[PATCH v2 2/2] msxml3: Don't unlink the XML declaration when saving document.

[Dmitry Timoshkov]

Dmitry Timoshkov <dmitry at baikal.ru> wrote:

> Nikolay Sivov <nsivov at codeweavers.com> wrote:
> 
> > On 4/5/21 2:40 PM, Dmitry Timoshkov wrote:
> > > Nikolay Sivov <nsivov at codeweavers.com> wrote:
> > >
> > >>>> Signed-off-by: Nikolay Sivov <nsivov at codeweavers.com>
> > >>> Please don't commit this, while the patch is correct, it causes a regression:
> > >>> ::load()/::save() leads to a bug: the saved file has double <?xml> header,
> > >>> and needs another fix: doparse() should never add an xml declaration to a just
> > >>> loaded XML.
> > >> Thanks for catching that.
> > >>> Nikolay, do you recall why that code was added in the first place?
> > >> I was a long time ago. I think the reason is that libxml2 does not use a
> > >> separate node for prolog PI, and msxml does.
> > >> After the loading we have to expose prolog node as a normal child.
> > >>
> > >> I don't think it always adds it, the meaning of 'doc->standalone != -1'
> > >> check is to add this node only if it was present in the input.
> > > Do you have an alternative solution to removing that code from doparse()?
> > > Because doing so breaks the tests. I'd appreciate a bit of help on this.
> > There is no good way that I'm aware of. Node has to be there,
> > applications expect that.
> > If the issue is save/get_xml() output, we could reimplement node
> > serialization similar to what libxml2, but in a way we want it.
> > 
> > To get rid of linking/unlinking we'll have to get rid of libxml2 as a
> > dependency, which probably could still use libxslt. But that's really a
> > lot of work and planning (sax, schema validation, xpath, parsing itself).
> 
> I'm not proposing to blindly remove the linking/unlinking the xml declaration
> code. What I would look into is revisit the doparse() code that adds the PI
> node with an XML declaration in case of doc->standalone = -2. According to
> libxml2 source doc->standalone = -2 indicates that the XML declaration *is*
> present however the standalone attribute is not, and in that case another
> (the 2nd one) XML declaration gets inserted as an PI node, however libxml2
> couldn't figure that out which leads to creating double <?xml> header.

I think that I've found a solution that doesn't break the tests. I'll send
a new version of the patch.

-- 
Dmitry.