), and the importer already reads it.
The docx package always writes the element; the page-background pass
prepended a second one, breaking the CT_Settings sequence so Word
repairs the file on open — and the repair drops every header and
footer reference. The schema test's DOCX leg now builds with the
page decor, line numbering and fold marks, which covers the pass.
Anchored exactly on the paragraph top (posOffset 0), LibreOffice's wrap
layout derails and paints a neighbouring inline text box's content
outside its shape; LO itself never writes less than a twip there.
Probed by bisecting a broken box against LO's own DOCX export of the
same document. The importer rounds the sub-visible offset back to none,
so nothing accretes on the round trip.
style:writing-mode alone leaves the paragraph left-aligned in
LibreOffice; the mode+end pair is what it writes itself (probed), so
both export paths add it where the block sets no alignment of its own,
and the importer suppresses end/right against an RTL block's default —
the pair is its normal look, not direct formatting. The render-
consistency leg now compares under spacing model 'max', the one both
formats express; thresholds re-measured.
Every cell already writes all four sides itself, and on a styled table
the table-level black default showed in Word, which lets table borders
beat a cell's "none" (only "nil" loses) — LibreOffice lets the cell
win, so only Word drew the grid over the style. Plain tables keep the
default grid via the per-cell fallback.
Three legs against the closed-loop blind spot the duplicate-style bug
exposed: package-lint.test.ts checks invariants the schemas cannot
express (unique style ids/names, dangling style/num/rel references,
balanced ranges, manifest completeness); lo-roundtrip gains a
soffice→fodt leg asserting the styles LibreOffice resolves (Heading
chain Arial/bold/sizes, Word defaults); lo-render-consistency.test.ts
renders the kitchen sink from both formats through LibreOffice and
compares pagination and pixels (ratchet thresholds). CI's LO job
gains poppler-utils; test:lo runs both LO files.
The docx package's Styles merge spreads our options over its factory
set, so a passed importedStyles array replaced the whole set —
w:docDefaults (default font, size, language) and the Title/Heading1-6
factory slots included. Every document using a table or list style
exported without document defaults; Word fell back to its own. The
style XML is now spliced into styles.xml by a post-pack pass
(applyRawStylesDocx), and the regression test builds with a table
style and asserts w:docDefaults survives.
Steffen Becker