At present, we seem to look for each inflected form in each work separately, which means that the order appears random: one form late in the work is output before another form which comes early in the work.
Possibly faster and better supported, by same author as XML::DOM::Lite.
I think this just requires adding strawberry\c\bin to the PATH, so that it can find libxml2-2__.dll
We should avoid re-parsing prefs file at each query.
Provide a way to search them by headword
Write dedicated xml to html-for-epub converter.
Make sure we export to XML hidden sources for fragments, as in Accius, Carmina
Especially for Latin, the PHI texts need to be supplemented with additional works from Perseus and DigiLibLT. Supporting this would require rewriting Diogenes so that it operates on the XML versions of the PHI and TLG databases. But much of the code could be taken from DiogenesWeb, after search has been implemented there.
Interface would be rewritten from Perl/cgi to html/js. No need for a server, except for morphological Ajax requests. Keep Perl infrastructure for converting XML and Perseus/Logeion server, at least for now. Eventually rewrite the morph server in Node.js.