Converting AbiWord (.abw, .zabw) to Markdown: Open-Source Word Processing Recovery

In the late 1990s and early 2000s, the AbiSource project created AbiWord, a fast, cross-platform, open-source word processor. Designed to run on resource-constrained hardware where bloated office suites choked, AbiWord became a staple on Linux distributions, GNOME Office desktops, and the One Laptop Per Child (OLPC) XO initiative.

To store documents, AbiWord engineered an open, human-readable XML-based specification using the .abw extension (and .zabw for GZIP-compressed files). Long before Microsoft Word adopted Office Open XML (.docx), AbiWord was already proving that rich word processing documents could be stored cleanly in structured markup.

However, on modern macOS, Windows, and mobile devices without AbiWord installed, opening a .abw or .zabw file often results in raw XML code or "unsupported file format" errors.

This engineering guide examines the internal architecture of the AbiWord document tree and explains how convrtr's ABW to Markdown extractor transforms vintage AbiWord documents into portable GitHub Flavored Markdown with preserved YAML frontmatter, tables, styles, and embedded illustrations—100% locally in your browser.

The AbiWord Document Hierarchy

An AbiWord document is a single XML file conforming to the AbiWord Markup Language (AWML) Document Type Definition (DTD). Its root element is <abiword>:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE abiword PUBLIC "-//ABISOURCE//DTD AWML 1.0 Strict//EN" "http://www.abisource.com/2001/xhtml/awml">
<abiword xmlns:awml="http://www.abisource.com/2001/xhtml/awml" version="3.0.5">
  <metadata>...</metadata>
  <history>...</history>
  <styles>...</styles>
  <section>...</section>
</abiword>

1. Dublin Core Metadata Mapping

AbiWord was an early adopter of the Dublin Core metadata standard. Inside the <metadata> block, document properties are serialized as key-value pairs:

<metadata>
  <m key="dc.title">Operating System Architecture</m>
  <m key="dc.creator">Linus Torvalds</m>
  <m key="dc.date">2001-05-15</m>
  <m key="dc.description">Technical notes on microkernel vs monolithic design.</m>
  <m key="abiword.keywords">kernel, memory, virtual filesystem</m>
</metadata>

When convrtr's engine processes this block, it serializes the Dublin Core tags into standard YAML frontmatter for Obsidian, Notion, and static site generators:

---
title: "Operating System Architecture"
author: "Linus Torvalds"
date: "2001-05-15"
description: "Technical notes on microkernel vs monolithic design."
tags:
  - kernel
  - memory
  - virtual filesystem
---

Paragraph Styles and Heading Levels

Within the document's <section>, textual content is stored in <p> elements. Paragraph formatting is declared through style and props attributes:

<p props="heading:1">Microkernel Architecture</p>
<p props="heading:2">Inter-Process Communication</p>
<p style="Block Text">"A distributed system is one in which the failure of a computer you didn't even know existed can render your own computer unusable."</p>
<p style="Bullet List">Message queues</p>
<p style="Numbered List">Initialize message descriptor</p>

The converter maps these properties directly into GitHub Flavored Markdown syntax:

• heading:1 or style="Heading 1" $\rightarrow$ # Heading 1
• heading:2 or style="Heading 2" $\rightarrow$ ## Heading 2
• style="Block Text" $\rightarrow$ > Blockquote
• style="Bullet List" $\rightarrow$ - Bullet list item
• style="Numbered List" $\rightarrow$ 1. Numbered list item

Character Spans and Inline Typography

Within paragraphs, styled runs of text are enclosed in <c> (character) elements using CSS-like property strings in the props attribute:

<p>
  This text includes <c props="font-weight:bold">bold words</c>, 
  <c props="font-style:italic">italic phrasing</c>, 
  <c props="text-decoration:line-through">strikethrough text</c>, and 
  <c props="font-family:Courier">monospace code</c>.
</p>

The parser inspects the props string and wraps inner text with corresponding markdown delimiters:

• font-weight:bold $\rightarrow$ **bold words**
• font-style:italic $\rightarrow$ *italic phrasing*
• text-decoration:line-through $\rightarrow$ ~~strikethrough text~~
• font-family:Courier $\rightarrow$ `monospace code`
• text-position:superscript $\rightarrow$ <sup>text</sup>
• text-position:subscript $\rightarrow$ <sub>text</sub>

Hyperlinks defined as <a href="...">...</a> are translated into standard markdown link syntax: [Label](URL).

2D Grid Table Extraction

AbiWord handles tabular structures with <table> and <cell> elements. Rather than relying on HTML-like rows (<tr>), AbiWord indexes cells using 2D grid coordinates:

<table>
  <cell x="0" y="0"><p props="font-weight:bold">Subsystem</p></cell>
  <cell x="1" y="0"><p props="font-weight:bold">Latency</p></cell>
  <cell x="0" y="1"><p>L1 Cache</p></cell>
  <cell x="1" y="1"><p>1.0 ns</p></cell>
</table>

The parser groups cells by their y row coordinate, sorts columns by x, and formats a clean Markdown table with headers and alignment separators:

| Subsystem | Latency |
| --- | --- |
| L1 Cache | 1.0 ns |

Handling Embedded Images and GZIP Compression

1. Inlined Base64 Images

Unlike DOCX or ODT files that store images as loose binary files in a ZIP container, AbiWord inlines raster illustrations directly into XML <image> tags:

<image mime-type="image/png" data="iVBORw0KGgoAAAANSUhEUgAA..."/>

The converter extracts the base64 string, strips whitespace, and outputs a self-contained Markdown image:

![Embedded Image](data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAA...)

2. GZIP Archive Decompression (.zabw)

To save disk space, AbiWord allows users to save documents with the .zabw extension. A .zabw file is a standard GZIP-compressed .abw stream. The converter detects GZIP magic bytes (0x1F 0x8B 0x08) and inflates the document in memory before XML parsing.

100% Client-Side Privacy

Confidential documents, manuscripts, and legal agreements should never be transmitted to third-party conversion servers. With convrtr's ABW to Markdown converter, all XML decoding, style translation, and image processing execute entirely within your browser using client-side TypeScript. Your documents never touch an external server.