text:id is deprecated

[Boudewijn Rempt]

On Thursday 15 December 2011 Dec, Jos van den Oever wrote:
> On Thursday, December 15, 2011 11:21:19 AM Thorsten Zachmann wrote:
> > > Well, no... Not that I could find. In fact, until yesterday encountering
> > > an xml:id when loading text or bookmarks always means "aha! rdf!", and
> > > the loading code doesn't have access to the rdf document, which it could
> > > use to figure out whether the id occurs in the rdf store as well. (I
> > > notice that KoTextMeta can save itself, but not load...)
> > 
> > Then this needs fixing. I think we should not keep the xml:ids around in
> > the loaded document.
> > 
> > Can you explain a bit more what the problem is here?
> 
> The xml:id is there because some part of the document references the element. 
> This reference should be present in some form. One form would be to have a 
> common way to reference that can be serialize as unique id's. Since different 
> parts of Calligra could reference a particular element and this reference is 
> written as xml:id, a common registry for these references is needed.

It turns out that currently, only the rdf system keeps the actual id strings after loading is finished, the other parts of calligra only use the id strings to build their mappings during loading, and recreate them during saving.

> Each element could suggest a prefix to make it easier to read, but the prefix 
> should be chosen by element and not by the code that references it, since 
> multiple parts could reference it.
> 
> A tricky part will turn out to be to make sure that keeping a reference does 
> not prevent delation. QWeakPointer could fit this role.

Well, for that I proposed yesterday to use QSharedDataPointer -- i.e., make the whole system work just like QString or QImage, but without the ability to get/set the actual identifying part out of the reference.

I've got sort of an action plan now:

1. Saving

* check all code that saves draw:id and text:id and make it also save xml:id
* keep track of the list of generated xml:id's and make double sure they are always unique. (I still would prefer uuids, like used for lists and rdf, but failed to convince zagge :-)
* only save xml:id (formerly text:id) for KoTextBlockData if the kotextblockdata is actually used for animation.
* make sure that even if two cross-references want an xml-id to be written to an element, that the links in other parts of the document are correct.

2. Loading

* fix the places where we use getAttribute("id") to be getAttributeNS(KoXmlNS:xml, "id");
* check all that loads draw:id or text:id and make it load xml:id first, if present, if not, fall back on draw:id and text:id.
* first load the objects that the xml:id's in the body may refer to, so can build the cross-references on loading

2. RDF

* make the KoDocumentRdf instance available to the KoDocumentResourceManager
* make KoDocumentRdf create a list of all xml:id's that are used in manifest.rdf
* on loading, when encountering an xml:id, check whether it's present in that list
*   if so, create the KoTextInlineRdf object, if not, don't (replaces the check for rdfid- in the xml:id string)
* replace keeping the xml-id string with an ElementReference class that can be used
during runtime, so we don't have to do the complex xmlid rebasing after saving we do now.
* make KoTextMeta actually load and save, since that was never implemented...

In passing: it would be good if we could directly attach rdf metadata to an element (like text:table) instead of always having to work with bookmarks.

3. KoTextBlockData

This class causes a text:id to be saved in all cases, but it's only relevant when animations have been added. So, only then should we save the xml:id.

-- 
Boudewijn Rempt
http://www.valdyas.org, http://www.krita.org, http://www.boudewijnrempt.nl