[Kst] : Huge performance problem under Windows

Barth Netterfield netterfield at astro.utoronto.ca
Thu Nov 19 18:26:18 CET 2009

On November 19, 2009 12:10:57 pm Peter Kümmel wrote:
> As I understand it the object store is not global, but
> many classes have an instance of it (search for "ObjectStore *_").

Not formally, but it is effectively, since it can be gotten from anywhere via 
the app/doc/etc.  As it is used as a global, It would be just as well, from my 
point of view, to be done with all the _store pointers and make it a global.  
There is only one of them after all.

> Anyway, as you said, using reference counted pointers should make
> ObjectStore superfluous. A container of refcounted pointers should be enough.

That is effectively what it is, with extra code globbed on to make adding and 
removing things... worse?

> And I also don't like all the casting when getting a object from
> the store, we are not using Java, with a master class 'Object', or
> Objective-C with 'NSObject'.  One advantage of using C++ is to check at
> compile-time for type-correctness not at runtime, with the help of
> templates.

Most of the casting is related to finding all of the objects of a certain type.  
Sometime you want
	All Objects
	All Primitives 
	All Scalars
	All DataScalars

One list + dynamic casts makes this possible (if annoying).  Some of the 
helper template functions in ObjectStore are designed to take care of this.

> A first fix would be to replace all ObjectStore member variables with
> a container holding QSharedPtrs. This also makes the code more
> readable.

ie, take stuff out of ObjectStore that doesn't belong there, and use QT's lists 
as is.  I think we do want the template functions which return typed sub-

I think we would prefer QExlicitlySharedPtrs though... what do you think?


More information about the Kst mailing list