DOC model

readDocModel(bytes, limits?) composes the three rich Word 97-2003 readers into one block-tree model for a consumer that maps it to WordprocessingML. It returns undefined for input that is not a readable unencrypted .doc, throws DocResourceLimitError when a budget in limits is exceeded (and Error for an invalid limit), and lists recoverable damage in warnings. It is read-only.

import { readDocModel } from '@christophervr/ole2';

const model = readDocModel(bytes); // DocModel | undefined
for (const block of model!.body.blocks) {
	if (block.type === 'paragraph') for (const item of block.inlines) { /* text, picture, fieldBegin, ... */ }
	else for (const row of block.rows) for (const cell of row.cells) { /* cell.blocks is a DocBlock[] */ }
}

Shape

Field	Content
body	the main story as blocks
sections	each DocSection with headers and footers by variant (default, even, first): { index, inherited, applies }, where index points into headersFooters at the entry that supplies the content
headersFooters	one entry per section slot, with its story blocks (empty when the section inherits)
noteSeparators	the footnote and endnote separator, continuation separator and continuation notice stories
footnotes, endnotes, comments, textBoxes	the stories, indexed by the ids used in the inline items
fields, bookmarks, blips	the typed field list, bookmark list and picture store the inline items point into
styles, lists, fonts, properties, facingPages	from readDocFormatting and readDocStructure
warnings	deduplicated warnings of all three readers

A story is { story, cpStart, cpEnd, blocks }. A block is a paragraph or a table.

• DocParagraphBlock: resolved props (style cascade plus direct formatting), markProps, style (istd), list (ilfo, level, the override, the definition and the effective level, when the paragraph is numbered), inlines, and endsSection (the index of the section whose break this paragraph's mark carries).
• DocTableBlock: gridTwips, rows of cells with gridColumn, columnSpan, rowSpan, vMerge, cell and row props (borders, shading, widths, height) and blocks. Nested tables are blocks inside a cell. A continue cell is covered by the cell above that restarts the merge.

Inline items

Control characters never appear in text. Tabs stay as \t; 0x1E and 0x1F become U+2011 and U+00AD.

Item	Source
text	formatted characters, with the run's resolved props
break	0x0B line, 0x0C page (not at a section end), 0x0E column
picture	0x01 with a picture location; picture.blip has the image bytes
ole	an embedded object, placed after its field separator; object.preview is the preview picture
fieldBegin, fieldInstruction, fieldSeparate, fieldEnd	complex-field marks sharing an id into model.fields; fieldBegin.field is the typed field (HYPERLINK url, REF bookmark, ...). Fields may span paragraphs, so the marks are flat; the instruction text is fieldInstruction, the result is ordinary items
bookmarkStart, bookmarkEnd	positioned at the bookmark's CPs; id indexes model.bookmarks
noteReference, commentReference	0x02 and 0x05 in the main story, with the id of the note or comment
noteMark, separator	the 0x02/0x05 mark at the start of a note or comment story; 0x03/0x04 in separator stories
shapeAnchor, textBoxAnchor	0x08 in the main or header story, with the floating shape; textBox indexes model.textBoxes

Header, separator and text box stories end with one extra bare paragraph mark that is not content; the model drops it. A section break (0x0C at a section end) ends its paragraph and is reported as endsSection, not as a break item.

Positions

Every cp* field is an absolute document CP. A DocStoryRange also carries the story id and story-relative start/end. The table, section, note and text box readers keep their story-relative fields (DocTable.start, ...) and document it.

What is shared

The three readers run over one StoryContext (openStoryContext): one text decode, one piece table, one CHPX/PAPX run index, one sprm parser (parseGrpprl), one sprm budget and one warning list. DocModelLimits is the union of the budgets (maxCharacters, maxPieces, maxParagraphs, maxRuns, maxSprms, maxCells, maxEntries, maxRecords, maxBlipBytes, maxNesting, ...); DocStructureLimits, DocObjectsLimits and DocFormattingLimits are subsets. readDocFormatting now returns the paragraphs of every story in stories (paragraphs is still the main story).

Not covered

• Symbol characters (sprmCSymbol), form-field data, hyperlink binary data and comment ranges (annotation bookmarks) are not decoded; revision marks, field-result caching flags and document protection are not mapped.
• Table row and cell properties beyond width, merge, vertical alignment, borders and shading stay in props.sprms.
• Floating shape content other than text boxes (groups, effects, gradients) is available only through the shape's raw properties.
• Page-level settings beyond the section properties (line numbering, footnote numbering formats, document grid) are not read.