Hi,
I've found the solution myself. Adapted RDF-import. Will be back for other questions...

*returns to hacking*

Matthias

Although I seem to have a conversation with myself I vastly improved the code from above (which was one big mess).

The code now really parses the pdf-file. It recreates the pdf-objects and the pagetree and looks for annotations (instead of just crudely throwing everyting inbetween "Contents()" in a note). It does not create a new Note for every annotation, but instead creates a new document with notes. Every note is accompanied by its pagenumber (btw. there really should be a way of storing those with a note) and a tag, telling whether it is a highlight or an annotation.

I would appreciate if somebody actually tried this - I can make a screencast if necessary.
Also, I would appreciate it if somebody could take a look at the "Store()"-function. While the parsing of the file is as efficient as I could make it, the actual storing of the annotations seems to create a huge SQL-overload - some bigger files (with several hundreds of annotations) take up to 10 min. to store. (compared to just 20 sec. of parsing the file and extracting the annotations).

Anyway, here's the code, make sure you get everything, I had to distribute it over two posts:

var objlist=new Object(); // stores the PDF-objects, stripped of all streams
var pagelist=new Array(); // stores the pageobject of each page

var pagenumberdifference=0; // indicates that the real pagenumber is different, has to be globel because function is recursive
var pagenumbertype=0; // 1=(normal) 2=[roman] 0=consistent with pdf

// returns failsafe either "" or the content of an object in the objectlist;
function getObj(objcode) {
var objcontent=objlist[objcode.replace(/^\s|\s$/g,"")];
if (objcontent!=undefined) {
return objcontent;
} else {
return "";
}
}

// creates a new Annotation Object
function Annotation(objcode) {
this.objreference=objcode;
this.contents="";
this.type="";

var objcontent=getObj(objcode);
if (objcontent=="") return; // Object does not exist

// extract contents
var pos=objcontent.search(/\/Type\s*\/Annot\s*/i);
if (pos!=-1) { // Annotation found
// extract type
pos=objcontent.search(/\/Subtype\s*\/Text\s*/i);
if (pos!=-1) this.type="Note";
pos=objcontent.search(/\/Subtype\s*\/Highlight\s*/i);
if (pos!=-1) this.type="Highlight";
pos=objcontent.search(/\/Subtype\s*\/Freetext\s*/i);
if (pos!=-1) this.type="Text";

pos=objcontent.search(/\/Contents\s*/i);
if (pos!=-1) { // simply hidden in Contents()-Tag
objcontent=objcontent.slice(objcontent.indexOf("(",pos)+1);
//objcontent=objcontent.slice(0,objcontent.search(/\)\s*\//i)); // weaker version, might be incompatible
objcontent=objcontent.slice(0,objcontent.search(/[^\\]\)/i)+1);

objcontent=objcontent.replace(/\\r/ig," "); // cleaning up linebreaks and other escaped sequences
objcontent=objcontent.replace(/\\\)/ig,")");
objcontent=objcontent.replace(/\\\(/ig,"(");
objcontent=objcontent.replace(/\\\s/ig,"");

// change type to pagerenumber and trim to number if Pagerenumber-Popup
pos=objcontent.search(/PageRenumber/i);
if (pos!=-1) {
objcontent=objcontent.replace(/[^\d\-\[\]\(\)]/g,"");
this.type="PageRenumber";
}

this.contents=objcontent;
} else {
// complexer version would require parsing of streams - not implemented
}

}
}

// creates a new Page Object
function Page(objcode) {
this.objreference=objcode;
this.realpagename=""; // caller usually handles the pagenumber, however, the constructor may modify it if he stumbles upon a PageRenumber-note
this.annots=new Array();

// now check whether page has annotations
var objcontent=getObj(objcode);
var pos=objcontent.search(/\/Annots/i);
if (pos!=-1) { // Annotations found
if ((objcontent.indexOf("[",pos)==-1) || (objcontent.indexOf("R",pos)<objcontent.indexOf("[",pos))) { // probably child object used, but this is just guessing
pos=objcontent.indexOf("R",pos)-4;
pos=objcontent.lastIndexOf(" ",pos);
objcontent=objcontent.slice(pos,objcontent.indexOf("R",pos)-1); // extract obj where childobject can be found
objcontent=getObj(objcontent);
pos=0;
}
objcontent=objcontent.slice(objcontent.indexOf("[",pos)+1,objcontent.indexOf("]",pos)-2); // discard anything except childtree
objcontent=objcontent.replace(/\s*$/,"").replace(/R*$/,""); // discard final R
var pagetree=objcontent.split(" R ");
for (var dummy=0; dummy<pagetree.length; dummy++) {
this.annots.push(new Annotation(pagetree[dummy]));
if (this.annots[this.annots.length-1].contents=="") {
this.annots.pop();
} else {
if (this.annots[this.annots.length-1].type=="PageRenumber") { // read real pagenumber from a popup note
this.realpagename=this.annots[this.annots.length-1].contents;
this.annots.pop();
} //else /*DEBUG*/ Zotero.debug(this.annots[this.annots.length-1].contents);
}
}
}
}

I just came across this and it sounds very interesting. I am, however, not sure how to get this code running? Can you help me?

Thanks!

See http://www.zotero.org/support/dev/creating_translators_for_sites , but I'm not sure that the code works with the current release of Zotero.

I haven't tried this, but Joscha has been doing some work with highlighting/annotating in Zotfile:
http://www.columbia.edu/~jpl2136/zotfile.html

mheim - while not Zotfile's main function, Joscha has indeed implemented exactly this functionality in the most recent version (I believe it's in beta). As I understand it's only working for Mac so far, though. See Joscha's post from August 18 here:
http://forums.zotero.org/discussion/18737/