Re: [xslt] No user context available when apply stylesheet loads an external entity

[Rush Manbert <rush manbert com>]

Rush Manbert wrote:
> Hi,
>
> I have an application that uses libxml2 to parse an XML file, then uses 
> libxslt to apply one or more XSL stylesheet transformations to the 
> document. The result of this is then extracted as a HTML string and 
> given to a web browser control to display.
>
> I have registered an external entity loader with libxml by calling 
> xmlSetExternalEntityLoader(). I have a set of search paths that are 
> registered, and the loader function handles searching them in the proper 
> order.
>
> The problem that I have is that I need to have multiple windows open, 
> each of which displays a separate XHTML document. Each window can have 
> its own set of search paths for file loading. This means that when my 
> external entity loader is called, I need to have enough context so that 
> I can find the proper set of search paths.
>
> For parsing the XML file, I discovered and started using 
> xmlCtxtReadFile(). I allocate a parser context, use the _private data 
> member to store my context, then retrieve it in the entity loader.
>
> However, when I apply a XSL stylesheet transformation to my document, 
> using xsltApplyStylesheetUser(), it often loads other XHTML source 
> files. This calls the loader, but the parser context that is passed in 
> has no context information that I can use to find the correct path set. 
> The myDoc member has a value of 0, and there is no pointer to the 
> xsltTransformContext, so I can't get at its _private data member. I also 
> tried setting the _private data member in the transform context, hoping 
> that it might be propagated to the parser context, but that doesn't happen.
>
> Have I missed something here? How am I supposed to know my application 
> context when I am in the entity loader? I suppose I could guard the XSL 
> transformation with a semaphore, and set the context info into some 
> global location that the loader can retrieve from, but that seems like a 
> nasty hack.
>
> I am using libxml2-2.6.19 and libxslt-1.1.14.
More info:
I found the xsltSetLoaderFunc() function, so I copied the function 
xsltDocDefaultLoaderFunc() and made my own version that adds this line 
right before calling xmlLoadExternalEntity():
	pctxt->_private = ((xsltTransformContextPtr)ctxt)->_private;

(The cast is required because the transform context ptr was cast to 
void* in the call.) I verified that the transform context pointer was 
the same as that passed to xsltApplyStylesheetUser(), so I thought I was 
home free. However, it turns out that xsltLoadDocument() is hard coded 
to call xsltDocDefaultLoaderFunc(), and the function to set a new loader 
has no effect!

I then searched on the libxslt site and found this:
http://mail.gnome.org/archives/xslt/2004-August/msg00034.html

so this seems to be intentional.

I would like to suggest that the call to xsltDocDefaultLoader() should 
pass the xsltTransformContextPtr value, not cast as a void*, and that 
xsltDocDefaultLoader() should copy the _private member from the 
transform context to the parser context. The parser context is allocated 
inside the function, so there can't be anything there to step on. This 
way we can set the private data member when we apply a stylesheet and 
use it in the entity loader. Otherwise there's just no context 
information available that relates back to the program that is using the 
library. Does this make sense?

Thanks,
Rush