htmlpurifier/docs/ref-strictness.txt

Is HTML Purifier Strict or Transitional?
    A little bit of helpful guidance

Despite the fact that HTML Purifier professes to support both transitional and
strict HTML, it rejects a lot of attributes and elements that are actually, indeed,
valid. You can investigate progress.html to find out precisely what we
are doing to these *deprecated* attributes.

However, users have found that Strict HTML imposes some quite unreasonable
restrictions on certain things. The start and value attributes in ol and
li (respectively) perhaps are the most contested. There's is currently no
widely supported browser method short of JavaScript that can replace these
two deprecated elements. It behooves us to allow these deprecated
attributes when the output is transitional.

Fortunantely, that's the only real bugger case. The others have near-perfect
CSS equivalents, and were presentational anyway. However, the other question
pops up: should we always convert these to the CSS forms when 1. the spec
allows them anyway and 2. older browsers support them better? After all, the
whole point about CSS is to seperate styling from content, so inline styling
doesn't solve that problem.

It's an icky question, and we'll have to deal with it as more and more 
transforms get implemented.  As of right now, however, we currently support
these loose-only constructs in loose mode:

- <ul start="1">, <li value="1"> attributes
- <u>, <strike>, <s> tags
- flow children in <blockquote>
- mixed children in <address>

The changed child definitions as well as the ul.start li.value are the most
compelling reasons why loose should be used.  We may want offer disabling <u>,
<strike> and <s> by themselves. We may also want to offer no pre-emptive
deprecated conversions. This all must be unified.