it is not valid to use (type & ND_TEXTNODE) before casting to SwTxtNode,
type of SwCntntNode has the ND_TEXTNODE bit as well, but SwCntntNode is not
derived from SwTxtNode.

>>= operator overload turn the right-value of the operator into a
left value... which confuse the heck out of the compiler's detection
of use of unitialized variables (and also C-reader like me for
whom >>= means something else altogether...
(why on earth are we right-bitshifting... with an unitialized variable...
oh! fracking operator overload!!!)

http://lists.freedesktop.org/archives/libreoffice/2011-December/021638.html

It's simply a bunch of documents with various math features in them.
It'd be nice to have an automated test built on top of them that
checks the formula is the same (roughly, in the meaning, it can't be
precise because import adds e.g. extra {}'s) after an export+import
roundtrip.

correct minor css rendering too

When writer would choose the same bias automatically that word
implies or this doc hard-codes, then elide the font-forcing.