The problem is that on return from the externalEntityRefHandler() call-back
the parser currently assumes that parsing of the parameter entity is finished,
and updates its internal state accordingly. I do not have the time at this
point to implement and test a solution. Being able to suspend while parsing
the DTD is also less desirable than doing the same while parsing content.
an XML_ERROR_NO_ELEMENTS condition could happen.
The reason was that the internal entity's start tag level was not preserved
when parsing was suspended. This has been corrected.
they don't produce an access violation when the value of eventPtr is
less than the value of positionPtr. This can happen when a handler
raises an exception or potentially in some border cases.
Also added some more updates of eventPtr.
to a parameter passed by reference (pointer). It seems that the processing
of return values is not standardized, even when calling convention and
platform are specified. This should make Expat more usable as a shared library.
Also removed some code that was never executed, because the condition
(prologState.documentEntity && role == XML_ROLE_INNER_PARAM_ENTITY_REF)
can never be true. Improved some comments as well.
- not all cases handled for switch based an enumeration:
added default: ; where appropriate
- char* passed where const char* was expected:
changed variable declarations to const char*
declarations; these are needed only on an obscure platform (Windows NT
on PowerPC using GCC), and were never in previous releases of Expat.
They caused way too many spurious warnings on several platforms where
they aren't actually needed but should be ignored silently.
Removing these cannot break working code.
the use of modified default calling conventions in client code.
To deal with this issue and generally clean up the mass of macros
being used to support bits of the machinery, two new macros are being
added:
- XMLCALL, which expands to whatever is needed to nail down the
calling convention for all calls across the library boundary. This
must match the convention used for the system's malloc()
implementation.
- XMLIMPORT, defined to be whatever magic is needed to mark an entry
point as imported from a dynamically loaded module (.dll, .so, .sl,
whatever).
These macros are used to define the XMLPARSEAPI macro already being
used to define the API entry points. In addition, XMLCALL is used to
define the types of callback functions, and all example code uses this
explicitly in both the distributed applications and the documentation.
See bug #765227.
* lib/internal.h:
(FASTCALL, PTRFASTCALL): only define these macros for the GNU C compiler
on i386 platforms. apprently, they do not work well on PPC ports.
expanding %percent; in the second of these entity declarations:
<!ENTITY % percent "%">
<!ENTITY %percent; y "value">
This patch was submitted by James Clark on the xml-dev mailing list.
Call storeAtts() for all element start tags; this is necessary to
ensure attribute defaults are properly processed for all elements
(needed to do proper namespace checking, at the very least), and that
tag names are properly cooked when there's an end-element-handler but
no start-element-handler.
This causes the new tests to pass, and closes the SF tracker issue.
Error with xmlns:prefix= with namespace processing enabled.
It seems that storeAtts() has the following properties:
- when called with tagNamePtr = NULL and bindingsPtr = NULL,
it does not process attributes incl. namespace declarations
- when called with all arguments non-NULL it processes
attributes (incl. default attributes) and namespace declarations
We also have these requirements:
A) for start of element event:
1) if we have a startElementHandler:
we need to process at least default attributes,
so we must call storeAtts with all arguments non-NULL
2) if we have no startElementHandler, but we have attributes:
we need to process namespace declarations,
so we must call storeAtts with all arguments non-NULL
3) if we have no startElementHandler and no attributes:
we need to store the name (for the end element event, where
the names are compared) and call only the default handler
Note: Storing the name is a pre-requisiste for calling
storeAtts with all arguments non-NULL.
So there really is no place for calling storeAtts with
tagNamePtr = NULL and bindingsPtr = NULL.
B) for empty element event:
1) if we have a startElementHandler:
we need to process at least default attributes,
so we must call storeAtts with all arguments non-NULL
2) if we have no startElementHandler, but we have attributes:
we need to process namespace declarations,
so we must call storeAtts with all arguments non-NULL
3) if we have no startElementHandler and no attributes,
but we have an endElementHandler:
we need to store the name for calling the handler,
but we need not process any attributes (default or not)
4) if we have no start- or endElementHandler, and no attributes:
we need to call only the default handler
Given that storeAtts will now always be called with all arguments
non-NULL we could remove a few internal checks in storeAtts,
if that improves efficiency. Not sure if that is worth it.
This patch should therefore fix the problem of namespace declarations
not being processed when no startElementHandler is set, that is,
it will fix Jeremy's NS processing patch.
the default handler does not get called even when no element handlers
are called, because storeAtts() may trigger the startNamespaceDeclHandler
where the application has a chance to clear the element handlers.
This patch fixes this with one issue still open, which applies to empty elements only:
When the endElement handler is called, but not the startElementHandler,
then the default handler is not called, which causes attributes to not
be reported at all. The other alternative would be to call the default handler,
but then the element would be reported twice.
