The Next Stage in Composition -- McIlroy
By all accounts XSL-FO can be considered a robust system, at least for technical documents. There's very little information out there yet on what works best, and what doesn't really work.
Probably the most detailed paper around is Eliot Kimber of ISOGEN International's presentation, "Using XSL Formatting Objects for Production-Quality Document Printing," offered at XML 2002 in Baltimore. As Kimber points out, "XSL Formatting Objects have unavoidable limitations from two principal causes: missing layout features and the limitations inherent in the two-step XMLxt-pages processing model." He says also that FO is "not a full solution for index generation." (Although this is addressed in the 1.1 version recommendations.)
Kimber, while generally very much on XSL's side, points also to a range of specific limitations, including an inability to deal with:
* Text that flows around arbitrary curved areas (but text flowing around rectangular areas is possible using side floats). There are no extensions that satisfy this requirement.
* Page-location-sensitive inclusion or exclusion of content. For example, there is no direct way to condition the text of a cross reference based on whether or not the target of the reference occurs on the same page as the reference itself. There are no extensions that satisfy this requirement.
* Any other presentation tuning semantics that require feedback.
Ken Holman echoes Kimber's theme when he writes, "Unfortunately there are many 'common' requirements that just couldn't be met with XSL-FO 1.0 that will be addressed in future versions. I understand that had the committee tried to add everything in the first version, it would never have been released due to feature creep. The first version was necessary to understand how it was going to be used." (The recommendations for version 1.1 were published in mid-December, but appear to be more of a "bug fix" for 1.0, than a new version.)