I've really got three paths, as I see it now... the first is
to make mswordview much like a static library (just lots of
functions), and do a high level dump into some intermediate
format (either something like markup through strings, or maybe a
contextual state machine interface), but I'd really like to
avoid this layer... but if it works, it works.
The second approach is to use much (but not all) of the
mswordview code as a static library, and basically model
the top-level parsing (and HTML generation) that goes on
from mswordview, but in our own code which would call the
proper piece table populate functions. We'd basically just
be merging in all the support code for mswordview (and there's
quite a bit of it), and recreating the code in the if()
statements to do our thing. This would be efficient, but
would require more understanding of how mswordview does its thing.
The third is to use Roberto Arturo Tena Sánchez's COLE
(the successor to oledecod, used in mswordview) as the stream
parser, and model Caolan's parsing logic to grab the juicy bits
from the stream. This last one seems like the cleanest and most
maintainable method, especially since COLE can evolve outside our
work, but it's also the most effort to get going. I've
already started hacking on a class that does this, but there's
a lot of redundant "support" code between AbiWord and mswordview,
so it'll take me a little while to get the top-level parser working.
What would you suggest, assuming you've been digging around
in the mswordview code a bit? Is it reasonable as a library?
-- Shaw Terwilliger