Created attachment 143213 [details]
Small sample source file

I just started to use docbook processing instruction (PI) in our data
sets. Running current xml2po on our data set shows that it picks up PI
along with translatable string.

On gnome-doc-devel-list, a few other people agreed on that "PIs should
absolutely not be marked for translation."

http://thread.gmane.org/gmane.comp.gnome.documentation.devel/459

to find out where, I dug a bit. I run xml2po on attached sample code
with enabling a debug code at the top of XMLDocument::normalizeNode()
in xml2po/__init__.py, I got:

> > <norm> (element) [<norm>PI Test</norm>]
> > <text> (text) [PI Test]
> > <norm> (element) [<norm>Section Title</norm>]
> > <text> (text) [Section Title]
> > <norm> (element) [<norm>sample table</norm>]
> > <text> (text) [sample table]
> > <norm> (element) [<norm><?dbfo bgcolor="#000000" ?>Translatable String</norm>]
> > <dbfo> (pi) [<?dbfo bgcolor="#000000" ?>]
> > <text> (text) [Translatable String]

it seems to me that normalized node with PI is judged
"worthOutputting".  Is it right to assume that
XMLDocument::worthOutputting() is the place to modify and make it to
return false if element has PI?