On 8/3/2010 3:44 PM, Lyublena Antova wrote:
I tried to use Xerces with the pluggable MemoryManager and I discovered
that on several occasions objects are instantiated with the global new
operator that does not use the memory manager. Here are some of those cases:
* initializing the EncodingValidator in EncodingValidator.cpp
* creating a DOMImplementationListImpl in DOMImplementationImpl.cpp
and DOMImplementationRegistry.cpp
* creating a DOMNodeListImpl in DOMNodeImpl.cpp
* creating a DOMDocumentTypeImpl in DOMImplementationImpl.cpp
* ...
In our code we essentially forbid the use of plain global “new” so the
above cases blow up when Xerces is linked against our codebase.
To my understanding the pluggable memory manager is used either:
* by making classes derive from the XMemory class which overloads
new and delete, or
* by using the global overloaded placement new operators that take a
DomDocument(Impl) object
The problem classes mentioned above are not derived from XMemory but
occasionally get instantiated with a plain “new” operator instead of the
placement “new”-s.
I have a fix that makes those classes inherit the XMemory class, and
thus get instantiated with the global memory manager. That caused some
problems because on some occasions the global placement “new”-s were
shadowed by the Xmemory member “new”-s which produced unexpected
results. The solution was to force the use of the global new (::new) to
avoid wrong resolving of operator calls.
Just out of curiousity, can you provide an example of where this
occurred in the Xerces-C code?
Was there any reason why the classes above do not inherit from XMemory
in the first place?
Just inheriting from XMemory isn't always the right fix. In some cases,
there may be an available MemoryManager instance, either as a function
parameter or as a class member. It may also have been done that way to
avoid multiple inheritance, particularly in the DOM implementation classes.
On a broader note, is there a particular reason why not have a placement
new operator that takes a MemoryManager instance? Perhaps deallocation
issues?
Do you mean a global placement new operator? If so, I suspect the
deallocation issue is why it doesn't exist.
Dave
---------------------------------------------------------------------
To unsubscribe, e-mail: c-dev-unsubscr...@xerces.apache.org
For additional commands, e-mail: c-dev-h...@xerces.apache.org
