The rules of Word found since 30 October, each with test documents laid out in Word and checked on the 368 existing documents:
Formulas are set by the numbers of Cambria Math, the font Word sets them in: fractions, roots, scripts, brackets and large operators. Lines with formulas stand within 0.42 pt of Word's, and formulas are as wide as Word's within 0.1 pt.
A justified line takes the next word in by squeezing its spaces only when each space loses at most half of what it would gain stretched. In a 204-page report, 549 of 550 such choices are now Word's.
In columns of different widths, a paragraph is set in the width of the column it actually lands in, and a paragraph that runs over into the next column keeps its space above it there. A report converted from PDF, in columns of 461 and 687 pt, now has its pages as in Word.
Text with a character scale, as converters from PDF write it, has the line height of the font Word sets it in, and a page is laid out at most 22 inches wide, as Word lays it out. Bookman Old Style, which Word for Mac takes from its own fonts, is set at Bookman's widths and line heights, also when the file names one of its faces; Garamond after a picture and a tab stands where Word puts it.
Tables: the text of a narrow cell is set in the cell's own width; a cell's room counts half of each side border in whole twips; every line of a cell with one decimal tab stands at the stop by its first number; a table nested in a cell takes an indent below zero as none; in a row of exact height, what sticks out of a cell is cut at its sides; an empty paragraph after a nested table takes no room; the borders of a merged cell are drawn part by part along its rows.
Tabs and hyphens: a tab that leads exactly to the right indent stays at the end of its line; a line may end after a dash that begins its paragraph; a line that ends at an optional hyphen ends in a visible hyphen, which must fit. Saving now writes optional and non-breaking hyphens as Word writes them: before, an optional hyphen came back visible in the middle of its line when the saved file was opened in Word.
Documents for Word 2010 and older: columns balanced above a continuous break count the space after paragraphs, turned text in table cells keeps no margin on the side its lines stack towards, and the numbers of .doc lists keep their own size. In the 257-page reference book in .doc format, all 15,207 lines that can be told apart by their text stand within 1 pt of Word's.
On the 368 existing documents, 31,916 of 31,951 pages begin with the same line as in Word, 99.15% of lines end where Word ends them, and in 354 documents every page holds the same text as in Word.
On this site: the name written apart, Vex Office, now leads to it too.
This week's rules of Word, each found with test documents laid out in Word and checked on the 368 existing documents:
In documents for Word 2010 and older, a paragraph that starts a page with its own page break keeps its space above it, and the line after a page break that opens a paragraph is its first line, with its first-line indent. In Word 97–2003 documents the paragraph mark after a page break stands at the top of the next page. The 257-page reference book in .doc format now begins 254 of its 255 pages as in Word, 244 before.
The columns of a fixed table are as wide as their widest cells, and every row stands on them, whatever widths its own cells ask for; the grid written in the file does not count. Plots that converters from PDF lay out as tables, with axis labels in narrow cells, stood up to 88 pt to the left.
Floating tables stand across the page where they ask to: from the column, the margin or the page, by their offset, centred or to the right. A table width in twips is that of its grid; a width in percent, in Word 2013 documents, reaches the outer edges of the side borders.
Columns balanced above a continuous section break follow Word 2010 in older documents: the later columns are no higher than the text of the first.
An object placed from its paragraph is placed from the top of the space above the paragraph, a paragraph that holds only the places of floating shapes is as high as its paragraph mark, and a table in a text box is cut at the edge of the box.
Decimal tabs align on the first point or comma, whatever the language of the system, and the one decimal tab stop of a table cell aligns text that has no tab before it.
Floating objects in table cells stand by their own offset from the cell; text that does not turn with its shape stands in the frame the turned shape shows; the text of rounded boxes and block arrows stands where their definitions put it; text turned on its side in a cell is cut at the edges of the cell.
In lists, the tab after a label goes to the hanging indent when the label ends short of it at all, and a Wingdings bullet makes its line as high as Wingdings. Character spacing no longer switches on the pair kerning of the font.
On the 368 existing documents, 31,916 of 31,951 pages begin with the same line as in Word, 99.13% of lines end where Word ends them, and in 352 documents every page holds the same text as in Word.
Five more rules of Word, found with test documents and checked on the 368 existing documents:
A line between table rows that is the bottom border of the cell above lies in the top margin of the cell below: the text stands under the line or under the margin, whichever is lower, not under both.
The text of a cell stands off the middle of its side border by the margin or by half the border, whichever is more: a thick border pushes the text in, and the table stands with the outer edge of that border at its indent, also when the border is the cell's own and not the table's.
A row of exact height cuts what does not fit in it above the bottom margin of the cell, not at the very edge of the row; and a row begins at the foot of a page only when its text fits there with the bottom margin that stays above the break.
Of several tabs before a word that does not fit, only the last goes on to the next line. Pages that converters from PDF set out with runs of tabs now stand as in Word: on one newspaper page all 308 lines stand where Word puts them, 257 before.
The font of the run a picture stands in gives a line beside text no height in documents for Word 2010 and older, and its whole height in documents of Word 2013.
On the 368 existing documents, 31,904 of 31,951 pages begin with the same line as in Word, 99.12% of lines end where Word ends them, and in 352 documents every page holds the same text as in Word.
A form whose header Word lays out its own way, a chart whose axis labels stood beside their places and a two-column form led to five more things Word does, and Vex Writer now does them too:
A DOCPROPERTY field in a header or footer shows the property as the document holds it, as Word shows it when it lays out the pages, rather than the result stored with the field, which can be out of date.
A right or centred tab stop beyond the right indent stands at the indent. In a justified paragraph, when the text after such a tab does not fit before its stop, the tab takes no room and the text simply wraps.
Text raised or lowered from the baseline makes its line taller where it stands, in lines with tabs and line breaks too, and the line spacing multiplier adds by the height of the text itself. 48 test cases laid out in Word (Arial and Calibri, raised and lowered by 7 and 3 pt, single and 1.15 spacing) agree within Word's own 0.24 pt grid.
A line has exactly the room between its indents, however narrow; before, it never had less than 20 pt. Text centred or set right in a narrow table cell now stays in its cell, and the labels of a chart axis made of table cells, in a document converted from PDF, stand where Word puts them.
Columns of their own width are as narrow as the file says, even under half an inch, and a paragraph after a column break is set in the width of the column it goes to. Word 97–2003 documents keep the direction of text in table cells: headings turned on their side stay turned.
On the 368 existing documents, 31,903 of 31,951 pages now begin with the same line as in Word, 99.08% of lines end where Word ends them, and in 348 documents every page holds the same text as in Word.
New test documents laid out in Word, 26 cases of merged cells and split rows, borders from 0.5 to 3 pt in every kind of row, and 14 tables that ask for no widths, showed four more things Word does, and Vex Writer now does them too:
A cell merged over several rows goes on on the next page when a page break falls inside the merge. Its lines that the rows on the page do not hold move to the part of the merge there; before, the whole text stayed in the first row and ran over the foot of the page. In documents for Word 2010 and older the first line of such a cell stays on the first page, as Word keeps it.
A row split between pages ends under its text, closed by its bottom border, and its part on the next page opens with the top border. A paragraph that begins that part keeps the space before it. The side borders used to run down to the foot of the page, and the space was dropped.
Each horizontal border of a table belongs to the row below it and is drawn from its line down. The text of rows of exact height now stands under its border, and what follows such a table stands under its bottom border, as in Word.
A table that asks for no widths at all, as tables converted from web pages often do, is sized by its text, as Word sizes it when it opens it: short columns as wide as their text, and long text taking the rest of the line. It used to get equal columns.
Later the same day: a cell aligned to the middle or the foot of a row split between pages stands so within the part of the row that holds its text; Word does not number a paragraph in a cell that continues a merge, and now neither does Vex Writer; a row that cannot be split leaves a column that begins low on its page for the top of the next page; and a line of exact height can be as low as 0.7 pt, as in Word, rather than 1 pt.
On the 368 existing documents, 31,901 of 31,951 pages now begin with the same line as in Word, and in 346 documents every page holds the same text as in Word. One long table of the documents made in Vex Writer is now read row by row in the text of its PDF, where Word's PDF reads column by column, so the comparison by text counts 111 of its lines as different, although every line stands where Word puts it.
A correction. The notes of 22 and 23 October said that in the 34 documents made in Vex Writer all 98 page breaks fall where Word puts them and every line ends where Word ends it. That came from a recount that compared our PDFs with our own earlier ones, not with Word's, and it was wrong. Against Word's PDFs: 96 of 98 page breaks (all 98 by the place of the lines; in two documents the text of Word's PDF reads in another order), and 5,352 of 5,560 lines, 96.3%. Most of the rest is comments and tracked changes, which Word prints in the margin and Vex Writer in the text, and equations. Those notes are corrected.
Test documents laid out in Word, now 72 cases of table rows at the foot of a page and 40 of line heights, showed more of what Word does, and Vex Writer now does it too:
A row of exact height shows only what fits in it, and Word cuts the lines of its cells at the edge of the row: a line half below the edge shows half. Vex Writer used to leave such lines out, or draw them whole. In documents converted from PDF, where rows of exact height are common, this alone put 1,500 more lines where Word ends them.
A row of exact height is never split between pages: when it does not fit, it goes to the next page whole.
Widow and orphan control in the paragraphs of table cells holds only in documents made for Word 2013 and later. In older documents, and in Word 97-2003 documents, a paragraph in a cell splits at any line, while in the body widow control holds in every version.
The baseline of a line stands where Word puts it: under the font's line gap rather than above it, at the descent above the foot of a line of at-least height, and at four fifths of a line of exact height. Calibri is measured by the typographic numbers of its font, as Word measures it.
A paragraph whose style, or the defaults of the document, ask for no hyphenation is not hyphenated. That request used to be lost on the way, so two documents had words broken that Word leaves whole.
Later the same day, on a 204-page test document of tables and diagrams: a table of Word 2013 is as wide as it asks to the outer edges of its borders; a row continued on the next page keeps its cells' margins at the break; boxes of a diagram anchored in a cell whose text is turned on its side are placed (they were missing); pictures kept as Windows metafiles go into a PDF as drawings with their text, as Word writes them; and shapes show their gradients and cast their shadows. 197 of its 204 pages now hold the text Word's do (177 before).
Now 99.06% of lines in the 368 existing documents end where Word ends them (98.9% before), and 31,895 of 31,951 pages begin with the same line. A new measure looks at the text each page holds, whatever order its pieces have in the PDF: 36,074 of 36,144 pages hold the same letters and digits as Word's page, and in 340 documents every page does.
In the 34 documents made in Vex Writer, 96 of 98 page breaks fall where Word puts them. (Corrected on 26 October: this note said all 98, and every line; the recount behind that compared our PDFs with our own earlier ones instead of Word's.)
More test documents laid out in Word, and the one Word 97-2003 document of the set, showed more of what Word does, and Vex Writer now does it too:
A list number or bullet makes the first line of its item as tall as its own font is above the line, never deeper below it, and line spacing does not stretch what it adds. A bullet in the Symbol font makes the line taller than one in the text font, because Word's own Symbol is taller than the one in macOS. In 20 test cases line heights now agree with Word to 0.04 pt.
Text set in the Symbol font shows the font's own characters, Greek letters and signs of mathematics. Before, a λ or Λ written that way came out as a Latin l or L.
In documents made for Word 2010 and older, and in Word 97-2003 documents, a paragraph kept with the next one goes to the next page whole, however long it is; only in Word 2013 documents does Word split it. And the next paragraph has to be able to begin on that page: with two lines when widow control is on, or whole if it has three lines or fewer.
The header rows of a table repeat at the top of each column of a page set in columns, not only at the top of the page.
Tables converted from PDF whose rows have cells of widths of their own keep those widths, row by row. In Word 97-2003 documents the cell margins a table names are read, so cells with no margins keep their text at the border, and inline pictures are laid out in whole points.
All 368 existing documents now come out on as many pages as in Word (367 before). The last one was the Word 97-2003 document: it now has its 257 pages, and each begins where it does in Word. 31,880 of 31,951 pages begin with the same line, and 98.8% of lines end where Word ends them. Later the same day: Cambria's own space, a hair wider than that of the free font Vex Writer sets Cambria in, and lines tried one word longer where Word sets Cambria narrower. Four exports of records, of 453 to 2,724 pages, now have every page begin where it does in Word. And in the evening four more: a cell continuing a vertical merge is as wide as the cell it continues, whatever width a converter wrote for it, which put 1,862 more lines where Word ends them; a floating picture anchored in a later line of a paragraph goes with that line to the next page when the line does not stay on the page with it; and a justified line ended by a line break is widened like any other. Now 31,890 pages begin with the same line, 98.9% of lines end where Word ends them, and in 342 documents every page begins where it does in Word.
In the 34 documents made in Vex Writer, 96 of 98 page breaks fall where Word puts them, and 98.2% of lines end where Word ends them. (Corrected on 26 October: this note said all 98, and every line; the recount behind that compared our PDFs with our own earlier ones instead of Word's.)
More test documents laid out in Word showed how it sets lines, and Vex Writer now does the same:
A tab after the last word of a full line does not hang in the margin the way a space does: it begins the next line and takes the word after it to its stop there. In 96 test lines, with a space before the tab and without, every line now ends where Word ends it.
Spaces, tabs and line breaks set in a larger font of their own do not make a line taller. A line that holds nothing else is as tall as its line break, or else as the paragraph mark. A no-break space does count. Documents converted from PDF often keep their tabs and breaks in another font than their text, and their lines came out taller than in Word.
Bold that a font does not have is made up by Word, and the made-up bold is a little wider: 0.0175 of the font size per character. Word also makes it up for a face that calls itself bold but is light by its weight.
In a Word 97-2003 document the formatting of every paragraph mark is now read, not only that of empty paragraphs: Word gives it to the number of a list item and to a line of nothing but spaces.
Later the same day three more: Word on the Mac gives Helvetica a line a fifth taller than the font itself says, as macOS does, and sets editions of Helvetica in Arial; a character scale written with a fraction, as converters from PDF write it, is cut to a whole percent rather than ignored; and in Word 2013 documents a table is placed by its edge, not by the text of its first cell, so tables no longer stand 5.4 pt to the left of where Word puts them. A report converted from PDF that Word lays out on 32 pages came out on 26, and now on 32.
In the 368 existing documents 367 now come out on as many pages as in Word (361 before), 31,850 of 31,951 pages begin with the same line, and 98.7% of lines end where Word ends them.
When Word opens a document, it fits the columns of a table that fits its text: a column narrower than its longest word gets just that width, and the other columns give it what it lacks, each in proportion to how much wider it is than its own longest word. Word's own copies of tables made in Vex Writer showed the rule to a tenth of a point, and Vex Writer now lays such tables out the same way. In the long table of the set every page now begins where it does in Word; before, none did.
In the 34 documents made in Vex Writer, 91 of 98 page breaks now fall where Word puts them, and 97.7% of lines end where Word ends them. In the 368 existing documents Word had already fitted the tables, and only one of them changed.
A PDF made in Vex Writer now also reads in order: a footnote number or a page number stands in its place in the line, and pictures and fields leave no stray characters in it (except in underlined lines), so text copied from it comes out in the same order as from Word's PDF.
Later the same day test documents laid out in Word showed four more rules, and Vex Writer now follows them: a line break right after a full line ends that line instead of leaving an empty one below it; an empty paragraph that only holds a section break takes no room; the height a table row asks for leaves out the margins of its cells, which Word adds; rotated text in a cell never makes its row higher, it wraps at the height the row has. Justified lines now widen only their spaces, as Word does, and no longer space out the letters. In the 368 existing documents 361 now come out on as many pages as in Word (330 before), 99.6% of pages begin with the same line, and 98.6% of lines end in the same place.
Not done yet: tables with no widths set, which Word lays out by their longest text; tracked changes and comments in balloons in the margin, the way Word prints them.
Thirty-four documents were made in Vex Writer and seventeen workbooks in Vex Sheets, typed and formatted with the apps' own commands, from a few plain paragraphs to a report of twenty pages and a model of three sheets. Then Microsoft Word and Excel for Mac opened them.
Word opened all 34 documents, each on as many pages as in Vex Writer. Excel opened all 17 workbooks and recalculated them itself: all 7,629 formulas gave the same results as in Vex Sheets, and the charts, tables, rules, notes, links and sparklines were all still there in the copy Excel saved.
On the way the check found faults, all fixed:
Excel did not open a workbook with a table that had a calculated column and a totals row. References to tables are now written in the form Excel reads from a file.
Excel dropped the notes of a new workbook.
A copy of a sheet lost its tables.
A figure number made in Vex Writer turned into plain text when Word updated it.
A watermark made in Vex Writer did not show in Word, and Word numbered its lines from 2.
Space between paragraphs, lines with superscripts, borders of paragraphs, footnotes and formulas among text now follow the rules Word lays them out by, found with test documents laid out in Word. So do a page colour, a column break and the first paragraph of a new section.
Microsoft Word for Mac and Vex Writer each made a PDF of the same 368 existing documents, and the two were compared line by line. Of 1,337,425 lines, 1,317,584 now end where Word ends them, 98.5%, and 31,322 of 31,951 pages begin with the same line. At the start of the day it was 97.3% of lines.
The comparison showed where Vex Writer still differed from Word. Fixed since:
Columns of tables of fixed width take the widths of their cells, as in Word.
A font that is neither on the Mac nor described in the document is set in the document's own font, as Word does.
Justified lines squeeze their spaces only when that keeps them closer to their natural width than stretching would.
Text in Calibri gets the right width of a space.
A .doc file is laid out in the compatibility mode of Word 2003, as Word lays it out, and keeps that mode when saved as .docx.
A link inside a run of text keeps the formatting of the run.
Text in Cambria breaks by the widths Word gives its characters. They were measured in Word's own PDFs of test documents, not taken from Microsoft's font.
Then every one of these documents was saved by Vex Writer as .docx and opened in Word. All 368 open, each on as many pages as the original. The only binary .doc among them showed two more faults on the way, both fixed: list overrides of .doc files were read from the wrong place, so Word would not open the copy, and tables of .doc files were saved as tables that fit their columns to the text, so Word laid the copy out on 265 pages instead of 257.
Both apps recognise a damaged file, say what is wrong with it and open what it still holds in repair mode, as a new document. The damaged file itself is left as it was.
Pictures in the EMF and WMF formats of Windows, common in older documents and in pasted charts, are drawn by the apps' own player.
Both apps carry Carlito and Caladea, free fonts with the widths of Calibri and Cambria, so that documents and workbooks set in those fonts keep their lines and pages. This is why the two apps now take 23 MB together.
Vex Writer: drawings of .doc files, columns within a page, and fonts missing on the Mac replaced by what the document says of them.
Vex Sheets: shapes and text boxes with their text, charts in the colours of the workbook, pivot tables recalculated, protection, cell styles, sparklines and checkboxes in cells.
Microsoft Word, Excel and PowerPoint were installed on the development Mac, to check VexOffice against them.
The page "How we tested" says what stands behind each figure on this site: which files were used, what was measured and how, what a figure does not mean, and what has not been checked yet.
Tables are made from a block of cells, with styles, a totals row and references by column name. Charts are made from cells, and notes, links and filters by conditions are added and changed. A page setup window sets paper, margins, scaling, headers and footers.
New functions: GROUPBY, PIVOTBY, PERCENTOF and the FORECAST.ETS family.
An edit recalculates only the formulas that depend on the cells it changed. Workbooks with close to a million formulas are calculated in seconds, and an edit takes milliseconds.
Cells are drawn in tiles that are kept while the sheet scrolls, and only the cells in view are drawn.
About 300 functions were added, with their Russian names, and the formula engine passed 500 functions.
