How does XSL FO work?

Core principles of XSL FO

Conceptual compatibility with DSSSL Compatibility with CSS properties Screen properties as well as print No compromises on internationalization Closely linked to XSLT transformation language Provides an abstract formatting language.

The FO output tree

The input is transformed into an output tree consisting of: page masters, which define named styles of page layout; and page sequences, which reference a named page layout and contain a flow of text. Within that flow, text is assigned to one of five (rectangular) regions (the page body, areas at the top, bottom, left and right) We also have allowance for floating objects (at the top of the page), and footnotes (at the bottom), and the model covers writing in left/right and/or top/bottom modes.

Inside regions and blocks

Within a region of text, we find one or more blocks tables lists and floats and within a block, we find inline sequences characters links footnotes and graphics

Object properties

aural properties borders, spacing and padding breaking colors font properties (family, size, shape, weight etc) hyphenation positioning special table properties special list properties BUT supporting all of them is not mandatory.

Simple example

Take this bit of TEI input The corr element marks correction

]]> and transform it, as an interpretation of what the corr element is supposed to do, with the following XSLT template.

Sample XSLT template [] ]]>

Sample FO output The corr element marks [correction] a a a mistake ]]>

Sample rendition

FO implementations

We know of six XSLFO implementations: RenderX's XEP, probably the most complete and reliable Apache's FOP open source Java, creating PDF or AWT, with a large amount of the XSLFO specification implemented. Sebastian Rahtz' PassiveTeX, a library of TeX macros which interpret XSLFO; Unicorn's XSLFO engine also uses TeX, but works by generating a normal TeX file from an internal FO tree. Arbortext's Epic is not yet in widespread use and its coverage of the specification is unknown Antenna House's XSL Formatter is different in that it formats for the screen, using Windows GDI calls. It has good coverage of the specification (albeit not complete), and is fairly robust. From which we can see that there is no 100% reliable implementation of the Candidate specification.

Sample 1: PassiveTeX

Sample 1: FOP

Sample 1: XEP

Sample 1: Antenna House

Discussion 1

The results are within fairly acceptable limits, though XEP's inability to hyphenate shows clearly how vital a good H&J engine is - the word and letter spacing on the 3rd line from the end is clearly unacceptable.

Let us now decrease the page width, and look at a harder paragraph.

Sample 2: PassiveTeX

Sample 2: FOP

Sample 2: XEP

Sample 2: Antenna House

Discussion 2

The results show two rather different formatting decisions taking place in FOP and XEP. The former opts to abandon sensible line breaking, and let the URL stick out into the margin, but XEP forces itself to extreme lengths to justify with letter spacing, while Antenna House makes an arbitrary break.