Both tables in $subject (in datatype.sgml and xfunc.sgml, respectively)
contain similar information (though the xfunc one mentions C structs and
header files, and the datatype one does not, but has a description column)
and seem similarly out-of-date with respect to the currently supported
types ... though not identically out-of-date; they have different numbers
of rows, and different types that are missing.
How crazy an idea would it be to have include/catalog/pg_type.dat
augmented with description, ctypename, and cheader fields, and let
both tables be generated, with their respective columns?