> > Gives this result:
> >
> > <foo>&#150;&#8211;</foo>
> >
> > I've checked the input file with a hex editor to make sure the
> > un-escaped dash really is 0x96.  Somehow the two characters are
> > treated differently, which is something I didn't expect.
> >
> > I think that 0x96 in the input XML read using Windows-1252 should
> > become #8211 when output using any encoding other than Windows-1252,
> > which is what is happening for the actual character 0x96, but the
> > character reference #150 gets serialised back as #150...
>
> Isn't this beause &#150; is a unicode entity? It's not a windows-1252
> entity. In other words a character entity never changes according to
> the input encoding.

Ahh of course, that makes sense.  The character for #150 is worked out
after the bytes in the document have be parsed using the encoding
specified in the prolog....

So 0x96 becomes #8211 though the mapping defined in Windows-1252, and
#150 remains as #150 because its a character reference and character
references are always unicode.

Thanks Nic!