Office documents — a ZIP of namespaced parts¶

The .docx on your disk and the .odt next to it are not single XML files. Each is a ZIP archive of many XML parts, wired together by relationship files. This page looks at both big families — Microsoft's OOXML (.docx, .xlsx, .pptx) and OASIS's OpenDocument / ODF (.odt, .ods, .odp) — because together they show what happens when a vocabulary gets large: dozens of namespaces, split across files, referencing each other.

OOXML: the main document part¶

word/document.xml holds the text. WordprocessingML's vocabulary lives almost entirely under one prefix, w:, with r: (relationships) reaching out of the part to images and hyperlinks.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

<w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"
            xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships">
  <w:body>
    <w:p>                                          <!-- (1)! -->
      <w:pPr><w:pStyle w:val="Heading1"/></w:pPr>
      <w:r><w:t>Quarterly report</w:t></w:r>
    </w:p>
    <w:p>
      <w:r>                                        <!-- (2)! -->
        <w:rPr><w:b/></w:rPr>
        <w:t xml:space="preserve">Bold </w:t>
      </w:r>
      <w:hyperlink r:id="rId4"><w:r><w:t>link</w:t></w:r></w:hyperlink>  <!-- (3)! -->
    </w:p>
  </w:body>
</w:document>

1. w:p is a paragraph; w:pPr is its properties (paragraph style, spacing). OOXML's pattern is rigid: a content element, optionally preceded by a …Pr properties sibling. Once you see it, the whole format reads.
2. w:r is a run — a span of text with uniform formatting — and w:rPr is its run properties (w:b = bold). Text always lives inside a w:t.
3. r:id="rId4" does not contain the URL. It is a relationship id that points into a separate part, word/_rels/document.xml.rels, where rId4 maps to an actual target. The r: namespace exists purely to express these cross-part links.

Note xml:space="preserve" on the bold run: the xml: prefix is the one namespace every XML document gets for free (bound to http://www.w3.org/XML/1998/namespace), used here so the trailing space in "Bold " survives.

The relationships part¶

The id from r:id="rId4" is resolved here. This is OOXML's answer to "how does one XML part point at another file in the package":

<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships">
  <Relationship Id="rId4" Type="http://…/hyperlink"
                Target="https://example.org" TargetMode="External"/>
  <Relationship Id="rId5" Type="http://…/image"
                Target="media/logo.png"/>
</Relationships>

The indirection (rId4 → real target) keeps document.xml stable when targets move, and lets binary assets (images) live as separate package members rather than being base64-stuffed into the markup.

ODF: the same problem, OASIS's answer¶

OpenDocument splits content the same way but makes a different namespace choice: instead of one big w:, it uses many small, semantic namespaces — office:, text:, table:, style:, draw:, fo: — so each prefix names a domain.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14

<office:document-content
    xmlns:office="urn:oasis:names:tc:opendocument:xmlns:office:1.0"
    xmlns:text="urn:oasis:names:tc:opendocument:xmlns:text:1.0"
    xmlns:table="urn:oasis:names:tc:opendocument:xmlns:table:1.0"
    xmlns:style="urn:oasis:names:tc:opendocument:xmlns:style:1.0"
    office:version="1.3">
  <office:body>
    <office:text>
      <text:h text:style-name="Heading_20_1" text:outline-level="1">Report</text:h>
      <text:p text:style-name="Standard">Hello
        <text:span text:style-name="Bold">world</text:span>.</text:p>
    </office:text>
  </office:body>
</office:document-content>

Editing a document by hand¶

Because the package is just a ZIP of text, you can change a document with nothing but unzip, an editor, and zip — no Office libraries at all. This is the most direct way to feel what the format really is:

unzip report.docx -d report/         # explode the package
unxml report/word/document.xml       # read the part compactly (or edit it)
# … change "Quarterly report" to "Annual report" in word/document.xml …
cd report && zip -r ../edited.docx . && cd ..   # repackage
# edited.docx opens in Word with the edited heading

Three things make this work — and one that quietly breaks it:

• The edit is text in, text out. As long as you keep document.xml well-formed and do not touch the [Content_Types].xml that names each part, Word reads your hand-edited package without complaint.
• Binary assets (the media/logo.png from the .rels) are copied through the zip untouched; you are only ever editing the XML parts.
• Mail-merge and report generation in the wild are often exactly this: unzip a template, string-replace tokens in document.xml, rezip. No COM automation, no headless Word.

Querying Office XML¶

Both formats are heavily namespaced, so — exactly as with SVG — //p finds no paragraphs. You must bind the prefixes in your XPath host and query //w:p or //text:p. The wrinkle that surprises people: the prefix you use in your query need not match the document's; only the namespace URI matters. A document could declare xmlns:foo="…wordprocessingml…" and //foo:p would still be wrong unless you bound foo to that URI too.

Things to note¶

• A "file" can be a package of XML parts; the interesting structure is across parts, not within one.
• Relationship indirection (r:id → .rels) keeps markup stable and keeps binary assets out of the XML.
• A large vocabulary can be organized one-prefix-per-app (OOXML) or one-prefix-per-concept (ODF) — two namespace strategies, with different trade-offs.
• xml:space and xml:lang ride the always-available xml: namespace.

Next: Atom and feed extensions, where the lesson flips — instead of one huge vocabulary, a tiny core that everyone extends.